If you enjoyed the post, consider sharing it with a friend or colleague; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

Over the years, I’ve realized that there’s one single trait that has outsized influence on someone’s level of impact at work. It’s not raw intelligence, and neither is it “people skills”.

It’s probably best illustrated with an example. Let’s say a manager has a small team with two people on it, Hannah and Lance. A new project comes up, and the manager asks Lance to own it. He agrees. A few days later, the manager checks in and asks Lance how he’s progressing. “I was waiting to get more clarity”, he says. “I didn’t really know where to start. What should I do? Who should I loop in? Can we scope this out together?”

The manager agrees, and they spend an hour putting a high-level project plan together.

When Lance comes back a week later with a “project update”, the manager hopes that he has some progress to share. Instead, he brings a list of questions. He is unfamiliar with a part of the project and asks if someone else can handle it instead. He describes three different approaches and asks which one the manager prefers. He’s wondering if the project timeline can be pushed out because he’s waiting to hear back from a key stakeholder.

The project continues like this. The only progress, it seems, is made during the live working sessions the manager and Lance have together.

Frustrated, the manager assigns the next project to Hannah. He mentally prepares himself for a similarly painful process; after all, Hannah has less experience than Lance. But he’s shocked to find that the opposite is true.

• Day 3: Hannah sends a message that she put together a working group, identified a similar project that’s running in parallel, and is working with the respective team to deduplicate efforts

• Day 7: She gives a brief summary of how they decided to proceed, and says they’ll continue unless there are any concerns

• The week after: The manager gets an excited message from another team lead. Instead of writing a lengthy doc and holding endless meetings to debate what to build, Hannah whipped up a prototype so people could react to something tangible. They’re asking if this can be the default approach from now on.

After the project is done, other teams start approaching the manager to ask if he can staff Hannah on their work streams as well. The manager wishes he had more than one Hannah on the team; but since he can’t clone her, he at least puts her up for promotion so she doesn’t leave.

What happened? Is Hannah smarter than Lance? Did she work longer hours? Not necessarily. The main difference is that Hannah has agency.

A brief primer on agency

If you have access to the internet, you’ve probably heard of the concept of agency by now. But in case you haven’t, let’s briefly discuss what it is. There are many definitions out there, but I think about it like this:

(High) Agency is the belief that you can shape your circumstances and achieve your goals through your own actions, rather than waiting for ideal conditions — and the ability to follow through on this belief.

In simpler terms: You think you can make things happen, and then you find ways to make them happen.

It’s kind of the Silicon Valley version of manifestation; but instead of waiting for the universe to make things go your way, you just go and do it yourself.

So what does this look like at work?

In the example above, low-agency Lance isn’t really doing anything. He outsources key decisions, is uncomfortable tackling new or challenging topics, and sees himself as a victim of his circumstances.

High-agency Hannah, on the other hand, just gets things done. She doesn’t get discouraged easily, she’s unafraid to make decisions, and she rolls up her sleeves and does what she needs to do — whether or not that’s “her job”.

How to develop agency at work

Every manager wants more high-agency people on their team, and everyone that wants to make an impact and get ahead in their career should strive to increase their agency.

But when I first learned about this concept, I thought: “That’s neat; but how do you actually develop this trait? You can’ just wake up and decide to have more agency, can you?”.

I still think that’s true; simply deciding to change the entire way you approach problems doesn’t do much. But over time, I’ve realized that agency is like a muscle: The more you practice this mindset, the easier it will become.

Once you have the first few “wins” under your belt, you realize that you can, in fact, change things that you thought were out of your control, and solve problems without waiting for someone to assign them to you or give you permission.

At a certain point, it just becomes part of how you operate, and you’ll be able to move an order of magnitude faster than others.

But as is often the case, getting started is the hard part:

• How do you spot opportunities to show agency? Sometimes, we’re not even aware we’re in a low-agency mindset, so it’s hard to break out of it

• What are concrete examples of high agency in action? Without knowing what it looks like, it’s hard to emulate it

This post aims to help with that. First, we’ll briefly cover three mindset shifts that will help you increase your agency. Then, for the remainder of the post, we’ll go through typical situations at work where people default to a “low-agency” mindset and discuss what you can do instead.

Three core principles for increasing your agency

Before we get into concrete examples in the next section, here are some general mindset shifts I found helpful in developing more agency:

1. Focus on what you can control instead of making excuses

When we feel stuck, there are often legitimate blockers in our way, or adverse circumstances that put us in a difficult position.

It’s understandable to want to give up or complain that things aren’t fair. Maybe somebody told you things were going to be easier than they turned out to be; or, there are other people doing similar things who don’t have to deal with the same challenges.

But we always have two choices: Use the challenging circumstances as an excuse to give up and not do what we planned to do, or focus on the things we can do to make it work.

You’ll be surprised how many ways to move forward you’ll find once you stop thinking about whether the situation is fair, or whose fault it is that you’re in it.

Fred Kofman, author of the book Conscious Business, articulates this well (and I highly recommend reading the full book):

“You are not responsible for your circumstances; you are response-able in the face of your circumstances.” - Fred Kofman, Conscious Business (emphasis mine)

2. Assume you can do anything until proven otherwise

Most people’s default is to assume they can’t do something until they figure out how to do it. The logical consequence is that they don’t even attempt a lot of things because — at first glance — they seem intimidating and beyond their (current) capabilities. And they never take a second, closer look to figure out if that’s even true.

This is one of the most common, and most severe, forms of self-sabotage.

Simply flipping this assumption on its head can massively improve your life, and your career. What helped me with this was realizing that most successful people didn’t get to where they are based on raw talent or intelligence; they got there because they assumed they could change every single thing around them and remove every obstacle in their way — and then they went and did it, even if they initially had no idea how to do it.

“A big secret is that you can bend the world to your will a surprising percentage of the time—most people don’t even try, and just accept that things are the way that they are. […] Almost always, the people who say ‘I am going to keep going until this works, and no matter what the challenges are I’m going to figure them out’, and mean it, go on to succeed.” — Sam Altman, How To Be Successful

In this context, it can be helpful to try not to overthink things. You won’t be able to figure out everything in advance anyways, and with every potential problem you identify your conviction will decrease, making it less likely you’ll see things through.1

Related to this: Be careful not to ask too many people for their opinion too early, or at least don’t take their opinions as fact. People tend to assume anything that hasn’t been done yet cannot be done, and they’ll come up with a million reasons why you’ll fail.

3. Don’t wait for someone to tell you what to do or give you permission

Another incredibly common way to avoid action is to outsource the responsibility to someone else. For example:

• You see a problem that you think should be fixed, but doing so would take a lot of effort. So you think “that’s not my problem” because nobody told you to fix it

• Or, you put a critical project on hold because one of 12 stakeholders hasn’t explicitly signed off yet

If you wait for others to decide on a path forward, or give you explicit permission, you’re often adding weeks to a project’s timeline; sometimes, you’re even killing it completely.

Not because the idea was bad, but because most organizations are incredibly inefficient at making decisions. If you want to get things done, it’s often best to keep the number of people involved to an absolute minimum, do the thing, and then get everyone else comfortable with the result.

Of course, whether you can get away with this heavily depends on the culture of your company. In some organizations, “just doing things” can get you in trouble, and you definitely shouldn’t go against direct guidance from your manager.

But even in more “sign-off-heavy” cultures there are ways you can unblock yourself; similar to the mindset shift discussed above, you can flip the default here as well.

Instead of waiting until everyone has explicitly signed off, you set a deadline for people to give feedback; after that deadline, everyone is assumed to be in agreement and you move forward. As long as you explicitly communicate this in advance and give people enough time to voice concerns, this often strikes a good balance between making stakeholders feel heard and moving reasonably quickly.

Concrete ways to be “high agency” at work

Alright, let’s get specific. In this section, we’ll cover concrete examples of high-agency behavior across multiple areas:2

1. Career development

2. Choosing and structuring work

3. Handling blockers

4. Managing projects & stakeholders

In addition to things you can do “the old fashioned way”, I’ll also discuss how you can leverage AI wherever it makes sense. Let’s dive in!

1. Showing agency in your career development

Common complaints from people who feel stuck in their careers are that their manager is not promoting them, or they don’t get any feedback, or they aren’t getting opportunities to take on more responsibility.

In almost all cases, what this means is that nobody is proactively offering these things, not that they asked for them and got rejected.

While it’s true that managers should be doing some of that on their own, simply waiting for this to happen shows a very low-agency mindset.

Nobody cares about your career as much as you do; and if you don’t get what you need from your manager, it’s up to you to take initiative and change that.

That means:

Your manager isn’t giving you feedback?

Schedule a 30-minute feedback session a week in advance, and send your manager a heads-up with a few bullets on what you’d like to discuss. Tons of managers forget to provide feedback regularly, but almost none of them would refuse to give it when put on the spot.

You’re not getting growth opportunities?

Figure out what you want, and share clear goals with your manager. Then, work with them to find ways to move in that direction (e.g. specific projects to be staffed on).

Worst case, they’ll tell you you’re not ready for the kind of opportunity you have in mind; if that happens, push them for concrete feedback and develop a plan to address it. The goal is to either get access to the opportunities you want, or get clarity on where you need to improve to be considered.

Your manager isn’t putting you up for promotion?

Take ownership of the process yourself. Figure out what it takes to get to the next level and work with your manager to identify any gaps, learn about the process and deadlines, and put together a plan. Collect evidence of your readiness for promotion along the way and don’t rely on your manager to compile your promo packet by themselves.

Simply doing solid work and hoping that it gets noticed is not enough.

🤖 How you can leverage AI:

Kicking off the conversation with your manager is an important first step. But if you leave it to them to develop a structured plan, you’ll likely be disappointed.

Instead, give the career ladder or rubric that your company uses for your role to gauge performance by level to AI, together with the feedback you received in the last few review cycles. Then, ask the AI to 1) highlight the key differences between the expectations at your current level and the next, and 2) summarize your main areas for improvement.

Prompt example:

Please review the attached career ladder for a [your role, e.g. Product Manager] as well as the feedback I received during the last performance review cycles.

I'm currently level [X] and want to get promoted to level [Y].

Output the following:

- An executive summary of what it takes to get from my level to the next
- A detailed comparison of what is expected at the next level vs. my current level across each dimension of the career rubric
- A summary of the feedback I have received and how it maps to the role expectations at my current level and the next one
- Suggestions for where I should focus to become "promotion-ready", incl. concrete practical ideas for how I can demonstrate these capabilities. Focus on one or two key themes rather than a tactical laundry list.

You want a role that doesn’t exist?

Invent it and pitch it.

It might not always work out, but if it does, you have the opportunity to write your own job description, and you’ll be the first (and potentially only) applicant.

2. Showing agency in how you approach your work

Over the years, I’ve managed a number of different teams; some of these I “inherited”, others I built from scratch.

And while every company, team, and person I worked with were unique, there has been a consistent goal across all of these: Increasing agency.3

From a manager’s perspective, there are essentially three “levels” of ICs:

Most people early in their careers start at Level 1. That’s natural; without a lot of experience, you typically lack the confidence to make judgment calls and figure things out for yourself.

But moving to Level 3 should be the explicit joint goal of any (ambitious) IC and their manager. Getting there is a win-win:4

• 💼 For the manager, it frees up bandwidth so they can spend more time making sure the team is working on the right things, has enough resources, works well with other teams etc.

• 🛠️ For the IC, it means having the freedom to solve problems creatively instead of feeling micromanaged and just executing orders. And you’ll be much less likely to get replaced by AI: AI can already execute straightforward tasks with detailed instructions well, but is (so far) unable to take a high-level goal and run with it

Unfortunately, this transition is hard. The more hands-off the manager is, the more ambiguity there is that the IC needs to push through. Being a high-agency IC means 1) trusting you’ll figure it out and then 2) pushing relentlessly to do just that.

Here are some behavioral shifts that can help you move in the right direction:

Get sign-off on your approach instead of asking what to do

If you don’t do any work unless your manager tells you to do something, you’re essentially asking to be micromanaged.

Asking your manager what to do is only slightly better; you’re proactively signaling that you’re idle and want to get started, but you’re not taking the initiative to figure out what needs to get done. You’re still asking your manager to do all of the heavy lifting.

Instead, tell your manager what you intend to do and give them an opportunity to provide input. Ideally, you already start executing while you’re waiting for a response (that, depending on how busy your manager is, might never come).

That way, you’re removing your manager as the bottleneck while still giving them an opportunity to weigh in.

🤖 How you can leverage AI:

In order for this approach to actually remove friction, you need to make sure that you give your manager sufficient information so that in most cases they will simply respond along the lines of “sounds good to me” instead of hitting you with a barrage of questions.

The easiest way to achieve this is to ask AI to review your approach and anticipate likely questions or concerns your manager will have. Then, you can proactively address them in your message.

Prompt example:

<goal> I'm about to send my manager an update re: a project I'm working on. Please review 1) the project context below as well as 2) my draft message and highlight any questions or concerns they might have (so I can proactively address them). </goal>

<context>

- [Either add context on your project here, or attach relevant documents (e.g. project plans, meeting notes etc.)]

- [Paste your draft message to your manager here or attach it as a document]

</context>

<output> In addition to a list of questions or concerns my manager will likely have in response to my update, please also provide a proposed edit of my message (add placeholders in the message wherever I need to insert additional context or information you don't have before sending it out). </output>

Note: This works especially well if the AI has context on who your manager is and how they think. You can achieve that by creating a Project (e.g. in ChatGPT or Claude) and uploading past questions and feedback you’ve gotten from your manager and then using this Project to run the prompt above.

Figure out how to make something work instead of finding reasons why it won’t

A lot of people, especially in larger organizations, love to point out reasons why something won’t work instead of trying to figure out how to make it work.

I think people do this because:

1. They think it makes them look smart to point out every risk, shortcoming, and challenge, and

2. Because it helps them cover their ass.

Getting something done means you have to take a stance and push through obstacles. If things don’t work out as planned, most people prefer to be the one that says “I told you so” instead of the one that put their personal reputation behind the failed effort.

However, this risk-minimizing approach will only get you so far. In high-performance organizations, you are ultimately not judged by how few mistakes you made, but rather by the level of impact you realized. So while constantly playing devil’s advocate might give you job security in the short term, it will severely limit your career trajectory in the longer term.

When faced with a new task, assume by default that it can be done, and pretend you have to do it; how could you make it work? Consistently practicing this mindset is one of the easiest ways to stand out.

Bonus points: Proactively identify problems and opportunities

Doing the things discussed above is already a solid start. But you can take it one step further: Instead of waiting for someone to hand a project or problem to you, pay attention and proactively spot opportunities to either unlock additional growth or improve efficiency.

The more senior you get, the more the expectation will be that you “own” an area of the business and help decide what work is worth doing. Practicing this even when it’s not officially part of your job description yet is a great way to show you’re ready for more responsibility.

🤖 How you can leverage AI:

I don’t recommend outsourcing this process of identifying opportunities to AI, as it doesn’t have enough business context to do a good job here.

However, once you find something that seems promising, you can use AI to size the opportunity and pitch it to your manager.

Just attach a document with notes on the opportunities you identified and add some context on your company so the AI can provide a customized sizing framework.

<goal> I have a list of opportunities that I identified across the company. I want to size their potential impact to figure out which ones should be prioritized. Your job is to help me put together a structured sizing framework for each one. </goal>

<context> [Add context on your company, team etc. here] </context>

<output> The framework you output should specify what metric can be directly impacted by the work stream, and establish a connection to a topline metric (e.g. Daily Active Users, Revenue etc.; this will be the ultimate business impact we can use to compare initiatives). Provide the calculation and list out the data that needs to be plugged in; in addition, provide benchmarks from comparable companies in case I cannot pull some of the data from our internal systems </output>

3. Showing agency in how you handle blockers

A classic symptom of someone with a low-agency mindset is that the moment something blocks their progress, they jump on the “excuse” to stop working. Progress is out of their hands; before they can move forward, someone else needs to unblock them.

In most cases, they either look to their manager or other teams for that.

To be clear, this is a very natural reaction to running into friction, and I’ve been there plenty of times myself. But these moments are the ideal opportunities to practice agency, and learning to unblock yourself will pay massive dividends. You’ll be able to move much faster, and your manager will love you for not trying to pass the monkey to them every single time something comes up.

Raising your bar for what constitutes a blocker

Many people consider themselves blocked when they can’t proceed with a project as planned. But that’s just a complication, not a blocker. There are usually multiple ways to get to a desired outcome, and just because your first attempt didn’t work out doesn’t mean you’re out of options.

Only once you’ve exhausted all of the options available to you, you are actually blocked and need somebody else’s help. Which brings us to the next point:

Reduce your reliance on other teams

A lot of the time when people on my team have told me they’re blocked, it’s been some version of “We’re waiting on this other team to do something and can’t move forward until they do”.

This seems like a neat excuse: You can’t move forward, but it’s not really your fault. However, not all dependencies on other teams are immutable things you just have to accept.

For example, when I was at Uber, the Operations teams managing the reliability of the marketplace on a day-to-day basis often felt blocked by the Product team.

• How are you supposed to avoid reliability dips during bad weather if the real-time pricing algorithms don’t take this signal into account? Product should fix this ASAP!

• Our tooling doesn’t allow us to target incentives at specific restaurants? Nothing we can do for McDonald’s reliability until Product builds this capability, I guess!

The best Ops managers, however, didn’t just throw up their hands in defeat and blamed Product. Instead, they implemented hacky band-aid solutions; for example, to address the pain points above, people built:

• A semi-automated process to pull in near-term weather forecast data, generate incentive recommendations, and send comms to couriers

• A Python script that created restaurant-level geofences so Uber’s geofence-based incentive system could be used creatively to boost specific restaurants

Note that this wasn’t done by engineers; these were people from non-technical backgrounds who learned basic SQL and Python to unblock themselves.

This high-agency approach had two benefits: Besides unblocking the business, it also created an incentive for the Product org to act faster. The hacky proof of concept showed that there was a real business case while the brittle nature of these band-aid solutions made Product highly motivated to replace them with something more robust.

🤖 How you can leverage AI:

Back when these challenges came up at Uber, AI wasn’t nearly where it is today. I.e. we had to build these things by hand.

Now, with AI, this has become an order of magnitude easier. Instead of learning Python, you can just ask an LLM to code up a simple tool. As long as you’re just using it for internal processes, it matters less whether it’s production-ready code.

Anticipate blockers and bottlenecks

The best way not to get blocked is to anticipate bottlenecks in advance.

Most people focus only on their part of a project and see everything else as “not their problem”. Unfortunately, parts owned by other people often become your problem at some point if you don’t manage dependencies proactively.

• You might be reliant on inputs from other teams? Then you need to scope out what those will be, and agree on the exact things that the other team will deliver (and deadlines for doing so)

• Your work needs to be reviewed, processed, or ingested by another team? Connect with them and align on the specifics so you can bake this into your plan

Don’t just trust that somebody else will be on top of these things; more often than not, you’ll find that nobody took the initiative.

4. Showing agency by just doing things

Agency influencers (yes, that’s a thing) love to repeat “You can just do things!”.

They’re right; you can (I checked!). And yet, many people seem to think their job is to just call meetings, send emails, and talk about things without ever actually doing anything.

Here’s how to stand out by going against this trend:

Put pen to paper

Sometimes, work can feel like you’re playing a game called “How many meetings can we have about a topic without reaching a conclusion?”.

In my view, one of the key reasons this tends to happen is because complex topics are hard to solve in your head. Somebody brings up a new consideration, another person revisits a point that you thought was already decided, and the discussion just keeps going in circles.

The easiest way to prevent this (but a heavily underutilized one) is to just put pen to paper. Here are a few concrete things you can do:

• If it’s a complex decision and many teams are involved, consider putting down clear roles and responsibilities

• Put forward an issue tree or decision framework that structures and simplifies the problem. Or, make a concrete recommendation that people can react to. Either way, it will be much easier to iterate on something concrete

• Document every decision that has been made, whether big or small. That way, you don’t keep reopening the same topics

🤖 How you can leverage AI:

With AI, mock-ups often take less time to build than it takes to write up a doc describing what you have in mind.

And since they give people something tangible to react to, they’re much more effective. When five different people read a proposal, they imagine five slightly different things since they’ll fill in the gaps based on their priors and expectations. With a mock-up (or prototype), you remove that ambiguity.

I recommend to get started with static (or slightly interactive) mock-ups. Both ChatGPT and Claude are great choices:

• ChatGPT if you want high-quality results fast

• Claude if you plan to iterate a lot (as the Artifact system makes this very convenient)

Often, this type of mock-up is enough to get stakeholders on the same page. If you want to flesh things out into a proper prototype, however, you can move over to a dedicated vibe coding / prototyping tool. I personally like Vercel V0; you can get something reasonably pretty and robust fairly quickly, and deploy it with one click to share it with others.

If there’s no clear owner (and you need it done), do it yourself

In one of my previous jobs, we’d been using the same take-home exercise for interviews for years. It was outdated, had been leaked online, and everyone agreed that it should be updated.

But for months, nothing happened. Everyone just… waited. For what? Who knows; maybe for someone else to take the lead, or for leadership to give top-down guidance.

Eventually, I just blocked off an afternoon and created a new exercise. It didn’t actually take that long; the difficult part was getting over the inertia and feeling of “That’s not my job”. What’s more, I’m pretty sure it was much more efficient this way compared to an “official” revamp where every team would have wanted to weigh in on what the exercise should look like.

Note: Of course, you’ll want to do this selectively only for things you care about a lot. Otherwise, you’ll quickly become the go-to person for every “unwanted” task in your org.

Just do small tasks instead of delegating them

I keep a mental “hall of shame” of the most ridiculous things I’ve seen people delegate over the years.

They range from “Can someone please rename this Slack channel?” to “Can you add Tom to the meeting invite?” or “Can you change the slide title from [X] to [Y]?”.

What do all of these tasks have in common? Writing out the ask took longer than just doing the thing itself. Regardless of whether you’re a project manager, Director, or VP: The only thing that delegating this kind of task accomplishes is increasing the (negative) corporate vibes of your culture.5

And while we’re at it: If you do delegate, address your request to a specific person. “Can we…” or “Can someone…” is a surefire way to trigger diffusion of responsibility; nobody will feel responsible, and it won’t get done.

Closing thoughts

Agency can be learned.

Some people might be born with more of it than others, but that doesn’t mean you can’t build this muscle over time. The important first step is to recognize when you’re showing a lack of agency; then, you can practice brainstorming (and executing) high-agency ways to deal with the situation.

If you do this diligently for a while, it’ll become second nature; and your career will take a different trajectory.

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

Most people see AI as a tool that can execute: Do research, draft documents, create images etc. And those are great use cases; the efficiency gains are massive, and you can reinvest the time you free up into more strategic work.

But if you stop there, you’re ignoring the full potential of AI. In addition to automating execution, it can also be an incredibly powerful thought partner. A collaborator, advisor, sounding board, and devil’s advocate on demand.

Using AI this way won’t just get you better results at work, it will also sharpen your thinking by constantly challenging you. In most organizations, you’ll get some feedback every few weeks or months — if you’re lucky. And that’s only with regards to the visible outputs you produce; your thought process is invisible to others and never actually gets pressure-tested and refined.

With an AI thought partner, you can get more rigorous every step of the way.

By the way, the emphasis here is on partner; you should never outsource your intellectual work completely to AI, for three main reasons:

1. Limitations: AI tools just aren’t yet able to autonomously handle any complex task that involves a lot of judgment and context1

2. “Generic” results: Even if they could; it’s difficult to differentiate yourself with AI outputs as they’re often solid, but rarely novel or exceptional

3. Skill atrophy: If you stop doing stuff yourself, you’ll lose those skills and you won’t even be able to properly review and feedback AI work anymore

But all three of these problems can be mitigated if you see AI as a partner that helps challenge and sharpen your thinking instead of doing the thinking for you.

In this post, we’re going to go in detail through how to do that. We’ll cover:

• How to set up your AI thought partner

• 4 principles for getting the best results

• Core use cases, including prompts and examples

• A detailed comparison of which AI tool will make the best thought partner for you

Let’s get into it.

As a reminder, this is Part II in my new series on how to do real work with AI.

It was great to see the positive reception to Part I; it seems like there is a real interest in more tactical, actionable AI content. Thank you so much to the 1,500+ of you that shared the last post.

If you haven’t read it yet, check it out here:

And to make sure you don’t miss future posts, subscribe here for free:

Subscribe now

How to set up your AI thought partner

Step 1: Creating a continued partnership instead of one-off interactions

Interactions with AI bots often feel like quick chats in the hallway with a colleague who doesn’t know exactly what you’re working on. In other words, they often lack context and continuity.

Not a great basis for thought partnership.

The simple first step you should take is to make sure that memory is enabled; with ChatGPT, that’s the little bubble icon on the top right. Under “Personalization”, you can then manage the memories:

But even with the memory feature, chats can still feel disjointed, and the AI doesn’t know which context from past conversations is relevant, which can cause confusion if you’re working on multiple things in parallel.

That’s where Projects come in.

They allow you to create a permanent space for working on a topic where you can add custom instructions and context and group multiple chats together, making things feel more like a continued partnership.

Note: Projects are available in ChatGPT, Claude and Grok. With Gemini, you have to create a “Gem” instead, which has somewhat similar functionality as a Project (custom instructions, persistent context etc.).

Step 2: Turning a “yes-man” into a thought partner

The other big issue with AI is that it acts as a cheerleader. Instead of challenging you and helping you sharpen your thinking, the chatbots often just tell you that your idea is great.

If you want them to be a true thought partner and poke holes in your reasoning when needed, you need to give them custom instructions.

You can do this under “Settings” for both ChatGPT and Claude, but I recommend doing it on the Project-level instead. That way, you can keep a more “balanced” AI personality for your day-to-day interactions and have a critical sparring partner when you need it.

The exact instructions should be based on your personal preferences; for example, you might have to tweak things a bit until you reach the desired tone, or balance between encouragement and criticism that works for you.

For your reference, here are some of the instructions I found helpful:

1. Reason from first principles and explain the thought process

Example instruction:

“Reason from first principles and explain your thought process. If you’re making assumptions, state them clearly.”

In my experience, if you ask AI to critically challenge you, you get a lot of pretty strong assertions without clear reasoning to back them up.

However, in order to decide if the critiques (or recommendations) make sense, you need to understand how they were derived. This instruction, or one like it, will prompt a more transparent thought process.

2. Be specific in your critiques

Example instruction:

“Be specific in your critiques. Mention the exact part of my reasoning you disagree with and explain why. Avoid generalized statements such as ‘That doesn’t work’.”

Along the same vein of the prior point, I quickly started to get annoyed by how broad many of the critiques I got were. For example, “Your analysis is superficial” isn’t particularly helpful feedback; to be actionable, I need to know specifically which key considerations I’m missing.

Once I started asking for more targeted, surgical feedback, the collaboration became much more productive.

3. Ask for more information if you need it

Example instruction:

“You don't have to "solve" the request in the first answer; if you need more information, request it instead of making assumptions or giving generic advice.”

By default, most reasoning models will give you highly generic suggestions if they don’t have enough context to tailor something to your situation. To avoid this, you need to prompt them to proactively ask for more information to fill these gaps.

4. Push for clarity instead of going along with every ask

Example instruction:

“When I’m asking you to help with a task, make sure there is clarity around 1) what needs to be done, and 2) why. If it seems like I’m jumping a few steps ahead, get me back on track and raise the fundamental questions that need to be answered first.”

When you come to your manager with a question, the best ones will help you take a step back and figure out if you’re even asking the right thing.

AI can do that, too, but you need to request this. By default, they’re just going to do whatever you ask them without questioning the premise.

Case in point: I asked ChatGPT to help me brainstorm ideas for a take-home exercise for a Data Scientist interview process. At first, it jumped right in; once I added the custom instructions, though, it pressure-tested the premise by exploring whether a take-home exercise is even needed before brainstorming ideas:

For reference, here is the full set of custom instructions I’ve been using recently (copy-paste version in the footnotes)2:

4 principles for getting the best results

With the basic set-up out of the way, let’s briefly talk about how you can get the best results from your AI thought partner. Then, we’ll dive into hands-on use cases with examples.

Principle #1: You go first & you bring it over the finish line

This article is about thought partnership with AI. As mentioned above, if you completely outsource your thinking, your logical reasoning skills will atrophy and you’ll eventually be unable to solve problems independently without AI.

To avoid that, first think through the problem on your own. Then, once you have a rough mental model, share it with the AI and ask for feedback. In addition to flexing your mental muscles, you’ll also get better results since you’ll be able to explore the problem freely instead of being anchored by what the AI proposes.

The second important point is that you shouldn’t just implement every suggestion you get. The idea behind using AI as a thought partner isn’t that AI is smarter than you; it’s that you get a second, independent perspective and uncover weaknesses and blind spots in your logic.

Providing these fresh inputs is AI’s job. Your job is to use your judgment and experience to decide which parts of the feedback to implement and which ones to ignore.

Keep in mind: If you’re instructing the AI to challenge you like we did above, it will do that. Even if your reasoning is solid, it will always try to find something to criticize.

Principle #2: A true thought partner needs context

If you’ve read my last post on how to get the most out of Deep Research, this one will sound familiar.

As with all AI use cases, if you want quality thought partnership, context is king. The good news: If you set up a project, you only need to provide the basic context once in the beginning, and all future conversations will benefit from it.

To start, I recommend sharing the following:

• 🏢 What your company does: Industry, product, business model, geographical footprint etc.

• 💼 What you do and where you fit into the bigger picture: Your responsibilities, who you report to, what teams you work with, what the swim lanes look like between teams etc.

• 🎯 What you’re trying to achieve: What you’re planning to do, what the desired impact is, what the timeline looks like (incl. hard deadlines), how you’ll measure success etc.

Then, every time you start a new chat, you only need to give a brief update on anything that happened since the last interaction (e.g. the outcome of a meeting), and share context on what you’re trying to do next.

Make sure you don’t just share “hard facts” but also nuanced context about processes and people; for example, if you’re preparing for a decision meeting with an executive, share how this person thinks so that the AI can help you prepare accordingly.

If you don’t remember where you left off, just ask; that way, you 1) don’t have to repeat yourself and 2) you’re not leaving out anything important.

The simplest way to make all this context available is to use integrations with your internal tools or to upload relevant documents to your project (e.g. strategy decks, Product Requirements Documents (PRDs), meeting notes etc.).

However, before you do any of this, make sure you understand your company’s policies on data and AI use.

If there are tight restrictions on what you can share, avoid bulk uploads and write a quick summary that leaves out any sensitive information instead (or, better yet, use your favorite speech-to-text tool).

Principle #3: Be “vulnerable”

When we’re getting feedback on our work from another person, we often wait until it’s somewhat robust out of fear of being judged. We want to impress the other person with how thoughtful and polished our work is, and anticipate as many feedback points as possible.

While that’s understandable, this behavior is a massive source of inefficiency because we often get fundamental input way too late and then need to re-work large parts.

With AI, there’s no need for that.

AI doesn’t judge you and won’t tell anyone how bad the first draft of your work was. So, share your unstructured thoughts, get feedback on shaky hypotheses, and ask “stupid” questions — you’ll be able to iterate much faster.

Principle #4: Keep things open-ended

For other AI tasks (e.g. Deep Research), giving hyper-specific prompts can be helpful so you get exactly the deliverable you need.

However, when looking for thought partnership, the opposite is true in my experience. After all, you want another perspective that helps you see things you might have missed, and it’s difficult to know what you don’t know.

As a result, instead of asking leading questions like “should I do X or Y?”, I recommend using open-ended asks such as:

• “What am I missing?”

• “What other ways could I approach this?”

• “Do you have any other ideas?”

Practical use cases and examples for using AI as a thought partner

Everything we do at work is essentially solving some sort of problem.

Doing that usually requires deeply understanding the problem, coming up with possible solutions, deciding on what to do, and then getting others bought into your proposal.

I’ve covered how to do this manually in a series of posts in the past:

It’s a critical skill-set in basically any job, but it’s also hard. But there’s good news: You’re not on your own anymore. AI can help with all of these steps.

Use case #1: Understanding the problem space

🎯 Goal: Deeply understanding the problem in front of you before you start thinking about solutions

Problems can be deceiving; it might seem obvious at first glance what the issue is, but if you look more closely, you realize it’s often just a symptom of an underlying problem a few levels deeper.

For example, let’s say you notice that in the last weeks, the number of new users who started using your app has decreased. You might think “Well, fewer people are signing up, so I need to promote it more”, but if you really think through the issue, you’ll realize that’s not necessarily the case.

AI can be a great partner in identifying potential root causes to give you “leads” you can investigate. For example, it can help you flesh out a hypotheses tree:

Then, once you have a number of hypotheses for what’s going on, you need to figure out which explanation (or combination of them) is correct.

AI can help with this as well; e.g. you can share your ideas for feedback, or (if you’re in a rush) ask it to take the first stab3:

“Now, I want to investigate what's actually going on. Please provide a detailed plan of how you would find the actual root cause, incl. 1) in what sequence you'd tackle the hypotheses and 2) how you'd prove / disprove them.”

Use case #2: Framing problems from different angles

🎯 Goal: Discover new ways to think about the problem in front of you

After you’ve spent a while thinking about a problem, it becomes difficult to take a step back and approach it with fresh eyes. However, there usually isn’t just a single way of looking at something, and framing a problem in a novel way can be a great way to unblock yourself if you feel stuck.

If you’re lucky, you have a manager or skip-level that can help you do that. But if you don’t, AI is a decent substitute.

For example, let’s say you’re a PM at a social media company and have to figure out how to deal with AI content. You’re stuck wondering “Should we label AI content for transparency’s sake even if it hurts engagement?”, so you turn to your AI thought partner.

In less than a minute, you have half a dozen additional framings, and chances are, one of them will help you move forward:

Use case #3: Brainstorming ideas

🎯 Goal: Going broad; AI can help you generate more diverse ideas than you would be able to come up with yourself

Finding the best solution for a problem involves two steps: 1) Identifying all relevant options, and then 2) deciding on the best one. That first part can be extremely tricky to do on your own, but it’s not realistic to hold team-wide brainstorming sessions for every project.

That’s where AI comes in: After you’ve come up with some initial ideas, ask your thought partner to flesh out the list. More often than not, you’ll get a bunch of ideas you wouldn’t have though of yourself, at least not this quickly.

For example, I asked ChatGPT to help me brainstorm ideas to create buzz around a startup launch. Within seconds, I had a list of 10 creative ideas for marketing stunts.

The problem: They were a bit too “out there” and risky for my taste. At this point, I could have gone back-and-forth to iterate my way to a set of ideas that might actually work.

However, I decided to give the list to Claude instead and asked it to refine it:

I use this approach of having one tool improve the output of another for a number of use cases since each one excels at different things (see the deep dive on capabilities at the end of the post for more on this).

In the case of brainstorming, I find that ChatGPT excels at coming up with ideas, while Claude does a good job of being a critical voice that brings you back to reality, so the ChatGPT → Claude flow works well.

Use case #4: Organizing your thoughts

🎯 Goal: Condensing your raw, unstructured thoughts into a coherent narrative or framework

When you’re thinking through a problem, it’s usually much easier to dump your raw thoughts on paper and then clean them up instead of trying to directly come up with something structured and coherent.

Let’s go back to the brainstorming example from above; after coming up with a list of ideas, you’ll need to translate them into a structured framework so you can pick the most promising ones and move forward.

AI can be a good sparring partner in the process, but I highly recommend practicing this skill yourself as it’s extremely valuable. At the very least, you should try to come up with a rough idea for a structure or mental model before you loop in the AI to clean up and execute your vision.

Use case #5: Challenging your reasoning

🎯 Goal: Pressure-test your reasoning before you share it more broadly with stakeholders

When someone challenges your thinking, one of two things can happen:

1. You logically defend your rationale; this will typically increase the amount of respect the other person has for you

2. You’re struggling to explain how you arrived at your conclusions, and people will lose trust in your recommendations

The problem: It’s difficult to poke holes into your own ideas. So to avoid scenario #2, you’ll need someone else to challenge and pressure-test your reasoning ahead of time. Luckily, as long as you give the right instructions (like we covered at the beginning of the post), AI is actually pretty good at that.

To start, explain your thought process and any conclusions you have come to. Then, ask AI to evaluate it. For example:

“Please pinpoint the weakest spots in my argument and highlight any leaps in logic.”

or

“Please highlight any implicit assumptions I’m making and pressure-test them. If you need more context to do that, please request it.”

Use case #6: Anticipating questions and objections

🎯 Goal: Anticipate questions from stakeholders so you can address them in advance

It sucks when you spend days preparing for a presentation, and 5 minutes into it you’re getting stumped by a question from the audience.

To prevent this, describe what you’re preparing for (e.g. a decision meeting for project X) and who the audience will be (e.g. the Head of Sales and CFO) and then ask your AI thought partner to come up with likely questions they’ll have.

The more context you provide (e.g. sharing the meeting deck or financial model you plan to present), the more targeted the question you get will be. For example, this is what I got from Grok as a reaction to a multi-tab Google Sheets model containing a Marketing forecast:

Use case #7: Planning your work

🎯 Goal: Minimizing the chance of dropping the ball or wasting effort by making sure you have considered all relevant operational aspects before you start implementing things

Moving fast is crucial in startups, and you don’t need a perfect plan before you can start executing. However, spending 10 minutes pressure-testing your approach with AI can save you hours of headaches and wasted time down the road.

Some of my favorite things to ask AI include:

🧐 Making sure the plan is comprehensive and operationally sound

“Please review the plan (attached) in its entirety. What am I not thinking of? Anything I’m missing that would be important to get this project implemented successfully?”

🚧 Figuring out likely bottlenecks and failure modes

“What likely bottlenecks do you see? Which part of my plan is most likely to not work out as planned?

📅 Sanity-checking your timelines

“What is your take on the overall timeline? Which estimates seem too conservative or optimistic?”

🤖 Asking for ideas to leverage automation or AI

“Which parts of the project would be a good fit to outsource to AI? Please propose how to break larger work streams into tasks of manageable scope and write example specs and prompts for these."

And when you get stuck during execution, you can always ask AI to help you figure out what to do next.

Identifying your blind spots

Having an AI sparring partner available on demand is great; but you likely don’t want to pull them into an important meeting or problem-solving session. To excel in those settings, you need to make sure you derive generalizable learnings and sharpen your “unaided” thinking.

Some of these learnings will happen organically over time; the more you have AI (or anyone else, for that matter) challenge you, the more you’ll start anticipating potential questions and critiques.

But it can still be helpful to get explicit feedback, and your AI thought partner is a great source for that as it will collect a lot of data on how you think over time.

Here’s an example prompt you can use:

"We've worked through several issues together. Based on your observations, what are my biggest blind spots or bad habits when it comes to problem solving?"

When I did this, ChatGPT told me (among a few other things) that I tend to be biased towards things I can quantify or measure, which is a fair point and something I should probably work on.

Note: Only do this if you’re prepared for some uncomfortable feedback. It can be pretty brutal to see someone list out all of your bad habits like this 😅

Which model is best as a critical thought partner?

Which model you use as your thought partner makes a big difference. Similar to Deep Research, each model has its pros and cons in this context.

However, while ChatGPT was the clear winner for in-depth research, you might be surprised to find that’s not the case this time.

Again, I’ve spent dozens of hours working through problems with these models, and I’m giving you insights on what actually worked from my real-life experiences, regardless of what official documentation or press releases claim.

Note: Although it’s a solid model, I’m not including DeepSeek in this comparison since it doesn’t have a Projects feature or an easy way to set up custom instructions in the UI. I’ve also excluded Perplexity since I find it to be a better fit for search than thought partnership.

Also, I’m only including one model per company (e.g. only o3 for OpenAI) to keep things manageable; in each case, I chose the one I found most helpful.

⭐ Overall recommendation

TL;DR: Grok is my favorite thought partner, but is terrible at image generation and lacks some collaboration features.

Grok is good to great at almost everything I’m looking for in a thought partner: It challenges you with logical arguments derived from first principles, and it explains its thought process very well. And instead of throwing out your drafts and suggesting “better” alternatives, it actually engages with your content and provides targeted suggestions for improvement.

The other models excel at specific things; so if these matter to you, give them a go. For example, Claude is the best at generating diagrams, and ChatGPT Canvas is great for iterative, version-controlled jam sessions on larger deliverables.

Below, we’ll do a deep dive into each tool’s strengths and weaknesses.

🧠 Reasoning & judgment

🏆 Winner: Claude; 🥈 Runner-up: Grok / ChatGPT

All of the models have reasoning capabilities that are sufficient to act as a critical thought partner. Most of the points they raise are valid, and conclusions are logically derived.

However, Gemini makes occasional smaller mistakes (e.g. creates issue trees that are not fully MECE or treating an issue as controllable when it’s not), and ChatGPT and Grok tend to overcomplicate things (e.g. suggest complicated scorecards when simple decision rules would be more helpful).

Claude is the most pragmatic model in my experience; unfortunately, it’s more difficult to collaborate with and doesn’t explain its reasoning in much depth (see below).

🤝 Collaboration

🏆 Winner: Grok; 🥈 Runner-up: ChatGPT

I want a thought partner that critically engages with my half-baked thoughts instead of forcing a completely different approach down my throat.

Grok is great at this. For example, when you share a draft framework you’re developing, it gives precise feedback and highlights any changes or additions, and it asks intelligent questions about any files (spreadsheets, presentations etc.) you’re sharing.

The biggest downside is that it doesn’t have real collaboration features that allow you to jam on something together, so you’e stuck going back and forth in the chat trying to describe what you want.

ChatGPT does have a dedicated collaboration environment (called “Canvas”), which is great for jointly editing a larger deliverable with AI; unfortunately, it’s worse than Grok at engaging meaningfully with your work and providing surgical feedback.

👉 TL;DR: No tool is perfect yet, so you’ll have to try multiple and see which one works for you.

💡 Quality & creativity of ideas

🏆 Winner: ChatGPT; 🥈 Runner-up: Gemini / Claude

In my experience, ChatGPT is the most creative brainstorming partner, regularly coming up with ideas I wouldn’t have immediately thought of. It also tends to offer quite a lot of them (sometimes 8 - 10) without additional prompting.

• Gemini and Grok tend to be a bit more conservative in their idea generation, mostly focusing on “obvious” suggestions. However, if you push back and ask for more creative inputs, they often deliver

• Claude provides good, creative ideas, but usually just a few, and it takes some back-and-forth to explore a problem comprehensively

🔎 Transparency of thought process

🏆 Winner: Grok; 🥈 Runner-up: ChatGPT / Gemini

As mentioned before, you need to understand your AI thought partner’s reasoning to be able to decide whether you should incorporate their feedback or not.

Grok makes this extremely easy; it walks you through its thought process in detail and explains where it agrees with you and where it wants to push back (and why):

ChatGPT and Gemini also do a decent job at this, but not quite at the same depth as Grok. Claude, on the other hand, tends to be very brief and I usually need to follow up to really understand its arguments (and underlying assumptions) in depth.

💬 Language & tone

🏆 Winner: Grok; 🥈 Runner-up: ChatGPT

Having a thought partner that challenges you is great, but if you get annoyed, you’ll stop interacting with them.

By default (or with very basic instructions), these chatbots talk very differently to you:

• ChatGPT and Grok quickly found a good middle ground: They highlighted the parts of my reasoning that were solid and used a constructive, “matter-of-fact” tone to push back on things that weren’t

• Gemini is fun to work with, but for my taste, it tends to be too encouraging by default and goes along with a lot of what you propose

• When I asked Claude to challenge me, it started to be pretty condescending (e.g. calling my thinking lazy) and talked like an influencer (“Now, here’s the uncomfortable question you don’t want to ask”). Out of all the models, it took me the most effort to get it to adopt a constructive tone that didn’t drive me crazy

🎨 Generating images & diagrams

🏆 Winner: Claude; 🥈 Runner-up: ChatGPT

I like to think in diagrams and frameworks, and I want my thought partner to be able to jam with me. That might not be the case for you; if so, feel free to ignore this section.

Either way, Claude is the clear winner here. It’s the only model that consistently generated neat, correct diagrams for me, and it does this well in various formats (e.g. PNG images, interactive HTML pages etc.).

An added bonus: It automatically creates the images as a version-controlled artifact, making it easy to iterate on them.

• ChatGPT does well with things like text-based issue trees and is happy to generate images, but anything complex usually has errors in it

• Gemini straight up told me it can’t generate images after failing 3 times; if you remember to click the “Image” button at the bottom, it will create one after all, but similar to ChatGPT they usually have issues

• Grok, unfortunately, is simply terrible at image generation; e.g. when I asked it to create a 2x2 matrix of options we’d discussed, it offered these two masterpieces:

Recap: Using AI as a thought partner

Having a sparring partner available on demand 24/7 is incredibly powerful. And not having to be afraid of sharing half-baked thoughts or work is even better.

If you’re like me and you’re used to solving most problems in your head, it takes some time getting used to working “out in the open” like that. But once you’ve configured the AI in a way that works for you and you’ve gotten over that initial hump, you won’t want to go back.

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage). Given this is an AI-focused post, the roles reflect that:

Replit: AI Chief of Staff | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Foster City, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $150k - $300k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: Flexible | 🚀 𝗦𝘁𝗮𝗴𝗲: Series B | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Khosla Ventures

Clay: GTM Operations | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: New York | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $140k - $160k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series B | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia

Notion: Chief of Staff, People | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NYC | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $185k - $215k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Index Ventures

Vanta: Head of Growth Marketing | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $250k - $294k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 10+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Craft Ventures

OpenAI: Growth & GTM Strategy - B2B Monetization | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $255k - $315k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 7+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Khosla Ventures, Thrive Capital, Coatue

Anthropic: Finance Analytics & BI, Finance & Strategy | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $265k - $315k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 6+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Menlo Ventures, Lightspeed

Attentive: Revenue Operations Analyst | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $80k - $126k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 3+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, IVP

Zip: Strategy & Operations Manager, Partnerships | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $130k - $170k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 2+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: BOND, CRV

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

The release of ChatGPT Deep Research was a huge “aha moment” for me. As a non-engineer, it was the first time that AI automated something end-to-end that was core to my job and that I was good at.

Instead of spending hours doing manual Google searches and compiling the results, I could now get a 20-page report in minutes. Mind-blowing.

But as always with AI, once I got over the initial excitement, I realized that the results were often pretty rough around the edges. Questionable sources, shaky methodology, barely formatted walls-of-text — you name it, I’ve seen it in a Deep Research report.

The potential was clearly there, but it didn’t seem like the promise of a “McKinsey analyst on demand” was entirely accurate. At least not without putting in the work to understand and address the weaknesses of the tool.

As I was going through this learning process, I was looking for guides online. But I was surprised to find that most of them were way too generic to be actionable.

So I decided to write the ultimate guide to AI research agents myself.

No hype — just my hard-earned lessons on what works, what doesn’t, and how to get the best possible output. Including actual side-by-side examples so you can see the impact of the best practices I’m proposing.

We’ll cover:

• How I think about the core use cases as well as limitations of DeepResearch

• Which AI research tool is best for what task (incl. cheaper alternatives to the $200 ChatGPT Pro plan)

• How to write effective prompts that give top-tier output every time

• How to get the most value out of the final report

And if you’ve been following the Operator’s Handbook for a while, you know it’ll be detailed. So grab a coffee and let’s dive in ☕.

P.S. This is the first part in a new series about how to get actual work done with AI; in the same in-depth, tactical style as other Operator’s Handbook posts. If you have topics you’d like me to write about, let me know.

Part 2 has recently been published; you can check it out here:

How I think about AI tools for research

AI research agents are incredibly useful — regardless of your job. “Research” might sound like something that’s mostly relevant in academia or certain entry-level roles; but once you look closely, I bet you’ll find that you spend a lot of time on this.

For example:

• As a Product Manager, you’re researching competitors’ products

• As a Founder, you’re learning about sales tax, payroll, or cap table math

• And if you’re in BizOps, your entire job is basically to ramp up ASAP on a new topic every week so you can make an impact and move on to the next thing

AI research agents can help with all of that and more. And they’re not just useful on the job.

Based on my experience, the high-level use cases are:

AI can reduce the time required to do these things by 80% - 90%. But: Do not completely outsource these tasks without oversight. If you blindly use the output for anything important, you’re guaranteed to have a bad time.

These AI agents, at least in their current versions, need serious handholding to produce something robust. And to play the role of the critical “manager”, you need to know what to watch out for.

Here are the five biggest issues to be aware of:

Issue #1: AI agents don’t ask for the context they need

As you’ll see in the detailed comparison below, most AI research tools don’t proactively ask for context and only use the information you provide in the prompt.

Like an overeager analyst, they jump straight into execution, even if they have no idea what you’re actually trying to do with the report.

And even if the agents do ask for context (e.g. because you told them to):

1. The questions don’t always cover all important aspects, and

2. They never follow up if your answers are insufficient

👉 What to do about this: Proactively provide as much context as possible about your situation and what you’re trying to achieve, and tell the agent to ask for anything that’s still missing

Issue #2: AI agents don’t know how to handle sources properly

I’ve observed a few major issues with how AI agents deal with sources by default (i.e. unless you tell them how to do better):

• ☝ Over-reliance on individual sources, resulting in biased assessments or tunnel vision

• 🧩 Mixing and matching data sources without highlighting caveats, creating a patchwork analysis that doesn’t make sense

• 🗑️ Use of low-quality sources (e.g. random Reddit posts from anonymous users)

• 📆 Outdated sources (e.g. data from 5 - 10 years ago even in areas where things are changing fast)

If you don’t watch out, you’ll end up with a fancy-looking report on a shaky foundation:

👉 What to do about this: Especially for data-heavy reports, you have to 1) provide guidance on what sources to prioritize and 2) request visibility into how sources were used.

Issue #3: AI agents lack access to a lot of high-quality data sources

Deep Research cannot access paid sources. However, depending on your area of research, that’s where a lot of the good data sits.

For example, for a lot of analyses in B2B SaaS, you might want to leverage data from LinkedIn’s API or market research firms. Right now, to do this, you’ll need to first download those datasets and then make them accessible to the AI.

In addition, AI agents can obviously only research what’s available online. If you’re looking into a nuanced, niche topic (e.g. a complicated legal matter), that’s a problem.

👉 What to do about this: In these cases, it’s best to treat Deep Research as a tool that can help you prepare for expert calls (so you know what questions to ask) rather than an end-to-end solution to give you a recommendation.

Issue #4: AI agents sometimes show mediocre judgment

The quality of reasoning in DeepResearch reports varies a lot; both between different queries as well as different tools (e.g. ChatGPT vs. Gemini). Sometimes I’m deeply impressed, and other times I feel like I’m reviewing the work of a high school intern.

In addition, you’ll find that the research agents often just rehash an opinion they found (e.g. in a blog post) instead of really thinking through an issue. In other words, you’re outsourcing judgment to a random person online.

👉 What to do about this:

• Most importantly, pick the most capable tool (more on that below)

• Make sure you review the research plan in advance and provide detailed instructions re: methodology

• Give critical feedback on the first draft to address any glaring issues (even if it makes mistakes, AI always has a good attitude and is happy to fix things)

Issue #5: The default output format is often not very helpful

By default, the research reports you get from ChatGPT and Gemini are massive walls-of-text that are difficult to make sense of.

Sure, you can get an AI summary; but in some cases, you do want to go deep. You just want information presented in a way that makes it easy to digest.

👉 What to do about this: Give instructions on how you want the report to be structured and formatted (e.g. adding a “TL;DR”, using overview tables etc.).

Alright, enough complaining; let’s talk about specific things you can do to get the most out of these tools.

1. First, we’ll dig deeper into the different tools and their strengths and weaknesses so you know which one is the best fit for your use case

2. Then, we’ll go through actionable best practices you can use to get high-quality outputs every single time

Which AI research tool is best?

You’ve probably heard of ChatGPT Deep Research, but it isn’t the only tool out there. Tons of other players released their own versions in the last few months.

The big ones are:

• Gemini Deep Research

• Perplexity Research

• Grok DeepSearch

• Claude Research

At first glance, they’re pretty similar. You drop a research prompt into a chat window, and 5 to 30 minutes later you get a detailed multi-page report.

However, the devil is in the details. I’ve tested these tools extensively over dozens of use cases, and there are massive differences in how they approach research and the quality of the outputs:

Note: I’m comparing the basic version of the models here so you know what you’ll get as a free user. I’ll cover the performance differences of the premium “advanced” models in a separate section at the end.

⭐ Overall recommendation

TL;DR:

• ChatGPT Deep Research is still overall the best tool for truly deep research out there; no other tool comes close to the depth and rigor at the moment

• Perplexity is great for brief, well-structured overviews on new topics

ChatGPT’s two biggest weaknesses are 1) how it selects sources and 2) that it creates massive walls-of-text with poor formatting.

Both of these are default behavior, though, that can be adjusted with the right prompts; we’ll get into that in the next section. In my experience, it boils down to this:

You can ask ChatGPT to use different sources or make the report pretty, but you can’t really prompt the other tools to be much more rigorous.

However, the $200 Pro tier is ridiculously expensive for individuals, and the free and Plus tiers have low usage limits. As a result, you might want to choose another tool as your “daily driver” based on the pros and cons discussed here and “save” ChatGPT Deep Research for your most important tasks.

If you already have a favorite, feel free to skip ahead to the prompting best practices.

💰 Pricing & limits

🏆 Winner: Grok; 🥈 Runner-up: Perplexity

Grok and Perplexity both have generous free tiers with multiple queries per day, making them the clear favorite for anyone trying to get Deep Research without a subscription.

In contrast, the free tiers of ChatGPT and Gemini have fairly restrictive limits per month, and Claude doesn’t have Research in the free tier at all.

As a heavy user, you’ll eventually face the question of whether you should upgrade to a premium tier. In addition to much higher usage limits, you supposedly also get access to better models and higher-quality outputs.

We’ll discuss towards the end of this section whether that makes a noticeable difference or not (spoiler: not really).

🤔 Research planning

🏆 Winner: Gemini; 🥈 Runner-up: ChatGPT

Especially if you don’t write a lengthy prompt with detailed instructions on how you want the agent to conduct the research, it’s key to understand what it plans to do before it starts.

For example, if you’re asking for a comparison of software tools, you might want to understand what criteria it plans to use to evaluate them so you don’t have to wait 5 - 10 minutes to get something that misses the mark.

Gemini is the only tool that shares its research plan by default. Among the others, ChatGPT, Claude and Perplexity usually provide one on request:

Grok, on the other hand, tends to ignore my requests and just proceeds with the research 🤷.

Note: To maximize the odds that the agent shares the research plan, ask for it both 1) in the initial prompt as well as 2) when answering follow-up questions.

🙋‍♂️ Context gathering

🏆 Winner: ChatGPT; 🥈 Runner-up: Claude

ChatGPT Deep Research is the only tool that reliably asks 3 - 5 questions for context, even when you don’t tell it explicitly to do that. Plus, the questions are usually on point (i.e. similar to what a strong analyst would ask).

The others do it when you ask them to in the prompt, although not reliably. Here are two things that have worked well for me:

• With Gemini, repeat your request to ask for missing context after the agent shares the research plan

• With Grok, include a request in your prompt to explain how [XYZ] applies to [my situation / my company]. It’ll then sometimes realize it needs additional information to do that

🧠 Reasoning & judgment

🏆 Winner: ChatGPT; 🥈 Runner-up: Perplexity

Reasoning and judgment are the foundation of a good research report.

If you don’t agree with the methodology the agent used or the conclusions it arrived at, it doesn’t matter how long or pretty the report is.

ChatGPT is the clear winner here; not only does it show strong judgment (e.g. when choosing evaluation criteria or deriving takeaways), it also gives strong recommendations with clear reasoning for how it arrived at that opinion.

All of the other tools (except Perplexity) regularly make questionable choices in their research approach, and it’s often unclear how they arrived at their recommendations. As a result, you have to read the reports very carefully and ask follow-up questions if you plan on using the output for anything important.

📖 Comprehensiveness

🏆 Winner: ChatGPT; 🥈 Runner-up: Gemini

ChatGPT and Gemini produce by far the most comprehensive reports; however, while ChatGPT typically goes deep on the important areas, Gemini often adds generic “fluff” that wasn’t requested in the prompt.

The obvious upside of ChatGPT’s massive reports is that you get much more detailed considerations for complex topics than with the other tools.

But it also means that you need to add instructions to improve the report structure and formatting (e.g. by adding summaries for each section), and/or get an AI summary for these reports if you need insights quickly. We’ll dig more into that below.

The other three tools suffer from the opposite problem; their reports aren’t really deep research. Even if they consider dozens (or in the case of Claude, hundreds) of sources, the output reads like a “TL;DR”, so don’t expect a lot of nuance, detailed comparisons or super tactical guides.

You can get somewhat more detailed reports by asking them to be “extra thorough” and “aim for at least [X thousand] words” (esp. from Perplexity), but it won’t reach ChatGPT levels.

✨ Report structure & format

🏆 Winner: Perplexity; 🥈 Runner-up: Grok

Perplexity and Grok both produce easy-to-read reports by default that make good use of bulleted lists and overview tables. If you’re looking for quick, well-formatted insights without having to tweak your prompt, these are the tools for you.

As mentioned before, the other tools need guidance in the prompt to produce something that’s easy to parse.

📚 Sources

🏆 Winner: Perplexity; 🥈 Runner-up: Claude

There are a few things that matter when it comes to sources for DeepResearch:

1. Whether it’s easy to select what (kind of) sources to prioritize

2. How many sources are being considered (to get a balanced overview that incorporates a wide range of opinions and data points)

3. The quality of the sources that the research agent selects

4. How easy it is to trace citations (so you can double-check the data behind arguments)

ChatGPT isn’t great at any of these (besides citations) by default. That’s not a massive problem in practice, though, since 1) its summaries and takeaways are often still on point and 2) it’s very responsive to feedback.

For example, when I asked for a market size overview of major EU markets, the first version used questionable methodology and data sources. But I was able to get a much better version in 5 minutes by doing two things:

1. Asking ChatGPT o3 to compile a list of the best data sources for market size estimation:

2. Asking Deep Research to re-do that part of the report based on this, and highlight where numbers from different reputable sources don’t align (+ provide hypotheses for why that could be).

Are the premium versions worth it?

After playing around with Deep Research for a bit, you might be wondering: Are the expensive premium versions worth it, or is the free tier enough?

Don’t worry; I bought all of them so you don’t have to.

TL;DR: In my view, only the ChatGPT premium version is worth it if you’re a heavy user. All of the other ones only make sense if you’re using the account for other things as well.

There are two main considerations here:

1. Request limits

Perplexity and Grok provide plenty of research credits in the free versions, so upgrading isn’t necessary for most users.

Heavy users of Gemini will likely hit a limit, and Claude doesn’t offer the Research feature for free at all; but in my view, you’re better off going with ChatGPT in that case.

2. Quality of outputs

In the ads for the premium tiers, you’ll see some pretty lofty claims around advanced reasoning, more in-depth research, or more sources / citations.

But is there actually a noticeable difference in practice?

In short: No.

• Reports from Perplexity Pro seem fairly similar to those of the free tier; I definitely didn’t get the advertised “10x more citations”. The main difference I noticed was that it produced a lot of fancy charts that weren’t actually that helpful

• Grok DeeperSearch also didn’t impress me. I ran the exact same queries on both models to do an apples-to-apples comparison; DeeperSearch did use about 2-3x more sources and in some cases produced a 30% - 60% longer report with added detail, but it got nowhere close to the depth of ChatGPT Deep Research

Even the premium versions still feel like summaries, while ChatGPT produces actually deep research that walks you through a topic exhaustively.

How to create a good Deep Research prompt

The quality of your prompt makes or breaks your research effort. If you submit a careless request, you’ll be waiting up to 15 minutes (or longer) only to discover that the half-baked mess you got back is unusable and you wasted precious Deep Research credits.

While each model has its quirks, the core structure of what makes a good Deep Research prompt is somewhat independent of which tool you end up using.

We’ll first go through the individual components, and then put it all together. Feel free to skip ahead if you want to get started ASAP.

Step 1: Clearly state your goal

What is the objective of your research effort, and what output do you want? As mentioned above, this could be an overview, a recommendation, a comparison, or a detailed step-by-step guide (or a mix of these).

• ❌ “Help me understand how SEO for AI search works”

• ✅ “Please briefly summarize SEO best practices for AI search engines, provide a checklist of concrete steps I can take to boost visibility of my content, and recommend a specific tool that can help me with this”

Step 2: Provide context

This step requires the most effort from you, but it’s also the most important. The magic of Deep Research isn’t that you get a 20-page summary of a topic — it’s that you can get a report customized to your situation as if you had a research analyst on your team.

Remember: For any bit of context that you leave out, the AI will just make an assumption or keep it generic so it applies to everyone.

What’s important depends on the situation, but you might want to include:

• 💼 Basic facts about your business or situation (e.g. product, business model, geography)

• 🙋‍♂️ Who the audience of the report is (e.g. you, the CFO of your company etc.) and what their level of familiarity with the topic is

• 🚧 What constraints you’re facing (e.g. things your company can’t / won’t consider)

• 🎯 What the downstream use of the report is (e.g. a specific decision you’re hoping to make)

Even basic context makes a massive difference:

• This is the super generic report Gemini created about VAT regulations and compliance without any context

• This is the much more concrete, actionable report thanks to one sentence of added context in the prompt (B2B SaaS startup from the US expanding to the EU)

Since it can be difficult to think of everything that might be relevant, you can ask AI to help you brainstorm. A simple prompt that has worked well for me (e.g. with O3) is:

I'm planning to generate a Deep Research report on [X] in order to [Y]. What context should I provide so that I get a customized, actionable report? Pretend you have no context from any prior conversations.

Lastly, if you want to be sure you didn’t miss anything, ask the Deep Research agent directly to get additional context from you:

Step 3: Specify what you want the output to look like

As mentioned before, the default output (esp. of ChatGPT) is a bit of a mess. If you want something that is easier to digest, you’ll have to specify that.

Some things that have worked well for me include:

Specify the content and structure of the document

Being explicit about the content and structure of 1) the overall document and 2) individual sections made a massive difference for me.

For example:

• Dictate any key things you want included in the deliverable (e.g. a specific comparison, templates, copy examples, code snippets etc.)

• Ask the agent to follow the Pyramid Principle and lead with the key takeaways or recommendations (both in the main summary and in each individual sub-section)

• Ask for an overview of the sources used, incl. what they were used for, their type (e.g. government vs. commercial vs. news outlet vs. newsletter / blog), the date they were created or updated etc.

Specify the desired format of the output

Just adding a few basic instructions to the prompt makes a massive difference in how the report looks.

• Ask to use bulleted lists and bolded text where appropriate

• State that you prefer tables over text summaries for any sort of comparison or overview

Step 4: Ask for the research plan and provide feedback

As you saw in the overview above, only Gemini reliably shares a research plan proactively. To prevent negative surprises, remember to ask for it before the agent kicks off any work.

For example, this is the proposed research plan I got from ChatGPT for a deep dive into vibe coding tools after explicitly asking for it.

Here are the key things I’d watch out for:

• Is the research plan comprehensive? Is there any analysis you want the agent to do, or a section you want them to write up, that’s missing?

• Do you like where the agent is focusing? If you have a particular area you’re interested in, make sure the agent is building its report around that

• Are there any implicit or explicit assumptions that you disagree with? If so, provide relevant context to plug those gaps

• Do you agree with the methodology (e.g. evaluation or comparison criteria)?

This is also a good time to specify if you want the agent to prioritize certain sources.

For example, you could ask to use data from independent third parties rather than company websites, or that you want only data that’s more recent than a certain cut-off date.

Optional: Give examples of what you want

If you have specific expectations (e.g. for what the output should look like), it can be helpful to give examples.

For example, you might be using Deep Research to automate the creation of reports you used to do manually. In that case, drop some best-in-class samples of your work into a document and upload it as context so that the AI agent can imitate them.

Putting it all together: What an effective Deep Research prompt looks like

Here’s an example of an end-to-end prompt that incorporates the tips from above:

And here is what ChatGPT produced:

• Account scoring guide

Pretty good, I’d say — and much better than what you would have gotten with a simple “create an account scoring guide” prompt.

Side note: As you can see, it promised to share a research plan before proceeding and then forgot about it. That’s why I recommend repeating the request when answering the context questions (which I didn’t do in this case).

🤖 How to quickly create high-quality prompts using AI

You can either design prompts yourself or ask AI to help (either to feedback your draft prompt or write one from scratch).

For example, if you ask ChatGPT o3 this…

…you get back this.

Note: I recommend not just blindly copy-pasting the AI-generated prompt, but tweaking it to your purposes. For example, you can see that o3 included a lot of very specific asks, and these might not be exactly what you’re looking for.

Either way, it’s a good starting point and faster than creating a prompt from scratch.

💬 Treating research as a conversation

Deep Research credits are limited, so crafting solid prompts is important. Still, I recommend to approach research with AI the same way you would with a human analyst.

There are several reasons why:

1. It’s much easier to feedback something than to imagine the perfect deliverable

Can you describe what the perfect analysis of a topic looks like? Every little thing it needs to contain?

Probably not. But once you see one, you’ll immediately know how it could be improved. In my experience, you get a strong result much faster by getting a first version and providing feedback than by obsessing over the perfect prompt.

This is especially true if you’re not an expert in the area. I often know so little about a topic that I don’t even know what questions I should be asking. So instead, I do the following:

1. First, I ask for a high-level overview with suggestions for follow-up deep dives

2. Then, after reviewing the initial report, I go deep on the most interesting areas one-by-one

This way, I avoid being overwhelmed by a 50-page report full of information I either don’t understand or don’t need. And with each deep dive, I get a better understanding and am able to refine my questions for the next “research job”.

2. The more prescriptive you are, the more you constrain the model’s reasoning

It’s the same as when you’re delegating work as a manager: You get some of the best deliverables when you give people some freedom in how they execute. For example, they might approach the problem from an angle you didn’t even consider.

If you give a super prescriptive research prompt, the best possible outcome is a report as good as what you would have created. But if you want to be positively surprised, you need to leave things a bit more open-ended.

3. A research report is like a house of cards: Without a strong foundation, everything else falls apart

Imagine you get an in-depth report for how to build an ML model for a certain use case, including checklists, timeline estimates, code examples etc., only to realize that you forgot to provide a crucial piece of context.

Not only is all of that stuff useless now; since the report is so detailed, you probably spent way too much time digging through it until you realized it was fundamentally flawed.

When it comes to complex topics, your initial focus should be on getting the core of the analysis right. Then, once you’re happy with it, you can ask the AI to create downstream deliverables like detailed project plans or anything else you need to actually implement the recommendations.

How to get the most value out of the final research report

As discussed above, ChatGPT Deep Research reports can be long; really long. Sometimes up to 20k words and more.

I don’t recommend reading them end-to-end the moment you get them, though. Instead, I’d treat them as a resource you can selectively reference whenever you need to go deep on a particular aspect.

To get a quick overview of the key points, it’s more efficient to feed the report back into ChatGPT (or another tool) and ask for a summary. And if you’re doing the research as part of a cross-functional project at work, I highly recommend asking for multiple summaries tailored to specific audiences (e.g. the PM, the Finance team etc.).

Lastly, humans aren’t the only ones who can get value out of Deep Research. You can also add the reports as context to any future AI conversations or projects:

Recap & outlook

DeepResearch is an absolutely mind-blowing feature — if you know how to use it. Hopefully this guide provides a shortcut to maximizing the value you get from it.

Over the next weeks and months, I’ll publish more in-depth guides like this. Subscribe below to get them in your inbox.

But don’t worry: I’ll continue doing non-AI content as well, and I have some meaty posts in the pipeline. Stay tuned!

Want more tactical AI guides? Keep reading with Part 2 of this series here:

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

Enterpret: Business Operations Manager | 🏠︎ Location: NY | 💼 Experience: 3+ YOE | 🚀 Stage: Series A | 🏛️ Investors: Sequoia, Kleiner Perkins

Pulley: Strategy & Operations Manager | 🏠︎ Location: Remote | 💰 Salary: $150k - $195k | 💼 Experience: 5+ YOE | 🚀 Stage: Series B | 🏛️ Investors: Founders Fund, General Catalyst

Alloy: Chief of Staff - Supporting CEO | 🏠︎ Location: NY | 💰 Salary: $225k - $260k | 💼 Experience: 6+ YOE | 🚀 Stage: Series C | 🏛️ Investors: Lightspeed, Bessemer

Decagon: Business Operations & Data Analytics | 🏠︎ Location: SF | 💰 Salary: $175k - $220k | 💼 Experience: 5+ YOE | 🚀 Stage: Series C | 🏛️ Investors: a16z, Accel, BOND

Decagon: Strategic Growth | 🏠︎ Location: SF | 💰 Salary: $200k - $280k | 💼 Experience: 3+ YOE | 🚀 Stage: Series C | 🏛️ Investors: a16z, Accel, BOND

Glean: Product Operations Lead | 🏠︎ Location: Palo Alto, CA | 💰 Salary: $185k - $235k | 💼 Experience: Flexible | 🚀 Stage: Series D+ | 🏛️ Investors: Sequoia, Kleiner Perkins, Lightspeed

Faire: Marketplace Strategy & Analytics Senior Associate | 🏠︎ Location: SF | 💰 Salary: $129k - $177k | 💼 Experience: 3+ YOE | 🚀 Stage: Series D+ | 🏛️ Investors: Sequoia, Lightspeed, Founders Fund

Perplexity: Finance Manager | 🏠︎ Location: SF | 💰 Salary: $180k - $230k | 💼 Experience: 5+ YOE | 🚀 Stage: Series D+ | 🏛️ Investors: IVP, NEA, Bessemer

Dropbox: Chief of Staff to the CEO | 🏠︎ Location: SF / Remote | 💰 Salary: $150k - $253k | 💼 Experience: 5+ YOE | 🚀 Stage: Public

📚 What I enjoyed reading recently

• The Next Great Distribution Shift by : A must-read on how to think about ChatGPT as the next distribution platform, and what we can expect to unfold based on similar patterns from platforms in the past

• Prompting is Managing by : An interesting contrarian perspective re: the recent MIT paper that stands out in a sea of “OMG I knew it, now MIT proved it” reactions

• Vampire Attacks: How Marketplaces Overcome the Chicken-and-Egg Problem by : A closer look at how many famous marketplaces grew by syphoning off users from more established platforms

• My (mostly) minimalistic AI setup as a Senior Engineer in Big Tech by : A good reminder that less is often more for everyone feeling overwhelmed by the number of AI tools

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

When you come across a new framework, what’s your immediate reaction?

Chances are, you’re not going to be excited. Many people have framework fatigue, and for good reason. For every type of problem or task, it seems like there are at least half a dozen frameworks to choose from. And on top of that, every consulting firm and influencer is trying to coin one themselves. Always claiming to be revolutionary, and always with a snappy acronym like RAPID or CIRCLES.

And even though there are so many frameworks, how often have you actually applied one of them and felt like it really solved the problem in front of you instead of just producing some pretty slides?

So I’m not surprised if you feel like you’ve had enough.

But at the same time, I’d argue that frameworks are one of the most powerful tools at our disposal. In fact, I credit a lot of my success at work to the application of frameworks.

How do you reconcile this? I think the dissonance comes from the fact that a lot of people use frameworks wrong. This article is trying to fix that.

We’re going to cover:

• Why frameworks are so valuable (and you shouldn’t ignore them)

• The problems with how people use frameworks

• The different types of frameworks and what they’re good for

• How to use frameworks effectively (incl. detailed real-world examples)

Let’s dive in.

Why frameworks are so valuable (and you shouldn’t ignore them)

In my view, frameworks provide two main benefits:

1. Frameworks make success repeatable

Using frameworks (correctly) allows you to de-risk your problem-solving. If you completely reject frameworks and wing every problem you encounter, you’ll probably get mixed results; sometimes you’ll stumble upon the right solution, and sometimes you’ll miss important things. Frameworks make it so that your success is easily repeatable.

In addition, it’s easier for others to follow along. The approach is clearly spelled out and everyone is speaking the same language, especially if they are already familiar with the same framework.

2. Frameworks provide a shortcut

At their core, frameworks are the insights and experience of some of the smartest people packaged up for you to borrow. Why would you pass up this opportunity?

You might arrive at these thoughts on your own, but using frameworks gives you a shortcut.

Sometimes, you have even been doing parts of what a framework suggests already, but having it formalized can still help “clean up” your partially-formed thoughts.

It’s no surprise that frameworks are at the core of how top consulting firms like McKinsey or BCG operate; they allow them to come in, quickly understand the core dynamics and challenges of a business, and frame clear recommendations.

The good news is: This stuff isn’t secret, and you don’t have to grind three years in consulting to learn it. That’s what newsletters are for 🙂.

The problem with how people use frameworks

Whether frameworks level up your work or give you burnout depends on how you use them. Here are the most common misconceptions I’ve observed that lead to subpar results:

Issue #1: You don’t always have to use a framework

This is a common theme especially with ex-consultants. If you’re used to thinking in frameworks, then there is an instinct to slap one on every problem you encounter.

But that’s not always needed. Good ol’ fashioned critical thinking is often more efficient; using a framework in these cases feels forced and overly academic and you might be viewed as someone who fits better into business school than a startup.

Issue #2: Frameworks aren’t a substitute for critical thinking

Many people use frameworks as a substitute for thinking through a problem on their own. They take a quick look at the situation, try to “categorize” it, and then just run through the motions of a framework they pick.

Without 1) deeply thinking through the problem first and 2) customizing the framework to the situation, this often ends up in disaster. For example, more candidates than I can count sabotaged themselves in case interviews I gave because the moment they decided on a framework to use, they got tunnel vision and stopped having a real conversation with me.

Frameworks are a great starting point, but you should never just blindly run through the motions or force it onto a problem.

It’s very similar to AI in that way: You shouldn’t outsource your thinking to AI completely, but rather use it as a help in your own reasoning process and critically engage with anything it generates.

Issue #3: Frameworks aren’t rules, they’re guidelines

Frameworks are designed so that they are comprehensive and “foolproof”. As a result, many of them have a lot of steps or components to them.

However, I’d argue that most frameworks have one or two simple key insights that create most of the value. Everything else is fluff around it.

For example, the Jobs-To-Be-Done (JTBD) framework is fairly complex and intimidating when laid out in its entirety:

But in my experience, 95% of the value comes from simply shifting your mindset to start with understanding the specific “jobs” customers need done, and then working backwards from there (instead of starting with your product or service and its features).

Sometimes it even feels intentional when people make frameworks seem needlessly complex and then offer consulting services around them (*cough* OKR implementation consultants *cough*).

But just because the framework is complex doesn’t mean you have to go through all of its steps. When people do that, it’s usually a sign that they haven’t deeply understood the purpose of the framework and how to apply it. It’s like reciting a definition from a textbook instead of explaining something in your own words.

Instead, pick the pieces that provide the most value and apply the best to your situation, and tweak them as necessary to fit your needs.

The different types of frameworks – and what they’re good for

At a high level, there are two types of frameworks:

1. General-purpose frameworks that give you high-level structure or best practices

2. Prescriptive frameworks that give you detailed instructions on what to do or what to think about for a particular use case

Imagine you want to assemble a piece of furniture. The first type of frameworks are like the tools you’re using; they do a specific thing, but they can be used for a number of tasks and in different contexts (e.g. you might be able to reuse the wrench that came with your furniture for something else).

The second type of frameworks are like the instruction manual that came with it. It tells you exactly how to do things: What part goes where, and in what order you should assemble them.

Both types of frameworks are helpful, just in different ways. We’ll double-click on each category next.

1. General-purpose frameworks

These frameworks help you think through a problem — or communicate something — in a structured way. In the process, they often help you reduce a problem to its essentials.

They are highly flexible and can be applied to a wide range of topics, but you’ll have to customize them based on the situation and “build them out” yourself.

Here are some of the most helpful ones (note: all of these have done a lot of heavy lifting over the course of my career):

Framework #1: Issue tree

This is my absolute favorite general-purpose framework. The idea is simple:

You take a problem and you break it down into its subcomponents. By organizing things in a tree structure, relationships are immediately obvious.

This can be applied in a ton of different contexts. For example:

Problem solving

An issue tree helps you to list out the various levers or options you can choose from to address a problem in the business.

You start by listing out the broad categories and then fill in more detail until you reach concrete initiatives you can prioritize.

Let’s look at an example:

• Imagine you are trying to increase the amount of time users spend on your app. That’s the goal you put at the top of your tree

• Then, you list out the high-level options you have. In this case, you can grow your goal metric by either 1) growing the number of users, or 2) increasing the time each user spends on the app

• Now, we break things out into more detail.

  • New users can come from a variety of different channels

  • Similarly, you can increase the time users spend by either 1) getting them to open the app more often or 2) growing the duration of each session.

By continuing this process of breaking out higher-level buckets into smaller ones, you create an issue tree step-by-step:

Investigations

Issue trees also provide an incredibly helpful structure for investigations into metric movements.

You can either do a purely mathematical decomposition of the metric you’re investigating (creating a “driver tree”), or you can create a set of hypotheses organized in tree form (called a “hypothesis tree”).

Both will yield similar results; here is an example of how you can investigate a revenue drop in B2B SaaS:

Decision-making

For recurring decisions, it can be helpful to lay out the key considerations in a tree (called a “decision tree”). The questions should be posed as “Yes” / “No” questions, and the most important considerations should be at the top of the tree.

Besides streamlining and standardizing the decision-making (e.g. when you want to delegate it to someone else) this also gives stakeholders transparency around how the decision was made.

For example, let’s say your team is regularly getting requests from stakeholders and they have to figure out whether to take them on or reject them.

A decision tree for this scenario could help set expectations:

Note: In my experience, creating actionable decision trees can be very difficult; for example, instead of simple “Yes” / “No” questions you might have to define what to do based on a spectrum (e.g. “If revenue impact >$X, then this, otherwise that”).

As a result, I’d recommend to only use this technique selectively in cases where you feel like the effort is worth it for standardization purposes.

Framework #2: MECE

MECE is more of a principle than an actual standalone framework. I think about it as a force multiplier: Once you adopt this technique, your frameworks will be 10x more impactful.

The core idea is that any time you break down a problem, that breakdown needs to result in buckets that are Mutually Exclusive and Collectively Exhaustive. This means:

1. None of the buckets or categories you choose can overlap (i.e. every single data point or example fits into just one bucket), and

2. They need to cover all possibilities.

This is especially helpful for creating clean issue trees:

I wrote an in-depth article describing how to apply the MECE principle in practice with tons of real-world examples as a guest post for ‘s High Growth Engineer newsletter; check it out here if you want to go deeper.

Framework #3: 2x2 matrix

The 2x2 matrix is simple but powerful.

To build one, you simply choose two dimensions that are relevant for your problem (those are your axes) and two variants for each (typically along the lines of opposites like “High vs. Low”), dividing your grid into four areas.

This simple visualization comes in especially handy when you’re trying to frame problems; instead of going in circles trying to consider every nuance, you’re just focusing on the essentials.

For example:

Let’s say you want to figure out if, and how, you should use AI for a task. To simplify this decision, you could categorize tasks based on:

1. Whether they are mostly execution-focused or require judgment / decision-making

2. Whether the stakes are high or low

The resulting grid then gives you four interaction models with AI depending on the task at hand:

Framework #4: Pyramid principle

The Pyramid Principle is the most effective communication framework I know. It’s powerful for a few reasons:

1. It makes sure the key point of your message gets across

2. It helps you be perceived as a senior contributor or leader, regardless of your actual level

3. It can be applied to basically anything

The Pyramid Principle doesn’t tell you what exactly to say; it just tells you to start with the key insight or recommendation and then add supporting arguments and data (to the degree they’re necessary to support your point).

• After an investigation, tell people whether the outage is fixed (and what you’ll do if not), then what caused it; only get into the details of your investigation if asked

• Give the CEO the recommendation first, then talk about alternative options and their pros and cons

• In an interview debrief, say first whether you’re for or against the hire, then list out your reasons

This is the exact opposite of the default approach many of us take; we go through our analysis and all the challenges we encountered along the way, and by the time we get to the takeaway, everybody already stopped paying attention.

Unlearning this takes some time, but it’s the most impactful change you can make to how you communicate.

Framework #5: 5 Whys

When you’re looking into problems at work, there are two common (related) challenges:

1. You’re fixing a symptom, but the root cause remains unaddressed

2. You’re completing a stakeholder’s request, but you didn’t have context on the actual problem they were trying to solve, so the result is subpar

Both issues can be addressed by asking “Why?” repeatedly.

Here’s an example:

With the additional information on why things are happening, you can address the actual root cause instead of coming up with bandaid solutions for the symptoms.

For example, you could propose that Marketing should also have a target on the conversion rate of their leads.

The key with all of these general-purpose frameworks is to practice, practice, practice until they become second nature.

These days, for example, I don’t actively try to make something MECE anymore. It’s my default, and if something isn’t, it gives me mental (and sometimes physical) discomfort.

This means:

1. For any problem you come across, always try to uncover the “true” underlying root cause. Then, try to create an issue tree to structure the problem

2. Test if your framework is MECE: Brainstorm examples and see if they fit into one, and only one, of the buckets you created

3. Come up with 2-3 different ways to frame the problem (or parts of it) using a 2x2 matrix

4. Use the Pyramid Principle every time you communicate your work. Share a minimum level of supporting data; then, notice what follow-up questions people ask and anticipate these next time

2. Prescriptive frameworks for a particular use case

The first category of frameworks consisted of flexible general-purpose tools —Swiss army knives, if you will. They are incredibly useful, but their biggest advantage is also their biggest weakness: They don’t really tell you what to do — only how to do it.

As a result, there is a huge category of frameworks that cater to specific use cases and give you prescriptive, sometimes step-by-step, instructions. They are easier to use, but less flexible and often feel more “forced” when applied to a problem.

Many of these frameworks are simply more specific versions of the first category. For example, there are several well-known frameworks such as the BCG Matrix or the Ansoff Matrix that are just 2x2 matrices tailored to a particular use case. Similarly, the popular Opportunity-Solution-Tree is just a specialized version of the issue tree presented above.

The biggest mistake people make when applying these frameworks is to study them in-depth and then apply them to the letter.

As discussed earlier, this approach often creates more harm than good. Instead of just checking all the boxes, try to think of these frameworks as:

1. 🤔 A starting point for your thinking, and

2. 🚧 A safeguard to ensure you’re not missing an important consideration completely.

Here are some of the most helpful ones:1

Substack won’t let me create a table in the editor, so I am adding links with more detailed information on each framework in the footnotes.2

How to use frameworks effectively

The general process for using a framework effectively looks like this:

1. Deeply understand the problem

2. Decide what output you need and choose the appropriate framework

3. Customize the framework to your use case

Step 1: Understanding the problem

I’ve written about this in more detail here; but the gist of it is that if you don’t fully understand the problem in front of you, you can’t come up with the best solution.

As a first step, you need to understand the problem in its entirety. Then, you simplify it by focusing on the factors that matter most.

At the very least you should consider:

• What is the actual problem you’re trying to solve?

  • Is there an underlying root cause you can trace the problem back to? (“5 Whys” from above comes in handy here)

  • What are different ways to frame this problem?

• What is the cost of being wrong?

  • If a decision is easy to reverse or the cost of being wrong is low, you can use a simpler, more streamlined problem-solving or decision framework (or not use a framework at all)

Once you’ve worked through enough problems, you’ll start to put them into certain “buckets” in your head. For example, after working on a few launches, any time you hear the words “new product” or “expand into new market” you might think “Oh, another launch problem; I know how to deal with that!”.

That’s perfectly fine — as long as you still go deep on the problem to see if there is anything unique about this situation.

Starting with a mental model based on experience is great, but no two problems are 100% alike, so you still need to go through the next steps to pick and customize the right framework instead of copy-pasting what you did last time.

Step 2: Deciding what output you need and choosing the right framework

Not every framework does the same thing.

• Some frameworks change your thinking or help you view a problem in a novel way

• Some create clarity on process or roles and responsibilities, or they give you a standardized process you can follow

• Others help you compile a list of experiments to try or areas to investigate

As a result, you need to work backwards: Decide first what you need, and then pick the framework that will be the most helpful in producing that.

For example, let’s say you need to make a decision. Which framework you should use depends on your needs:

Step 3: Building out your problem-solving approach iteratively

Most problems are too complex to solve them end-to-end by applying a single framework or analysis. As a result, you’ll have to pick a starting point and then iterate from there.

This can mean using a formal framework as a basis and adding some custom problem-solving on top, or it can even mean combining multiple frameworks.

For example: Imagine you want to figure out how to improve profitability for a business.

1. As a first step, you could create a tree that breaks out profits into revenue and cost, then cost into the individual fixed and variable cost components etc. This will give you a good idea of the different drivers you could optimize

2. However, you still need to brainstorm concrete initiatives to actually drive improvements

3. And as a third step, you’ll need to decide which ones you should focus on. For this prioritization exercise, you could then plot the different initiatives into a 2x2 matrix with expected impact on one axis and effort or time to impact on the other

A hands-on example of frameworks in action

Now, to make it really tangible what effective use of frameworks looks like in practice, I’ll go through a real-life example. I’ll focus on an interview case I used to give since in these situations I was able to see how somebody solves a problem end-to-end.

However, all of the points I’m making here apply equally to problems you’re solving on the job.

The question I would give candidates was along the lines of:

“Imagine you’re on the BizOps team and working with the Head of Customer Support. The team is heading towards their annual peak season and they want to make sure they’re prepared. Specifically, they want to avoid building up a ticket backlog.

How would you approach this problem?”

There were three types of candidates:

Approach #1: Straight into execution (no framework)

Many candidates would immediately start brainstorming concrete things they could do. For example, they’d throw out ideas like “We could use AI to automate a lot of the tickets" or “We should improve training and create better templates for common problems”.

While some of the ideas were actually pretty good, they weren’t structured in any way.

This always caused one or several of the following issues:

• The lack of structure meant they would run out of ideas a few minutes into the case

• They didn’t ask for more context, so a lot of the ideas were fairly generic and it was unclear which ones would be most impactful to the team’s specific situation

• It wasn’t clear if they had actually solved the problem since they didn’t have a structured way to tie their ideas back to the “zero backlog” goal

Approach #2: Going through a generic framework from A to Z

The second type of candidate would think for a moment, maybe ask a clarifying question or two, and then announce which framework they would use.

The frameworks they chose were typically generic ones that provided a broad structure for their brainstorming. For example, several candidates chose the “3P Framework”, which allowed them to group their ideas into People-, Process- or Product-related ones.

While this gave more structure compared to the first group of candidates, the approach still had major issues:

• Organizing ideas by qualitative buckets still didn’t help us understand if, and how, we’d get ticket backlog to zero

• Candidates usually went through their entire framework, wasting valuable time and making it feel like they were checking the boxes rather than solving a problem

• The rigid structure of the frameworks made a real conversation difficult, and candidates usually struggled to incorporate any new information they learned along the way

Approach #3: Developing a custom framework on the spot

The most successful candidates all took the time to deeply understand the problem and then tailor their approach to it.

In this case, they would realize that preventing a backlog meant the support team’s capacity during peak season had to equal the time required to solve all the tickets that come in on a given day.

They would then figure out that to solve this problem, they would 1) first need to forecast the expected number of tickets during peak season and compare that to the team’s current capacity to understand the potential gap, and then 2) evaluate the potential levers they have to plug that gap:

1. ⬇️ Reduce the # of tickets the team will receive during the peak period

2. ⬆️ Increase the # of tickets solved per hour

3. ⬆️ Increase the total # of hours worked

Most candidates that got to this point would then realize that this type of problem is an ideal fit for an issue tree and would start building one out.

While you’re building out an issue tree, it can be helpful to apply other frameworks to brainstorm ideas. For example, some candidates used the 3P framework mentioned above to brainstorm initiatives specifically to reduce the number of tickets (the right branch in the tree).

Prescriptive frameworks aren’t bad per se; it’s just that they are rarely the right first step in solving a problem.

Finally, once the issue tree is built out, what’s left is to prioritize the most promising levers to plug the expected capacity gap. To do this, you could use a framework like Meta’s traffic light matrix to compare your options across multiple dimensions (e.g. expected impact, time to implement, what would happen after the peak season is over etc.).

It doesn’t really matter what exact approach you choose; but the strongest candidates all had a structure for evaluating their options in a principled way.

The bottom line is: There isn’t just one right way to solve a problem. However, there are key considerations that you need to think through to get to a robust solution. And whether — and how — you use frameworks plays a big role in how easy it will be to do that.

TL;DR

I’ve relied heavily on frameworks throughout my career. Without them, it would have felt like I had to reinvent the wheel for everything I’ve been doing.

As with everything in life, frameworks aren’t inherently good or bad. They’re just a tool, and whether they level up your execution or result in generic output that misses the mark depends on how you use them.

Hopefully, this article helps you unlock their value and turn them into a competitive advantage.

P.S. Given the sheer volume of examples I had to whip up for this post, I’m sure I made a mistake somewhere. Hit me up if you find something and we can nerd out about frameworks.

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

Blank Street: Senior Product Strategy Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $100k - $130k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 2+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series B | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: General Catalyst, Tiger Global

GlossGenius: Strategy and Operations Manager, Payments | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $145k - $180k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Bessemer

Writer: Chief of Staff, CRO | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF, Chicago | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: Flexible | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: ICONIQ, Radical Ventures

OpenAI: Business Operations, API Pricing | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $265k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Thrive Capital, Khosla Ventures, Coatue

Anthropic: Finance & Strategy, Product | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $230k - $310k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 7+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Menlo Ventures, Lightspeed, Bessemer

Faire: Chief of Staff, CFO | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $158k - $218k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 6+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Lightspeed, Founders Fund

Stripe: Strategy & Operations, Card Networks | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $186k - $280k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 10+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Sequoia, Founders Fund

Anduril: Special Projects Manager, Air Defense | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Irvine, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $108k - $162k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Founders Fund, Thrive Capital

Airbnb: Programs and Business Operations Lead | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $164k - $204k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 10+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Public

📚 What I enjoyed reading recently

• How custom GPTs can make you a better manager by / : A highly actionable deep dive into how to leverage AI to scale yourself as a manager

• Math to Evaluate Sales Reps & GTM by : A brief, helpful overview of efficiency metrics for sales reps

• AI's Missing Multiplayer Mode by : A brief preview of what more collaborative AI tools might look like in the future

• How to use the AI-boom to level up in your career and land a job by and : Practical tips on how to leverage AI with concrete prompt examples. Most of it is helpful to engineers and non-engineers alike

• An analytical guide to people-watching in NYC by : It’s people watching season, and this is your guide to NYC’s best spots

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

When you look at discussions around AI developments and their impact on jobs, there are two main camps that dominate the conversation.

Camp #1 is in denial. It sounds something like this:

“AI will never replace what I do. Real quality and craft will win against AI slop!”

Camp #2 are the doomsday prophets. They keep repeating versions of:

“We’ll all be out of jobs! In 5 years, there’s nothing AI can’t do. Better say ‘Please’ and ‘Thank you’ to ChatGPT to get on the good side of our new overlords!”

Neither extreme take is 1) accurate nor 2) helpful in my opinion.1

• If you think AI will never replace what you do, you become complacent (which will actually make it more likely that AI will take your job)

• If you think there’s no hope, you’re giving into a defeatist attitude — instead of seeing yourself as response-able, you are waiting for someone else (e.g. the government, or the AI labs) to do something.

Both attitudes are equally passive.

Look, I get it: The progress of AI is dizzying, and I’m not sure I am excited about a world where Clippy will eventually be smarter than me.

But denying or ignoring AI progress — or throwing up your hands in defeat — won’t help.

So, what can you do?

In my view, the only viable strategy is to proactively engage with AI, understand its capabilities and shortcomings, and lean into the things where you currently have an unfair advantage.

But what are those?

Where AI is at a disadvantage (for now)

First things first: I am not trying to predict the development of AI here; nobody can predict the future, and I have no insider information. Anything that AI can do today, whether at the level of an entry-level analyst or a seasoned expert, will only get better, and new capabilities will be added as well.

However, there are things where AI is at a structural disadvantage simply because it is, well, AI — something that (today) exists purely in the digital realm and relies on data we feed it. None of these obstacles are impossible to overcome, but models simply getting “smarter” won’t be enough; it would require very fundamental changes to how companies operate, and that will take time.

Based on that, and over a decade of experience seeing how work actually gets done in the real world at companies from startups to FAANG, I picked a few areas where I expect humans to continue to add a lot of value for at least the foreseeable future.

AI challenge #1: Limited data access & context

Generally, AI is good at things that it has a lot of training data on. And since the web is full of data on almost anything, AI is (becoming) good at a lot of things.

However, what AI is lacking is organization-specific context. Even if you gave it access to all your internal systems and documents: Anyone who has ever worked anywhere knows that a lot of important information and context only exist in people’s heads.

The problem is exacerbated by the fact that AI is not (currently) in the habit of asking a ton of questions to gather context. Even when you give custom instructions to do this, chatbots like ChatGPT will typically only ask for limited surface-level information before starting to generate a recommendation or deliverable.

And even if it did ask for more context: Who wants to answer a series of 57 questions in writing? And can you actually articulate everything that matters explicitly and concisely?

Now, the more we integrate AI into our daily work and the more tasks are handled (or supported and observed) by AI instead of humans, this data gap will decrease. But at least for now, this is putting AI at a disadvantage.

AI challenge #2: Not all work is digital (yet)

AI can execute a lot of tasks — writing code, creating visuals, drafting documents etc. — at tech companies well because deliverables are completely digital and execution can be done by one person (or AI agent) end-to-end.

However, the hard part of any large initiative at work isn’t this execution — it’s aligning different teams and keeping them in sync, getting executive buy-in, deciding ownership and securing resourcing etc.

The larger the company, the more important (and challenging) this is. And then there are roles that are almost exclusively based on human interaction — Sales, for example. As long as most decision makers are human, there will be a need for humans to influence them.

Even if AI can make convincing arguments on paper: A chat transcript doesn’t sell things, people do.

Again, if we ever reach a point where large chunks of companies are run by AI agents, then a lot of these human friction points (and the need for people to handle them) will disappear.

But as long as you have humans in key positions, the difference between a failed project and a success will be how well you navigate these situations.

AI challenge #3: There is no alpha in commoditized intelligence

Large financial returns aren’t generated by average companies, but by outliers. These companies did something differently than the competition; they took a different approach to building product, positioned themselves differently, unlocked novel distribution channels, or took an unusual approach to execution. Or they created an entirely new market altogether where others didn’t see an opportunity.

Similar to how in the stock market all public information is already priced in, leaving no opportunity to generate outsized returns from it, using the same publicly available AI tools as everyone else is unlikely to give you a competitive edge (unless you have large, unique datasets to train them on that others don’t have access to).

For some companies, that might be acceptable — in fact, “good, but average” results might be better than what they’re getting right now. But those that want to win in competitive markets will want an edge over others, and you don’t get that from putting things on the same autopilot everyone else is using.

It’s the same reason why you don’t outsource any core competencies like product strategy or marketing today, and why it makes more sense to bring in McKinsey to help cut costs instead of hoping they’ll create the next breakthrough product for you.

AI challenge #4: Accountability

Like it or not, executives love having a single person responsible for an area of the business or a critical initiative.

When they are not happy with the progress, they have someone they can drag into a meeting room and grill on turnaround plans.

But if ChatGPT messes up your product launch, what are you going to do? “Fire” ChatGPT and never use it again? Sue OpenAI? Or — god forbid — take responsibility yourself?

This will not necessarily protect entry-level IC positions, but it does suggest there will be a continued need for “managers” overseeing work (whether that’s done by humans or AI).2

What skills and traits provide a competitive edge in the age of AI?

Now, based on the above, let’s talk about what concrete skills you should focus on to maximize the value you bring to an organization.

In my view, it comes down to two key areas:

1. Having strong judgment and convictions

2. Being able to influence people to get stuff done

1. Having strong judgment and convictions

AI is good at solving somewhat straightforward problems with a clear right answer. Where a lot of the value lies (and will lie going forward), however, is in making good decisions in the face of incomplete data and ambiguous situations.

Even if AI gets better at this; if 1) everyone has access to the same AI solutions and 2) AI can’t really be accountable the same way a person can, then developing good judgment — and building a reputation around it — will continue to be a key differentiator.

You’ll need this both to handle things that executives are not comfortable handing off to AI as well as to critically evaluate AI outputs.

But what does good judgment actually look like?

In a nutshell, good judgment is about making good decisions, recommendations or assessments. To do this, you interpret hard facts and data through the lens of situational context and personal experience.

There are a few “ingredients” to this:

Knowing what “great” looks like (and why it’s great)

You can’t properly evaluate something if you haven’t seen examples of excellence in that area.

For example, if you want to judge whether an email drafted by AI (or your team) is good, you need to 1) have seen examples across the entire quality spectrum and 2) form an opinion on what makes some better than others (and a few much better than the rests).

This will always be subjective to a degree. Call it “taste”, if you like.

The best way to develop this is to work directly with, or for, people at the top of their craft. The second best way is to study their work and try to understand how they operate and think (e.g. through courses, podcasts, and newsletters).

Knowing what works and what doesn’t

The vast majority of AI models people use are trained on publicly available data.

The problem: Similar to how scientific journals only publish findings with statistical significance, companies mostly write about successful projects (and only the final approach that worked, not the 100 things they tried before that didn’t).

In addition, the crucial details and nuance that matter when you’re actually trying to operationalize it are left out.3 The only way to learn this is through hands-on experience.

Case in point: When I try to work with an AI like ChatGPT through complex problems like building a marketing attribution model, it gets the general approach 95% right. And even when asked about potential challenges, it does fairly well and rattles off most of the obvious ones — almost as if it has experience doing this.

But when I keep pushing, it becomes clear the whole thing isn’t “thought through”. It’s the kind of proposal you get from a new grad that has learned about a topic in school but has never actually done the work.

As long as you know what you’re doing, you can actually iterate your way to a decent result4, but if you just asked an AI to “run with it”, you’d get something half-baked that won’t hold up in the real world.

So, to build out a competitive edge, you need to:

1. Do the work yourself (i.e. don’t just outsource it to AI to make your life easy)

2. Work through all of the setbacks and challenges you’ll inevitably encounter

3. Over time, identify what approaches work and which ones don’t

4. Use those insights to pressure-test your own proposals and those of others (incl. AI)

If that sounds like a lot of work, that’s because it is. But unfortunately, there is no shortcut.

Knowing what drives business impact vs. is a waste of time

Execution isn’t the only blind spot AI has. Another area it (currently) shows poor judgment is when asked to evaluate whether a project idea is even worth pursuing.

Given its lack of data on what projects led to business impact vs. which ones went nowhere, it cannot give a robust assessment here.

I keep a list of “Types of project to avoid” based on years of painful lessons, and I asked ChatGPT (4o and o3) to evaluate a few of them.

Given the project ideas I pitched all sounded reasonable at surface level, the responses I got were encouraging across the board, despite custom instructions to challenge me.

Just because a project feels important and interesting doesn’t mean you should do it, though. In fact, if people describe projects like this, I try to avoid them like the plague as it’s typically a sign that they don’t know exactly how it will drive impact (besides satisfying curiosity).

Being able to boil things down to the essentials

AI is getting better and better at giving detailed overviews or generating in-depth research reports.

However, it takes a lot of custom instructions, prompting and iteration to get to really clean, simple frameworks that boil complex decisions down to the essentials. Doing that, and knowing how much you can simplify without losing important nuance, requires judgment.

To practice this, first try to frame complex topics in the most simple terms possible. For example:

• Framing decisions as a simple trade-off between two things

  • E.g. growth vs. profitability, speed vs. accuracy, etc.

• Reducing an issue to two key dimensions and plotting it on a 2x2 matrix

  • E.g. grouping data science projects by business impact and effort

• Expressing something as a function of a single (most important) factor

  • E.g. showing the ideal team structure based on team size

Then, see what decision you’d arrive at and, if you’re not happy with the result, add back nuance and complexity step-by-step until you reach a level that seems sensible.

Forming strong convictions

Finally, using your judgment, it’s important to form strong convictions.

Let’s say you’re working through a problem and, after a careful analysis, you reach your conclusion.

Now, to actually move forward, you’ll have to convince other people of your take. Throughout this process, people will challenge you, bring up additional considerations, and give their own opinions. Without strong convictions to guide you, you’ll be easily swayed by all this new input and start going in circles.

To be clear, I’m not saying you shouldn’t listen to feedback. But in ambiguous situations, you need someone to cut through the noise and advocate for something they strongly believe in.

This is what makes working with AI so frustrating at times. With no skin in the game and no real experience to draw from, it’s often a bit of a pushover. And while you can instruct AI to stand its ground and give strongly worded recommendations, it’s not the same as having actual convictions grounded in lived experience.

You can see the same difference between a consultant that doesn’t have to see the work through vs. the operator that actually needs to stand behind their recommendation and implement it.

2. Influencing people to get stuff done

As mentioned before, AI is rapidly getting better at creating outputs (code, documents, images etc.).

But the actual execution work is only a small portion of any project:

For example, when we built an account scoring model in my last job, writing the code for the first version of the model was actually the easy part. We spent the majority of the time 1) getting everyone to agree on what exactly to build and then 2) getting people across multiple teams comfortable with the outputs of the model.

These were also the phases of the project where it was most at risk of failing.

A model is completely useless if it’s not integrated into any critical business processes

So if the SDRs and AEs didn’t trust the model, they wouldn’t make decisions based on its output, and our work wouldn’t have driven any impact.

And at least for now, this process of getting people bought into your work and driving adoption is a deeply human challenge that requires an intentional approach and lots of context.5

Context about people

The more you know about how people operate, what they care about, and how they think, the easier it will be for you to work with them — and ultimately get their support or sign-off.

There are a few different aspects to this:

What different people respond to

It’s easy to fall into the trap of thinking that other people think like you. And that as a result, the best way to convince them is to give a presentation that you would find convincing.

However, I’ve learned that’s not true. For example:

• When I worked at Rippling, our founder and CEO always wanted raw data in monthly updates so he could dig into the details and get unfiltered insights. He rejected any slides that had a narrative “spin” on the data

• At Meta, on the other hand, everything that went to executives was packaged up into a clean narrative. Data was quoted selectively to support arguments in the summary, but no raw data tables were shared

The thing is: You can only get this context through direct exposure. These meetings are often not recorded or documented in detail, so an AI won’t have any data on this. You won’t either if you’re not in the room, so jump on any opportunity to listen in.

What incentives and priorities people have

People will ultimately always act in their own self-interest. As a result, understanding each person’s incentives is key so you can frame your proposals as a win-win (or at least make sure you don’t make their lives harder).

This goes beyond the metrics they are goaled on. For example, my team at Uber was officially responsible for both 1) keeping supply and demand balanced, and 2) staying within efficiency guardrails.

However, in practice, nobody cared if we spent more than budgeted. We did, however, get grilled whenever the marketplace was undersupplied — even if just for a few hours late at night.

As a result, we cared way more about reliability than efficiency. We didn’t say that out loud or documented it as a principle for the team, but everyone who’d been around for a while knew it to be true.

Context about process

How teams operate and work together

For almost any impactful project, you have to collaborate with other teams.

In addition to understanding their incentives and personal styles, you also need to understand how to work with them from a process perspective.

This goes way beyond official org charts (that an AI is aware of).

For example:

• How do you get the data team’s support for the important experiment you want to run?

• Do all requests have to be surfaced during quarterly planning, or are they flexible enough to adjust mid-quarter?

• If they don’t have the bandwidth but you need the support, how do you escalate the issue?

Or:

• How do the Marketing & Sales teams align their plans to make sure the overall Go-To-Market plan makes sense?

• If the first version of the plans does not ladder up to the top-down goals from Finance, how do you iterate your way to a version that works?

I’d argue that given that execution is becoming commoditized, knowing how to navigate these processes will be one of the key differentiators of high performers going forward.

How decisions are made

When you’re trying to get a project off the ground, or move to the next phase, you often need to get approval from leadership.

Doing this well can speed up progress by weeks — if not months — and will massively reduce the risk that your project will just get “stuck” at some point and never see the light of day.

Since it’s such an important topic, I wrote a separate deep dive on it.

Interpersonal relationships and reputation

Finally, even if you know how to convince another team in theory and you are doing everything “right”, you’ll often hit a wall.

You submit a request and it gets deprioritized. You escalate, but it goes nowhere. The only reliable way to avoid this is to invest in strong relationships with your key partners that allow you to “cash in” on the credit and reputation you built with them when you need to.

The best way to build up goodwill with another team is to be helpful to them first. If you show that you’re willing to be pragmatic and help them out, circumventing official processes and red tape, then they’ll be much more likely to do the same for you.

Interestingly, since both gaining context and building strong relationships take time, it might incentivize people to stay in their roles longer than they do today (especially in more mature companies where things are more stable, both in terms of “how things work” as well as the people you work with).

What this looks like in practice

Besides the challenges discussed at the beginning of this post, there is also a much simpler reason why AI (currently) struggles with these things at work: They are hard.

But when someone does have strong judgment and is skilled at influencing stakeholders, what does that look like?

You get:

• The Data Scientist that:

  • Can tell whether a stakeholder request will result in business impact instead of just a pretty dashboard that nobody looks at (and can successfully push back on things that would be a waste of time)

  • Knows what needs to happen for a predictive ML model to be trusted by business stakeholders

• The Strategic Finance analyst that:

  • Knows that adding more features and complexity to the model will just be false precision and create more of a black box, which will erode trust in the forecast

  • Can pinpoint and challenge the weak points in the GTM plans based on experience and get teams to take on more realistic goals

• The BizOps manager that:

  • Designs a planning and operating cadence that creates a helpful amount of structure and discipline without restricting operating flexibility too much

  • Gets Marketing and Sales to align on a joint attribution model despite different incentives and political tension between teams

• The Legal team member that:

  • Has a good sense for how much exposure and regulatory risk is acceptable to take on in exchange for faster growth

While nobody can predict the future, none of these profiles are about to be replaced by AI tomorrow. So if nothing else, leaning into these skills extends the “runway” of your current career for a few years.

Closing thoughts: Being adaptable is key

In the areas I’m discussing above, you can still add value at the moment.

However, AI is developing rapidly and what is a human competitive advantage today might not be as valuable anymore a few months or years from now — at least from an employer’s perspective.

That means it’s unrealistic to expect that you’ll be able to keep doing your job the way you’ve done it in the past indefinitely.

Even if it doesn’t get replaced completely, it will keep changing, and if you want to be successful in that new world, you need to lean into the change and adapt.

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

Vannevar Labs: Business Operations | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $150k - $170k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series B | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: General Catalyst, DFJ Growth

Talkiatry: Sr. Associate, Strategic Finance | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $140k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, General Catalyst

Mercury: Chief of Staff to the CEO | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NY / Portland / US remote / CAN remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $175k - $243k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Sequoia, Spark Capital

dbt Labs: Marketing Strategy & Planning Manager | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $144k - $194k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Sequoia, Google Ventures

OpenAI: Sales Strategy & Operation | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $265k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 7+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Thrive Capital, Coatue, Khosla Ventures, Microsoft

Faire: Revenue Operations Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $88k - $121k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 2+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Lightspeed, Founders Fund

Relativity Space: Business Operations Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Long Beach, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $93k - $119k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 1+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: ICONIQ, Coatue, General Catalyst

Ethos Life: Strategy & Business Operations, Senior Associate / Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $85k - $138k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 2+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Google Ventures, Accel

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

When I told my professors at MIT that I was joining Uber, they all told me that I was making a big mistake.

“That business will never make money”, they said.

What they meant, however, was “Uber is losing a lot of money right now, and it doesn’t seem to be getting better, so it’s probably a bad business model.”

So why did I have confidence in my decision when all these smart people were telling me I was jumping on a sinking ship?

Because I’d interned at Uber over the summer and had seen 1) how the business was operating at the time, and 2) all the things it could improve — over time and at larger scale — to eventually turn profitable.

Based on these insights, I’d created a simple scenario model that had shown me the potential of the business and what needed to happen to get to profitability. As long as I believed that those things were realistic, I’d be fine.

And it turns out, Uber did start making money:

Scenario modeling is a powerful tool, and once you get the hang of it, it’s not particularly difficult to do.

In this post, we’ll dig deeper into 1) what scenario modeling is, 2) why it’s useful, and 3) how to do it.

After that brief primer, the majority of the post will be focused on an in-depth real-world example to show you what it looks like in practice.

Let’s dive in.

Part 1: What is scenario modeling, and what is it good for?

In the example in the intro, running a scenario analysis helped me decide whether to accept a job offer or not. But you’ll be able to use it on the job, too — once you know how to do it, it comes in handy all the time.

But what exactly makes scenario analyses worth the effort?

The simple reason is that execution is hard, and it’s often costly to reverse bad decisions.

Even if you know something is possible and what you need to do, it’s going to be difficult to actually do it. But if you don’t even know that, you’re adding a massive additional risk — you might be wasting your time entirely or focusing on the wrong things.

A scenario analysis helps by:

• Telling you whether it’s even possible to get where you want to go

• What the key drivers are that you need to be on top of

• What numbers you need to hit for things to work out (so you can track progress along the way)

For example, when I was at Meta, a common question for every new initiative was:

“What would it take to make this a $1B business?”

If we didn’t have confidence that it could eventually reach that level, we would focus our limited time and resources on bets with higher potential.

Similarly, when I worked in B2B SaaS, we were wondering: “What has to be true for us to profitably scale an Outbound motion?”

Figuring this out in advance allowed us to structure our approach in a way that would maximize the odds of success (what customers to target, how many people to hire etc.) and focus on the most important things we needed to get right.

In all of the examples above, we’re trying to figure out how to make something work. To do that, we’re working backwards from a goal (e.g. profitability). However, there is a different — forward-looking — kind of scenario analysis as well: Starting from today, you can project different scenarios for the future.

By using different sets of assumptions (e.g. aggressive vs. conservative), you can get an idea of where you might be headed. This allows you to prepare in advance; for example, you can calculate out how many people you need to hire at a minimum to keep up with future growth.

This article will focus on the first kind of scenario analysis (working backwards from a goal); I have covered some basic principles of the forward-looking approach in my post “How to estimate”, but might go deeper on this in the future.

Part 2: How to do scenario modeling

The high-level sequence of steps for this type of analysis is as follows:

• Step 1: Define the goal

• Step 2: Lay out the driver tree

• Step 3: Assign baseline values to each driver

• Step 4: Quantify the gap

• Step 5: Identify key drivers & sensitivities

• Step 6: Find scenarios that allow you to hit your goal

We’ll go through each step in more detail below.

If you want to jump into hands-on examples immediately, you can skip directly to the next section (Part 3) and come back here as needed.

Step 1: Define the goal metric

The first step is to define the goal you’re trying to hit.

Your goal needs to be clearly defined and quantifiable. For example, it’s not enough to say “I want to make this market launch successful”.

What does “successful” mean in this case, and how do you break this down into metrics?

A better definition of your goal in this case would be something like “I want to grow this new market to $10M in ARR”. That way, you have a (somewhat) clearly defined metric and a number you need to hit.1

Step 2: Lay out the driver tree

Once you have your goal metric, you need to decompose it into its key drivers. This will form the basis for your analysis.

I won’t lay out a full driver tree here since we’re getting into a detailed example below, but even the first few levels give us a pretty good idea of the levers we can pull:

Step 3: Assign baseline values to drivers

To limit complexity, you won’t want to play with a range of assumptions for every single driver. But you also can’t completely leave them out of the analysis; otherwise, your result won’t be accurate.

As a result, you need to assign baseline values to each driver; then, in the next steps, you’ll model out assumptions for a select few.

The baseline values can come from a number of sources:

Step 4: Quantify the gap

Now, since you assigned starting values to all drivers, you can calculate how close (or far away) you are from your goal.

This delta can help you decide which drivers are worth focusing on. For example, if you see that you need to improve your target metric by 3x, there’s no point looking into drivers that can only realistically improve by a few percentage points.

Step 5: Identify key drivers & sensitivities

Based on your understanding of how much of a gap you have to bridge, you should now identify your most promising drivers.

The key considerations are:

• 1️⃣ Which drivers are directly correlated with your goal?

  • Generally, you’ll want to focus most of your effort on global drivers, i.e. those that affect the entire business instead of just a subsegment (e.g. a specific market or marketing channel)

• 2️⃣ Which drivers have the biggest headroom?

  • As mentioned above, the more of a gap you have to your goal, the more you need to focus on the drivers with the biggest room for improvement

• 3️⃣ Which drivers are under our direct control?

  • The less directly you control the driver, the higher the risk that your assumptions won’t come true in the real world

  • For example, if your scenario model relies on organic website traffic growth, you are at the mercy of the Google algorithm if you want to hit your goal

• 4️⃣ Which drivers can you move with the least cost?

  • Moving some drivers has negative consequences for the business, either immediately or in the future

  • The lower the cost (or negative consequences) of moving a driver, the more likely you are to get buy-in for your plan

Step 6: Find scenarios that allow you to hit your goal

As long as your goal isn’t impossible to achieve, there will be an infinite number of different paths to reaching it.

In my experience, picking scenarios that represent a “random” combination of assumptions (a little bit of an improvement here, a small change in performance there) isn’t very helpful as it doesn’t give you a clear strategic direction.

Instead, try to tease out one or two distinct scenarios that make it clear where the priorities should lie. For example, you might find that there is a driver that is 1) directly under your control and 2) has a lot of headroom, and improving this one driver alone could get you to your goal.

That is a clear message that can help you get the team aligned on the path forward. We’ll get into examples of what that looks like in practice below.

Lastly, at this stage of the exercise, you might notice that the target seems impossible to hit. If that’s the case, there are two options:

• Option 1: You have a very wrong assumption somewhere

  • To rule this out, ask other people to sanity check your work — when you’re heads-down modeling, it’s easy to miss the forest for the trees

  • For example, your numbers might be too low because you only looked at the US market and forgot about the potential of international expansion.

• Option 2: The target is actually impossible to hit

  • While this might not be the result you were hoping for, it’s still valuable. Instead of spending months trying to make the initiative work, you now have a quantitative argument for why you shouldn’t invest resources in it

Part 3: Hands-on scenario modeling example

Before we get into it, I want to make something clear: Scenario modeling is a flexible tool you can add to your toolkit and it can be applied to a wide range of problems.

Your work doesn’t have to be at all similar to what I’m describing here for it to be relevant to you. The purpose of this example is simply to illustrate how to apply the technique in practice.

Example: Running a profitable food delivery marketplace

Scenario

Let’s say you’re working a food delivery company like DoorDash, UberEats, etc. The company is currently losing money on each transaction, but obviously, the goal is to eventually turn profitable.

For simplicity’s sake, let’s define “profitable” as generating a positive contribution margin with your average order, defined like this:

Let’s take a closer look at each bucket.

• Revenue (the stuff you make money from):

  • Fees you charge the end user (mainly a % fee applied to the basket size)

  • Fees you charge the restaurant (typically a % fee applied to the basket size if there is a contract with the restaurant)

• Cost (the stuff you have to pay for):2

  • Courier pay

  • Cost of unfulfilled orders

With that, we have our goal metric (“profitable contribution margin per order”) and a simple driver tree (steps #1 and #2 from the framework above).

Quantifying the status quo

With our drivers laid out, let’s assign a baseline value to each to understand where we’re at compared to our goal.

In this example, since we’re already operating a live delivery marketplace, we’d be able to pull this data from our internal data warehouse.

What should jump out at you: We are currently losing $2.85 on every order. Keep in mind, that’s before we even try to cover any of the fixed costs of the business.

It’s easy to see that something fundamental has to change.

For reference, what we just did corresponds to steps #3 and #4 from our framework above.

Figuring out how to hit the goal

Now, let’s look at the levers we have and see which ones have the potential to get us to our goal.

At a high level, we could either increase the amount of money we make from each order, or reduce cost.

However, we can rule out a few options pretty quickly based on the priorities of the business:

• End-user fees: We don’t want to raise the % fees we charge end users as it will hurt demand

• Restaurant fees: Selection is a key part of our strategy, so we don’t want to renegotiate with existing restaurants or increase fees for new ones

However, that doesn’t mean we don’t have options on the revenue side:

Over-the-top (OTT) orders

Above, we called out that we are only getting money from restaurants that we have a contract with. In our example, that percentage is 50%.

The other restaurants are those we added to the platform without an explicit agreement (called “Over-the-top” or “OTT”); this helps us offer better selection to users, but hurts our unit economics.

We could either 1) stop doing OTT orders or 2) try to convince these restaurants to enter contracts with us based on the order volume we provided to them. Given that we want to maintain strong selection, our priority should be option #2, but this would take longer to show an impact.

Potential impact: Up to $3.00 (in other words, stopping OTT orders completely would turn the contribution margin positive)

Basket size

We could also do various things to increase average basket sizes, which would result in more revenue per order even if we keep restaurant and user fee percentages the same.

For example, we could offer promotions and discounts that incentivize users to order more, and/or we could add up-sell steps to the check-out process.

Potential impact: Every $1 increase in basket size would increase our contribution margin by $0.25; i.e., we’d need a $11.40 (+57%) increase in basket size to get to a neutral contribution margin.

That seems like a stretch without materially changing our restaurant selection or user base, so we should look at cost next.

Reliability cost

If we can’t find a courier to deliver an order, we have to refund the end user. In addition, in cases where we have a contract with the restaurant, we have to pay them for the food that was prepared and not picked up (and won’t charge a fee).3

This is costly, but it only happens occasionally. While we should continuously work on improving reliability, the savings are limited to $1 per order (if we brought the rate of unfulfilled orders to 0).

Given that we need to improve our contribution margin by $2.85, let’s focus on higher-impact levers.

Courier pay

Courier pay rates are set dynamically by the system based on supply and demand.

Since the pay structure in this example is based on distance, we could try to reduce the average trip distance to save money. However, the main way to do this would be to shorten the delivery radius for restaurants and since this would hurt demand, it should be a last resort.

So are we out of options on the cost side? Not exactly.

While we can’t reduce the cost of a delivery trip by much, we can increase the throughput of these trips.

But what does that mean exactly?

Most of the time, our couriers are picking up and delivering single orders. Sometimes, they pick up two or three at the same time if 1) the orders are from the same or close-by restaurants and 2) the users who ordered live in the same direction.

This is called “batching” orders.

If we assume that the pay for the courier remains roughly the same whether they deliver one or multiple orders on a trip (as long as the distance doesn’t change materially), we can reduce the effective cost per order by increasing the batch rate.

The key trade-off that we have to watch out for is that if we batch too aggressively, users will have to wait longer for their orders which might lead to reduced engagement or churn.

Potential impact: Right now, we deliver 1.1 orders per delivery trip on average. If we increased this to 1.5, we’d save ~$1.85 per order.

At 1.9 orders per trip, we’d break even without changing anything else.

Summary

Now, where does that leave us?

Before we did the scenario modeling exercise, we only knew that we were losing $2.85 per order. However, we didn’t know if, and how, we could solve that issue.

Now, we have two clear paths to breaking even and eventually turning a profit:

1. Over-the-top (OTT) orders: We could either stop doing OTT orders or make a concerted effort to enter contractual agreements with these restaurants so we can charge a fee

2. Batching: We should invest in 1) increasing demand density in key areas and during key times (to have more orders eligible for batching) and 2) improve our batching algorithm; this will reduce costs per orders without affecting demand

With these insights, we now have concrete initiatives we can rally cross-functional teams behind instead of trying to tweak everything we’re doing a little bit and hoping that it will be enough to make the business work.

TL;DR

Many people blindly jump into execution without fully understanding the problem they’re trying to solve.

While it’s important to move fast, this approach is almost guaranteed to fail. To maximize your chances of success, you should know what you’re trying to achieve and have at least a rough idea of 1) whether it’s theoretically possible and 2) how you can make it happen.

Decomposing your goal into a driver tree and doing a simple scenario analysis might cost you a few hours, but it can save you weeks of wasted effort. Or, conversely, it can prevent you from missing out on opportunities just because they look impossible at first glance.

That’s time well spent, if you ask me.

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

Traba: Strategy & Operations - General Manager (Special Projects) | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $135k - $150k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series A | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Founders Fund, General Catalyst, Khosla Ventures

Talkiatry: Senior Manager, Business Operations | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $135k - $165k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z

EarnIn: Strategy & Ops Manager | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Mountain View, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $146k - $179k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 6+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Ribbit Capital

OpenAI: Product Operations Manager, ChatGPT | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $180k - $240k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: Flexible | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Thrive Capital, Khosla Ventures

Anthropic: Strategy Business Partner, Applied AI | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $190k - $270k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Lightspeed, General Catalyst, Bessemer

Scale AI: Strategic Finance Manager, GenAI | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NY | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $112k - $140k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 1+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Founders Fund, Index Ventures

Lyra Health: Sr. Associate, Growth Strategy, Provider Ops | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $95k - $146k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Coatue, Greylock, IVP

Airbnb: Senior Program & Business Operations Lead | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $200k - $247k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 12+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Public

📚 What I enjoyed reading recently

• Data and Defensibility by : An epic essay on what constitutes a data moat (and what doesn’t)

• There Are No New Ideas in AI… Only New Datasets by : A deeper look at what has actually been driving progress in AI

• The MCP Revolution by : MCP continues to be a hot topic in AI, and this post provides a good overview of what it is and why it matters

• [Paid] Tech hiring: is this an inflection point? by : Solid overview of the recent developments in hiring (e.g. fully-remote loops becoming almost impossible due to use of AI)

• What should Allbirds have done differently? by : A brutal (and fun) analysis of Allbirds’ rise and fall

• How to turn chaos into clarity with Investigation Docs as an engineer by and : A spiritual successor to my post on MECE structure on Jordans’ newsletter. If you’ve ever been in a metrics war room, you should read this.

Usually, every week or two, I share actionable advice to help you grow your career and business, based on operating experience at companies like Uber, Meta and Rippling.

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

Being a lone wolf isn’t really a viable strategy in any organization.

I have seen many smart, capable people get passed over for promotions because they weren’t able to work well with others or couldn’t keep cross-functional projects on track. Some were even managed out.

And it makes sense: Almost everything we do at work requires collaboration, regardless of your role.

• Marketing needs support from Data Science and input from Design, and needs to coordinate growth plans with Sales

• Data Science needs to gather business requirements, relies on Data Engineering for pipeline work, and needs to partner with stakeholders for implementation

And so on.

Collaboration is already critical at the IC level, but it becomes table stakes once you’re a manager. And as with many similar topics, it’s more complex than it might seem at first glance.

That’s what the “Nailing the basics” series is all about: Helping high performers get to the next level by going really deep and providing tactical advice on deceptively simple topics.

As a reminder, this is part 4 in the series; check out the first 3 posts if you haven’t yet:

Part 1:

Part 2:

Part 3:

In this post, we’re going to cover:

• How to identify and manage cross-functional dependencies

• 8 tactical tips to make collaboration successful

• How to become a person people want to work with

Let’s dive in.

Part 1: Identifying and managing cross-functional dependencies

The first — and arguably most important — step is to be intentional about collaboration.

If you just “let it happen”, it will often end in chaos. To avoid this, you need to make an effort to figure out 1) when you need input from others or 2) when your work affects others.

Then, you need to manage that dependency in a structured way.

Figuring out where your work intersects with other teams

There are several ways you can identify these dependencies:

Understanding what’s up- and downstream of your team’s work

If you’ve ever worked at a large company, you’ll be able to relate to this:

When I worked at Uber, I was in charge of managing the supply / demand balance in many of our US markets. To do that, we were using several internal tools to analyze data, automate processes etc.

One day, we got an all-company email that one of these tools was going to be deprecated. This caused widespread panic; dozens of core workflows were leveraging this tool, and there was no viable alternative available.

When we followed up with the team owning the tool, it turned out that they had no detailed knowledge of how people were using it. They had assumed all major use cases were covered by other tools, but hadn’t actually confirmed that.

This, to me, is one of the key symptoms of dysfunctional collaboration between teams (or the complete lack of any collaboration).

If you don’t know how your work fits into the bigger picture — i.e., which teams’ work you’re relying on, and who is leveraging your output — now is the time to find out.

And since context like this often gets lost when people switch teams or leave the company, consider taking a few minutes to document it once you do.

Scoping projects

In cases like the example above, one team is consistently producing or maintaining something that another team needs. But that’s not always the case; many times, cross-functional dependencies come up as part of one-off projects.

In these cases, it’s crucial to scope the project as early as possible to identify what you need from others.

That doesn’t mean you actually have to start working on the project right then and there; you just need to figure out who you need to reach out to so you can do it early.

If you don’t, you risk causing an unnecessary fire drill for another team.

Reviewing other teams’ priorities

Make it a habit to read other teams’ planning decks or roadmap documents to see if there is any overlap or dependencies.

You’d be surprised how often teams commit to doing something that they can’t deliver without your (or your team’s) help, without reaching out first.

Similarly, you might find that another team is working on something similar to what you’re doing and you should join forces or at least coordinate your approach.

This is a valuable exercise even as an IC. Your manager will not always be on top of these things, and it shows that you see the big picture and think strategically (which will come handy during promo time).

Build in public

If you wanted to get married in Germany before 1998, you had to publicly announce that in advance so people could voice any concerns.

Why? Because back in the day, it was difficult for the church or authorities to know whether there were any reasons you shouldn’t be allowed to get married (e.g. because you were already married elsewhere). Crowdsourcing this information was easier.

It’s the same at work. It’s hard to identify every single person that your projects affect, so the easiest way to surface potential issues is by making your work public.

Some ways you can do this:

• 💬 Share your project roadmap on Slack

• 🎤 Present during all hands meetings

• 📅 Bring up your current priorities in meetings with people from other teams

Instead of you having to do all the legwork, people will now come to you.

Managing dependencies

Once you know that your work is an input to other people’s work or vice versa, you need to manage that dependency to avoid bottlenecks.

The more complex and important the project is, the more you should rely on formal structure and process.

Step 1: Create a workback plan

You don’t have to create a detailed GANTT chart in Excel, but you need to clearly document what needs to happen by when, and who is responsible.

To do this, you work backwards from the final deadline:

1. 🪝 Who is on the hook for the final deliverable?

2. 🙏 What do they need from others to put that together?

You then keep repeating this until you get to a point where there are no more upstream dependencies (i.e. the work can be done without input from others).

Here’s an example of what that looks like in practice:

• Sales needs to deliver their headcount plan and quota roll-up to Finance

• To do that, they need Marketing to provide a Leads forecast

• Marketing, in turn, needs Data Science to forecast certain metrics like website traffic

That means for Sales to be able to meet their deadline, two other teams need to deliver.

To make sure that happens, you need to work backwards and agree on dates for when Marketing and Data Science will complete their work.

If you expect there to be controversial points (e.g. the Marketing forecast might be too low for the Sales plan to work out), schedule early review meetings ahead of the actual deadline so that there is time to make changes, if necessary.

Step 2: Avoid bad surprises

Set clear expectations

The most common failure mode I’ve seen in cross-functional projects comes from misaligned expectations: One team delivers what they think the other team needs only to realize it’s not exactly what they were looking for.

The core of the issue is that people use terms like “analysis” or “forecast” as if they were standardized. For example, if someone says they’ll give you a “marketing forecast for France”, what do you think you’ll get?

• A spreadsheet of monthly Leads in aggregate?

• A daily CSV with Qualified Opportunities split by marketing channel?

• Something else entirely?

The only foolproof way to avoid this is to write down the exact specs of what will be delivered. This can feel like micromanagement, but nobody wins if you’re talking past each other.

Maximum transparency

The second most common issue is that people align on what needs to be done once in advance and then don’t talk to each other again until the deadline.

• If you’re waiting for an important deliverable from another team, set yourself a reminder a week or two before it’s due and check in with them

• If you’re on the hook for something, give regular proactive updates to your stakeholders. If something comes up, flag this early so there’s time to find alternative solutions

What not to do:

❌ “Sorry, I wasn’t able to finish this by today. There was an issue with the data pipeline, and Data Eng took a week to fix it, and then we couldn’t get feedback from our Director in time, and also…”

Explanations like the one above are common, but not helpful. You might feel the need to explain what happened, but that’s irrelevant at this point.

Instead, flag the issue as soon as it comes up and focus on what can be done:

✅ “There is an unexpected issue with the data pipeline which will delay the project. However, to keep the original timeline, we can switch to another data source. We won’t be able to do the analysis on the country-level, but everything else can be done in time. Does that work?”

Part 2: Making collaboration successful

Similar to effective communication, there are many small tweaks you can make to how you work with others that make a big difference.

Especially if you’re working remotely (or with people across different offices), it’s important to address things that could cause friction or confusion.

Here are 8 concrete things you can do:

Tip #1: Make your time zone and working hours visible

It sucks when you quickly need an answer from someone only to realize they are already offline because they are three hours ahead. Similarly, you don’t want to get 7PM meeting invites on a Friday.

To avoid this, set your calendar to your time zone and configure your working hours:

In addition, add your timezone to your Slack name (or whatever tool you’re using) so people know during what times you’ll be responsive:

Tip #2: Be responsive

When I was busy, I used to avoid incoming messages completely; I didn’t feel like I had the time to truly digest the message and answer thoughtfully, so I didn’t respond at all.

However, over time, I realized that there is a middle ground: If you skim the message to see if it’s urgent, acknowledge that you saw it (e.g. by using the 👀 emoji) and jot down a quick reminder to get to it later, you can close the loop for the sender (and get them off your back) without wasting more than a minute of your time.

If you want to set expectations with regards to when you’ll properly respond, you can send a simple “Seen this, will respond by tomorrow morning”. If that doesn’t work for the other person, they’ll let you know.

Note: To be clear, I am not saying you need to check every Slack notification in real-time; focus blocks without distractions are valuable. But when you do see a message, give a quick indication you’re aware instead of ignoring it until you have the full answer ready.

Tip #3: Be hyper-explicit and overcommunicate

You can’t complain somebody missed your point if you buried it in a wall of text. Make sure the important parts are 1) front and center and 2) phrased hyper-explicitly.

Here’s what that looks like in practice:

If you say the above, you’re not going to get a preview in time. The request is not worded strongly enough, and nobody feels responsible for it.

Here’s what to do instead:

Lastly, repeat yourself. It might seem weird at first, but in my experience, if you don’t repeat something at least 2 - 3 times, there’s a high chance people will simply forget about it. So:

• Start meetings by recapping recent discussions or upcoming milestones

• After meetings, share decisions and action items via Slack

• Start messages with reminders like “Since we decided to do XYZ…”

Tip #4: Use screenshots and recordings for things that are complex to explain

Many things are difficult to explain in writing, or it simply takes too long to type it up. In these cases, you can often get the point across quickly by using a screenshot or a screen recording.

For example, let’s say you’re giving someone feedback on a homepage redesign. Instead of describing what you want them to change, just take a screenshot and highlight / annotate the relevant piece:

Tip #5: Get sign-off in writing and document it

If you don’t have proof somebody agreed to something, it didn’t happen. In practice, that means if someone approves something verbally (e.g. in a meeting, or in the hallway), you need to follow up and get confirmation in writing.

And make sure you keep a permanent record of it; I learned that lesson the hard way when I got important approvals via Slack, only to realize the messages were auto-deleted 60 days later.

If you need sign-off from multiple people, you can add a tracker to your proposal:

Tip #6: Don’t gatekeep information

Gatekeeping information doesn’t make you important — it just creates unnecessary friction.

Unless something is highly confidential, it should be shared as openly as possible. It’s already difficult enough to get information to flow through a large organization if you actively want it to, so don’t make it even harder.

This means:

• Be overly generous with who you copy on emails

• Minimize the use of 1v1s and DMs and discuss important things in public channels

I’m not alone with that opinion. NVIDIA CEO Jensen Huang, for example, actively discourages 1v1 meetings:

Tip #7: Protect what you don’t want other people to touch

If you’re collaborating with others on a spreadsheet, there’s a risk they’ll mess everything up. And you can’t really blame them if you don’t make it clear what’s off-limits.

Even better, just eliminate the risk altogether by locking the cells you don’t want people to edit:

Otherwise, you’ll find yourself having these conversations over and over again:

Tip #8: Use version control intentionally

First things first, unless you’re an investment banker in the 2000s, there’s no excuse to send Excel files via email. Google Sheets / Slides does the trick in 99% of cases.

However, if you’re collaborating with multiple people, it gets difficult to keep track of changes in documents.

Luckily, there is a version history feature in Google and Microsoft 365 tools. Almost nobody uses it (I’d estimate about 1% of spreadsheets I’ve seen have named versions), so it’s a great way to stand out.

To make your (and others’) life easier:

• Name important versions in the history

• Make a copy of really important ones and create a simple tracker to manage them

Part 3: Be someone people want to work with

Being someone that people want to work with pays off — in fact, I’d argue it’s one of the major drivers of accelerated career trajectories.

Here are some things I’ve seen happen first-hand:

Benefit #1: Skipping the line

If you build a reputation for being easy to work with and helping others out, you can “skip the line” in the future.

For example, you can ping a data scientist that you have strong rapport with directly instead of filing a ticket through the official intake form.

Benefit #2: Access to high-profile projects

Not every project is staffed through your manager. Many of the most impactful opportunities I have gotten (and have seen others get) were the result of executives and managers from other teams approaching me because we had successfully worked together in the past.

But how do you become someone people want on their project or team?

In addition to the tactical tips in the section above, there are a few key traits that make a massive difference:

1. Get work done

A shortage of ideas is rarely the bottleneck in a project; it’s usually execution. If you want to add value to a project team, you need to roll up your sleeves and get things done.

There are two ways you can do this:

1. 📋 Structuring and framing problems: Efficient execution requires clarity, so simple frameworks, decision trees or summaries that boil things down to the essentials will get you noticed

2. 🚀 Executing work independently: An occasional meeting for alignment or a problem-solving session as a group can be helpful, but what people really want is for someone to just raise their hand, take a part of the work stream, and get it done

Volunteering for unpopular work (e.g. reviewing a bunch of support tickets for common themes) can be a great way to build a reputation for being a “doer” early on. With AI, you should be able to get this done fairly quickly, but people will be thankful nonetheless.

This doesn’t mean you should become the go-to person for this kind of stuff, but it helps to show you’re not above doing what needs to get done.

2. Be accountable

It’s almost a cliché at this point, but it’s worth repeating: There are few things as powerful as saying what you’re going to do and by when you’re going to do it, and then doing that.

Besides building trust with your stakeholders, it also makes you more productive.

I’ve found that whenever I am not making progress on a deliverable as much as I want to, simply setting a realistic — but ambitious — deadline and letting a stakeholder know (so that there’s external accountability) gives me the focus and motivation boost to actually do it.

If something comes up that truly makes it impossible to deliver, come up with a mitigation strategy and tell your stakeholders.

The focus shouldn’t be on what happened, but how you’ll deal with it (see below).

3. Always be response-able

Being accountable is easy when things are going well. But when things get challenging, you have a real opportunity to stand out.

Most people give up and blame external factors. While that’s understandable, it’s a defeatist attitude. And the time you’re spending trying to cover your ass and coming up with an explanation for why you couldn’t deliver, you could have spent finding an alternative solution.

The best articulation of this that I’ve seen comes from Fred Kofman’s book “Conscious Business” (which was somewhat required reading at my last job at Rippling). The concept he introduces is “response-ability”.

You might not be responsible for the delay or additional challenges in the project (i.e. it’s not your fault), but you are still able to respond.

“You are not responsible for your circumstances; you are response-able in the face of your circumstances.” - Fred Kofman, Conscious Business

4. Respect other people’s time

Everyone at work is always strapped for time, so the most valuable thing you can do for other people is to respect theirs.

Here are two concrete things you can do:

Making it easy for others to help you

Reaching out to others for help is not a big deal as long as you’re respectful of their time.

Here’s what a bad ask looks like:

❌ “The number of Leads between Sales and Marketing reports doesn’t match. Can you help me understand why?”

This type of ask is simply lazy.

• What data source are you looking at?

• Are all the numbers different, or just a few? How big is the difference?

• What have you already investigated and ruled out?

Without asking a bunch of follow-up questions, nobody can help this person.

Here is what to do instead:

Yes, this is much longer, but it saves a lot of back-and-forth by adding all relevant information, and makes it much more likely you get a response.

Do the heavy lifting

There are many cases where a few minutes of work on your end can save other people a lot of headaches.

For example, when you’re editing a document in Google Docs, the other person will be bombarded with dozens of notifications and highlights.

Important questions or changes get buried in a mountain of small copy tweaks or changed font sizes. But there’s an easy fix: Send a separate message summarizing the important stuff.

5. Celebrate wins and share credit (& never steal it)

When you’re heads-down executing, it’s easy to forget to take a moment every now and then and celebrate wins.

But doing this will do wonders to team morale, and people will remember if you give them a well-deserved shout-out (especially if they are on teams that typically get less visibility, like Data Engineering).

And while it’s important to create visibility for your own work as well, don’t do this at the expense of others. For example, even if you did the majority of the work on something, as long as others were involved, share the credit.

It’s not a zero sum game.

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

Vannevar Labs: Product Operations Lead | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series B | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: General Catalyst, DFJ Growth

Ghost: Business Operations Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: NY / LA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $85k - $100k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 3+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series C | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Union Square Ventures

dbt Labs: VP, Finance & Strategy | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 15+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Sequoia, Google Ventures

Figma: Figma Ventures , Corporate Development & Strategy | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NY or remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $136k - $288k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 6+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Sequoia, Greylock

OpenAI: Special Projects, GTM | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $265k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 6+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Thrive Capital, Khosla Ventures, Coatue, SoftBank

Anduril: Business Operations Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Costa Mesa, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $94k - $141k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 1+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Founders Fund, General Catalyst

Plaid: Business Operations | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $135k - $202k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Index Ventures, Ribbit Capital

Airbnb: Manager, Strategy & Operations EMEA | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: London / Paris | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 10+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Public

Instacart: Chief of Staff to CTO | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Remote | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $194k - $260k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 8+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Public company

📚 What I enjoyed reading recently

• 7 Real-World Lessons from Data Leaders at LinkedIn, Dropbox & More by : A collection of 7 webinars with hands-on advice for leaders (and everyone else) in data

• MCP (Model Context Protocol): Simply explained in 5 minutes by : The hottest topic in AI explained in practical terms

• Offensive vs. defensive data analysis by : A reminder to be more proactive in data analytics incl. a list of helpful real-world examples from top companies

• The 13 software engineering laws by : 13 laws, and why they matter, summarized in one place; a great resource to bookmark for later

• Take BIG Bold Career Risks by and : An insightful discussion with a seasoned operator (early at FB, COO at Quip)

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

Solving problems sounds like it’s mostly about analysis and execution. And while those are key, it’s equally important to get everyone aligned on the path forward.

In a small startup, you might be able to just “go and do things”.

But once your organization hits a few hundred employees, has multiple layers of management and several teams that want to have a say, you need to approach important decisions in a structured way.

That doesn’t necessarily mean you need to introduce a formal decision-making framework like RAPID (we’ll get to these in the end); but it does mean you need to:

1. 🔍 Identify the decision maker

2. 📝 Prepare by aligning with other stakeholders

3. 👩🏻‍🏫 Frame the problem & present your recommendation

4. ⏳ Make sure the decision gets made efficiently and in time

5. 🔒 Make sure the decision doesn’t constantly get reopened

Many projects fail, or take unnecessarily long, because people are unable to tee up a clean executive decision. In this post, we’ll cover how to do it well.

I will first go through what works from my experience, and then cover popular decision-making frameworks that are widely used at top companies.

Note: Similarly to what I wrote in my guide on solving problems, you should only apply a structured approach like the one described in this post for high-stakes decisions. Everything else can be handled using your best judgment.

When in doubt, remember: It’s usually easier to ask for forgiveness than to get permission.

As a reminder, this is Part II in a loose series on how to solve problems. If you haven’t yet, check out Part I here:

Step 1: Identifying the decision maker

You can’t make a decision without the decision maker.

What might sound obvious is actually the cause of a lot of failed meetings I’ve been a part of. In my last job, we’d sometimes have heated debates for 45 minutes between Marketing and Sales only to realize that the person who can truly make this decision wasn’t even in the room.

So, how do you figure out who the decision maker is?

In most organizations, it’s the person that is on the hook for the outcome of that decision (if that’s not the case in your company, then you might want to address that). In rare circumstances, it needs to be multiple people (e.g. when transferring ownership of something from one team to another).

Unfortunately, while it’s often immediately clear which team owns the outcome and thus the decision, it’s not always obvious at which level the decision is made.

For example, if you want to get sign-off on testing a new AI agent for customer support, do you approach the support lead on whose team you want to test it? Or does the Head of Customer Support need to sign off?

Most companies don’t have these kind of decision-making rules clearly spelled out.

What worked well for me: Approach the person who directly manages the impacted team and copy leaders 1 - 2 levels above them on the message. That way, there’s a very low risk that the topic will be reopened later because an exec felt like they didn’t have visibility into the decision and don’t want to own it.

Step 2: Preparing for the decision meeting

For important decisions, I recommend scheduling the decision meeting first and then working backwards to prepare. This ensures that you get the decision made by when you need it.

Once you have that scheduled, you need to prepare. The easiest way to create chaos is to use a decision meeting with an executive to discuss the topic with other teams for the first time, so you should avoid that at all costs.

Instead of discussing with you how to move forward, they will:

• Ask basic questions about the situation

• Argue about how important and urgent the problem is

And so on.

If you want to have a clean, efficient decision meeting, the only person getting any new information in the meeting should be the executive making the decision

Ideally, you’ve aligned with the other meeting participants on every aspect in advance. But that’s not always possible.

Here’s the hierarchy of what you absolutely need to agree on in advance vs. what can remain up for debate:

• [Must have] You need to agree on what the problem is and that it’s important

• [Very important] You should agree on how to think about the problem (more on that below)

• [Important] Ideally, you should come with a joint recommendation

So, how do you actually do that?

I recommend you meet in person with the team(s) most directly affected (e.g. if it’s a Go-To-Market decision, Marketing & Sales should be in the room).

For everyone else, you can share a brief write-up of what you’re proposing and give them a chance to weigh in with feedback by a certain deadline. That deadline should be at least a few days before the decision meeting so that you have a chance to follow up on any controversial points and resolve them in time.

If you skip this step, you risk a flood of complaints once you announce the decision, which will make it difficult to move forward. So, don’t.

Step 3: Framing the decision

The executive making the decision is not as in the weeds as you are; in fact, they might not have thought about the topic in question before.

So it’s your job to clearly frame the decision you want them to make. This is not about giving a recap of how the problem came up, or your analysis. It’s about describing in simple terms what has to be decided, and how to think about the decision.

Example: Short-form video at Meta

While I worked at Meta, we were in the process of slowly rolling out Reels (Meta’s short-form video product) across Facebook and Instagram.

Initial tests showed revenue drops across the board. If we had evaluated the decision whether to roll out Reels broadly to all users based on normal experiment guidelines, the answer would have been “No”.

However, Reels was a new, unoptimized product and not meaningfully monetized yet. In addition, if Meta didn’t cannibalize its own existing business with this new product, TikTok would. Therefore, the key question for the roll-out wasn’t whether Reels was a net-positive addition at the moment, but rather whether we believed it could, in its end state, be incremental to engagement and revenue.

Leadership decided to make this long-term strategic bet and it paid off: In its Q4 2023 earnings call, Meta announced that Reels had become a net positive contributor to company revenue, and it’s now one of the main drivers behind the continued growth.

While it’s important to give context, you need to be concise. In most companies that aren’t Amazon, executives won’t read a multi-page document ahead of the decision meeting. So you need to summarize the key information in an easy-to-digest way.

Give them the information they need to make the decision. Not more, not less.

As a general principle, the framing of the problem should be presented by one person, even if multiple teams are involved. That’s why it’s crucial to agree on a joint framing in advance (see Step 2 above).

Sequencing decisions

Some complex problems require multiple decisions as you’re working through them. The initial decisions guide the rest of the work, so you should get those locked in as early as possible.

Based on the guidance you get, you can then continue the work and come back to frame up another follow-up decision.

For example:

Let’s say you work at a video platform like YouTube and are trying to figure out whether to increase the number of ads.

If you have made similar decisions in the past, you might have a principle in place that tells you how to make this decision. In that case, you just have to present your recommendation and the data that backs it up to get sign-off.

But if this is the first time, you need to start with the fundamentals.

Increasing ads on a media platform is ultimately a trade-off between revenue and user growth. Users don’t like ads, so more ads will results in less engagement or even churn. However, the (remaining) time users spend on the site will generate more revenue.

The first decision you have to tee up is:

“Are we optimizing for revenue or user growth?”

Without clarity on the strategic goal, you can’t decide if (and how much) you should ramp up ads.

➡️ If you’re optimizing for user engagement, you shouldn’t run more ads; the work stream comes to an end

➡️ If you’re optimizing for revenue, increasing ad load is an attractive option. The follow-up decision you then need to tackle is: By how much should you increase ads?

Step 4: Presenting your recommendation and the options

Lead with the recommendation

Once you’ve framed the problem, you should share your recommendation.

Why give a recommendation at all?

It’s your job to not just dump information on an exec, but provide a recommendation (for them to agree with or challenge). That’s because:

1. You’re closest to the problem, so you are uniquely positioned to give a recommendation, and

2. Otherwise, you’re leaving all the heavy lifting to them

But why lead with the recommendation?

This is the core idea of the “Pyramid Principle” (if you’re a regular reader of this newsletter, you might have heard about it before 😬).

If you don’t lead with the recommendation, people will be distracted and wonder what it is while you’re presenting the analysis.

In addition, it allows them to interpret the data and arguments you present in the context of the recommendation, making it easier to follow along.

As mentioned before, if there are multiple teams present, you should ideally present a joint recommendation. If you really can’t agree on one, you can present two “competing” recommendations as a last resort. In that case, start with the framing of the problem (which all teams should agree on), and then let each team present their recommendation for what should be done.

Let’s continue with the example from above. For simplicity’s sake, let’s assume your company has determined in the past that it wants to optimize for revenue, and the decision at hand is how much (if at all) you should increase ads.

The recommendation you might present is:

“We should increase ad load (ads / hour) from 5 to 6 by increasing overlay ads”

Following the pyramid principle, you now need to back up that recommendation.

Provide supporting arguments and show alternative options

To be specific, we actually made two separate recommendations, and we need to provide reasoning for both:

1. Recommendation #1: “We should increase ad load from 5 to 6”

2. Recommendation #2: “The increase should come from overlay ads”

Diagrams and graphs come in handy here. A single crisp chart can communicate what might take you multiple paragraphs to explain.

Let’s start with recommendation #1. To explain our reasoning, we could show a chart that shows revenue as a function of ad load (based on experiment data):

This chart clearly explains how we landed on our recommendation and shows the other options (in this case, the spectrum of ad load choices). Our objective is to maximize revenue, and that happens at an ad load level of 6.

Below that, we’re leaving money on the table because our revenue per hour of watch time is not high enough. Above 6, decreases in watch time (users watch less due to ads or churn altogether) outweigh the incremental revenue per hour.

This type of diagram is well-suited for when you’re trying to solve a quantitative problem and decide which point on a spectrum you should pick.

But not all problems are like that. The second part of the question (“Which ad type should we scale?”) requires you to compare several discrete options. In this case, the Traffic Light framework, popularized by Meta, is a good fit.

It’s simple; you just need to:

• 1️⃣ Decide what criteria you care about (that's your y axis)

• 2️⃣ List out all options (that's your x axis)

• 3️⃣ Score each option for each criteria using red / yellow / green color-coding

The result is a simple matrix that lets you quickly see the pros and cons of each option.

The important part: While you should evaluate all options as you’re working through the problem, you shouldn’t present all of them to the decision maker. Your job is to focus the discussion and narrow things down to the most relevant options.

For completeness’ sake, you can add the remaining options to the appendix and link to that.

Lastly, keep in mind that even with color coding, interpreting a list of pros and cons to make a decision still takes some mental effort. As a result, I’ve found it helpful to present your recommendation and the main competing options in plain English:

Other frameworks for presenting recommendations & options

I can’t write an article on framing options without mentioning one of the most useful frameworks ever: The 2x2 matrix.

It might seem simplistic, but that’s the point. There’s a reason why it’s one of the favorite tools of consultants at McKinsey, BCG etc.

It forces you to boil a complex topic down to just two relevant dimensions. The result places all options on a grid with four areas.

Here’s an example. Let’s say your team is responsible for driving user engagement on a social media platform, and you’re on track to miss your goals for the quarter.

Instead of presenting your manager with a long list of projects, ranked by a dozen different criteria, you can create a simple 2x2 matrix:

This presentation makes it easy:

1. You start with the ideas in the top left (high impact, short time to impact)

2. After you exhausted these, you move to the bottom left (lower impact, but can still help you hit your quarterly targets)

3. The ideas in the top right can be revisited in the next planning cycle (high impact, but take too long to help your current situation)

4. The ones in the bottom right you shouldn’t touch

Step 5: Moving forward after the decision

You got the decision maker to make a call — great!

But there’s still work left to do. You’ll want to make sure that:

• 🤝 People are bought into the decision (or will at least not actively fight it)

• 🔒 The decision doesn’t get reopened over and over again, and

• ⚙️ People actually execute the plan

Announcing the decision and explaining the rationale

People who were in the decision meeting know what was decided and why.

Everyone else doesn’t have that context. In fact, they might not even know that the decision was made at all!

So, it’s up to you to make sure everyone’s in the loop. Depending on your company’s culture and preferred communication tools, you can send an email, a Slack announcement etc.

The important thing is that 1) it’s in writing and 2) you share the rationale for the decision.

If you don’t share the context, you’ll likely get a lot of pings from people asking whether you considered this or that factor or looked at certain alternatives.

Documenting the decision

Nothing erodes morale faster than constantly revisiting decisions that were already made. When that happens, it doesn’t just feel like you’re not moving forward — it feels like you’re taking a step backwards.

So make sure you’re creating a permanent record of what was decided and why. When someone wants to reopen the topic (for example, new people joining the company tend to do this), you can point them to it.

One thing I find helpful in this regard: Keeping a decision tracker. That way, you can easily dig up the relevant documents.

Enforcing “disagree and commit”

It’s important that everyone gets heard. That doesn’t mean everyone gets what they want all the time.

As long as you give people a chance to give input by a reasonable deadline, you’ve done your job. Even if someone disagrees with the decision: In a functioning organization, they have to accept that and move forward (i.e. support the execution of the plan).

It’s important to be adamant about this, or you’ll set a precedent that anything can constantly be revisited. Even if someone brings up a valid new point that you didn’t consider at the time, it’s usually not worth reopening the discussion. As long as the original decision and rationale are still reasonable, it’s better to keep moving forward.

“A good plan violently executed now is better than a perfect plan next week.” — General George S. Patton

Assigning a party responsible for execution

If you want to get something done, it’s best in my experience to have a single person in charge of coordinating the execution. This can be a PM, BizOps manager, Program Manager etc.

That doesn’t mean they have to do all the work — it just means they are making sure that the right people are working on the right things and any deadlines etc. are hit.

A checklist for efficient decisions

Whew, that was a lot. In case you want a TL;DR, here’s a checklist to help you efficiently facilitate decisions:

• 👨‍💼 Decision maker: A single person for most decisions; but give visibility to execs 1 - 2 levels up the chain

• 1️⃣-2️⃣ Sequence decisions: Tee up fundamental decisions (e.g. “What are we optimizing for?”) early on to give clarity & minimize throwaway work

• 🤝 Avoid chaos: Align with other teams in advance to present a joint framing of the problem. Ideally, you present a joint recommendation as well

• 🎯 Lead with your recommendation before diving into supporting data & arguments or alternatives

• 🖼️ Presenting choices:

  • Use graphs to illustrate optimal choices for quantitative problems

  • Use a traffic light matrix for comparing a small set of options across multiple criteria

  • Use a 2x2 matrix to boil an issue down to the essentials

• 📅 Set a deadline: Set a date by when the decision needs to be made, schedule the meeting, and then work backwards from that to prepare

• 💬 Communicate the decision and rationale: People need to know about the decision in order to follow it, and need context so they can get behind it

• 📝 Document the decision: Create a written record of what was decided & why so it doesn’t constantly get reopened. If there are a lot of important decisions, create a tracker for easier reference

Popular decision-making frameworks used by top companies

The guide above is a collection of tactics that have worked for me working in tech companies of different stages.

While it’s not based on a particular formalized decision making framework, there is material overlap with some of them. Below, I’m giving an overview of some of the most commonly used ones.

Note: Using formal frameworks like this is typically only a good idea once your org reaches a certain size and level of complexity, and you’ve exhausted other methods of creating clarity (e.g. simplifying the org chart).

For example, Uber only introduced RAPID shortly before the IPO when it became clear that the more informal approach it had applied so far wasn’t working anymore at scale.

Frameworks that help you work through a problem & make a decision

S.P.A.D.E.

S.P.A.D.E. stands for Setting, People, Alternatives, Decide, and Explain. Gokul Rajaram developed and popularized this framework while at Google, Facebook and later Square.

• 🎯 The very first step (not even part of the acronym) is to classify the decision by urgency and importance. Only use the framework for important decisions

• 🖼️ Next, you establish the setting (what, by when, and why)

• 👨‍👩‍👦‍👦 Once you have that laid out, you need to include the right people

  • 1) decision-maker (this person is accountable for the outcome and responsible for execution)

  • 2) approver, and

  • 3) consultants (give input); you should consult as many people as possible

• 💡 Then, you lay out alternatives. These should be feasible, diverse and comprehensive (AKA “MECE”). This should be done via a public brainstorm

• ⚖️ Finally, you decide on the best option (you can hold a private vote beforehand if you want)

• 💬 Once you’ve made your decision, it’s time to explain it. First, you hold a “commitment meeting” with the approver and consultants to get their buy-in. Then, you broadcast the decision to the wider company.

Further reading:

• Gokul Rajaram’s S.P.A.D.E. toolkit

• S.P.A.D.E. template

• S.P.A.D.E. log

• S.P.A.D.E. example #1: ATS selection

• S.P.A.D.E. example #2: Feature naming

Coinbase framework

Brian Armstrong, CEO of Coinbase, walked through how Coinbase makes decisions in this Medium post.

The three key steps are:

1. Set the parameters

2. Deliberate

3. Decide

Step 1: Set the parameters

Define what is being decided, by when, by whom etc.

One unique thing about this framework is that it includes a date by which the decision will be revisited (at the earliest). A neat idea you might want to use even if you don’t apply the whole framework.

Brian Armstrong recommends to get input from 3 - 8 people (in contrast to S.P.A.D.E. that encourages to consult basically everyone).

Ideally, those input providers should represent all teams affected by the decision.

Step 2: Deliberate

You get everyone in a room to share information, brainstorm options and vote. The framework suggests doing one round of voting after collecting all options and data, and a second round after discussing pros and cons to see if anyone could be convinced.

Step 3: Decide

The last step is that the decision maker decides, communicates the decision, and documents it.

Frameworks that mainly clarify roles & responsibilities

RAPID

Many companies, including Uber while I worked there, adopted this framework.

The acronym stands for: Recommend, Agree, Perform, Input and Decide. Its main benefit is to clarify roles during the decision-making process to avoid going in circles.

In chronological order, the decision-making process with RAPID works like this:

• Input: Multiple parties provide input (subject matter experts, the party responsible for execution etc.)

• Recommend: The Recommender drives the process and makes a recommendation (that's where most of the work happens)

• [Optional] Agree: A designated person makes sure the recommendation meets certain requirements (e.g. Legal)

• Decide: One person is responsible for making the decision

• Perform: One person / team implements the decision

DACI

Similar to RAPID, the main purpose of DACI is to clarify the roles in the decision-making process:

• Driver: The person who is in charge of making the decision happen by the date it needs to happen. I.e. they manage stakeholders, organize meetings etc. to make sure the process stays on track

• Approver: This is the person who ultimately makes the decision

• Contributors: These are the subject matter experts that provide input

• Informed: The people who need to be in the loop because they’re affected by the decision (but don’t have a say)

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

HeyGen: Special Projects | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / Palo Alto / LA / Toronto | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $150k - $160k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 2+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series A | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Thrive Capital, Benchmark, BOND

Blackbird Labs: Market GM - San Francisco, CA | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $200k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series A | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Union Square Ventures

Inworld AI: BizOps Lead - USA | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Mountain View, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $170k - $210k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series B | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Lightspeed, Kleiner Perkins, Founders Fund

Anduril: Chief of Staff, Software Platform | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Costa Mesa, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $190k - $285k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 8+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: a16z, Founders Fund, Thrive Capital, General Catalyst

OpenAI: Strategic Finance Generalist | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $210k - $265k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 7+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Khosla Ventures, Thrive Capital, Microsoft

Anthropic: Senior Revenue Operations Analyst | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $150k - $190k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Menlo Ventures, Amazon, Google

Harvey AI: Business Operations Lead | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: SF / NYC | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $200k - $260k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 6+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Kleiner Perkins, Google Ventures

ScaleAI: Business Operations & Strategy Manager | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $160k - $200k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 4+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Index Ventures, Founders Fund, Coatue

📢 What I’ve been up to

I recently wrote a guest post for on “How to become the data expert everyone wants to work with”.

Over the last decade, I’ve been on both sides of the collaboration between data & biz teams. During this time, I have noticed a few key patterns that distinguish those working in data that drive outsized business impact from those that don’t.

This article covers 5 skills that I think will help you stand out. Check it out and give the newsletter a read in general; puts out some of the best data-related content you can find.

📚 What I enjoyed reading this week

• 35 by : 35 pieces of advice for his younger self (that anyone can benefit from)

• Engineer to CEO in 3 years: These key lessons got me there by

  & : Three key lessons for accelerated career growth

• Memory in AI Agents by : An insightful discussion of how memory can help improve the performance & user experience of AI agents, and what it takes to make this happen on the technology side

• How to feel bad and be wrong by : A good read on how our mind tends to substitute hard questions with easy ones (and why we should try to push back on this every now and then)

• Same Data, Different Questions by : Great illustration of how presenting the same data in different ways can spark and answer different questions for the business

• Why The Algorithm Hates You by : A fun illustration of sampling bias in action

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

When you read job postings, one of the most common requirements is “strong problem-solving skills”.

That seems reasonable at first glance; companies hire people because they have problems they need solved.

But what does it really mean to be good at solving problems?

In my experience, it boils down to this:

Most people can only solve problems they have encountered before. For example, you hire someone who built Marketing attribution at another company to do the same at yours (and they’ll pretty much copy/paste what they did the last time).

Very few people, however, are good at reliably solving problems they’re encountering for the first time.

After all, you can’t just do what you did last time, and there might be no guide for this exact problem.

But it’s an absolutely crucial skill, and not just for “generalists”. In high-growth companies, what got you to the current stage is not going to get you to the next level, so even if you specialize in an area like Growth Marketing, you need to deal with new challenges all the time.

The goal of this article is to help with that and give you a general purpose playbook for solving any problem that comes your way.

It will be split into at least two parts; in this post, we’re going to cover how to analyze the problem and find potential solutions.

In the second part, we will discuss how to present these options and decide on the path forward. Check it out here:

Let’s get into it.

Why problem solving fails so often

The number one mistake people make when it comes to problem solving is to start discussing solutions before they’ve fully understood the problem.

It’s enticing; we often feel like we know already how to solve the issue and want to move forward. I’ve seen this especially in high-growth companies where “moving fast” is one of the most important principles.

But if you don’t take the time to deeply understand and define the problem first, there is a high risk you choose the wrong solution and have to re-do things later.

In other words, moving too fast can actually end up taking more time than a structured, deliberate problem-solving approach.

For example, in my previous job at a B2B SaaS startup, we wanted to put an account score in place for the Sales team. Most people immediately started talking about vendors we could work with, or brainstorming what type of ML model would be best suited for this problem if we built it in-house.

This is an example of jumping into the solution space before fully exploring the problem space. Instead of discussing what we were going to use the score for and, as a result, what it needs to look like, they were trying to go straight into execution mode.

But how can you decide if any of the options in front of you are a good fit if you don’t know exactly what your requirements are?

Assessing the nature of the problem

If you want to solve a problem, the first step is to understand that type of problem you’re dealing with.

Is it a one-way door or two-way door decision?

When you’re trying to decide whether to optimize for speed or finding the optimal solution, the key deciding factor is the cost of being wrong.

Besides the business impact of the decision, what determines the cost of being wrong is how hard (or costly) it is to reverse.

If it’s easy to reverse, you can just go ahead and try a solution. If it doesn’t work or turns out to be not an ideal fit, you just roll it back and try something else—no harm, no foul.

For example, instead of analyzing in-depth what format your team meeting should have, just try something and iterate based on feedback.

For decisions that are costly—or impossible—to reverse and have substantial business impact, however, you should put in maximum effort to find the ideal solution. The cost of being wrong is high, so you need to minimize the likelihood that you’re wrong by using a structured problem-solving approach like I’m describing in this post.

Essentially, you’re trying to minimize the expected cost of your decision:

Jeff Bezos calls these one-way door and two-way door decisions. Here’s a helpful overview of how he thinks about them at Amazon:

Can you break the problem into smaller steps?

Many problems seem impossible to tackle. There are simply too many unknowns and dependencies to make a decision that we feel comfortable with.

Luckily, we can often make things simpler by splitting one big decision into a sequence of smaller ones. Ask yourself:

Is there an initial small test you can do to gather data and decide on next steps?

For example, deciding whether you should enter the UK market and open a local office is a big decision that’s hard to unwind (you’d have to break your lease, let people go etc.). However, people in the UK speak English, so you could start reaching out to potential customers with your existing Sales reps in the US (they just have to work slightly odd hours).

If the response is good, you can start hiring your first dedicated Sales people in the UK. Then, if deal close rates and contract volumes justify further investment, you can hire management and supporting functions and build out proper local operations.

All of a sudden, instead of deciding whether to make a multi-million $ investment in a new market based on research and hypotheses, you just have to decide if it’s worth asking one or two existing employees to shift their focus for a few weeks.

What are the opportunity costs of the decision?

Decisions don’t happen in a vacuum.

Often, deciding to do one thing means not being able to do something else. As a simple example, hiring one candidate for a position means saying “No” to all other candidates.

Similarly, our future options for solving a problem might be limited based on a choice we make today.

For example, when I was at UberEats, DoorDash made some controversial pricing and UI changes in their app. We had to decide if we wanted to publicly slam them for their choices or not.

This decision had serious consequences for our own product strategy in the future. If we took a public stance that these choices were bad for couriers or end users, we would essentially commit to never implementing any of them ourselves (even if that might put us at an economical disadvantage).

Takeaway: The more downstream implications the decision has, the more effort you should put into analyzing the problem and deciding on the path forward.

Is it a recurring or one-off problem?

Recurring problems require a different approach compared to one-off decisions.

There are two types of recurring problems:

1. You have to deal with the exact same thing on a regular basis

For example, every week, you have to go through hundreds of customer support tickets to identify common themes.

In this case, you’ll want to automate the solution (e.g. asking an AI agent to parse the tickets and highlight important topics) so you don’t have to keep solving the same problem over and over again.

2. You deal with similar problems on a regular basis

For example, you might be wondering on a regular basis whether you need to get legal sign-off for a new project or feature you’re working on.

In these cases, you’ll want to develop a principle that gives you clear guidance on how to handle these things going forward. So by solving the problem comprehensively once, you solve all future instances as well.

Symptoms vs. root causes

One final thing to consider: If you’re dealing with similar problems over and over again, you should ask yourself whether you’re actually solving the underlying problem, or just fighting the symptoms.

For example, when I was at Uber, we had constant debates between the Central Operations team (which I was on) and the local City teams over pricing and incentive decisions.

In the beginning, I thought I needed to define a principle that told us when we should or shouldn’t launch additional incentives. However, after a while, I realized that this didn’t actually solve the issue and I was just playing corporate whack-a-mole.

The reason we kept having these conversations was because we hadn’t clearly defined the roles and responsibilities of the teams.

So what we really needed to do wasn’t to decide under which conditions we should run promotions in a given market, but define clear swim lanes with leadership so that one team could unilaterally make the call.

Defining the actual problem you’re trying to solve

Most of the time, people don’t throw problems your way, but instead give you tasks. In other words, they only share with you what they think the solution should be, and they don’t share the context of the underlying problem.

The issue with this:

Without deeply understanding the problem, you can’t do a good job on the solution.

For example: Somebody might ask you to pull a list of the number of businesses by country. But that’s not the actual problem they’re trying to solve.

• Once you ask why they want that information, you learn that they are trying to rank countries by market size

• If you ask “why?” again, you learn that they are trying to decide which market is most attractive to enter next

Only once you have that context can you start being a thought partner in solving the problem. For example, you might realize that many registered businesses are 1-person sole proprietorships—that your company wouldn’t sell to—and suggest additional filters to apply to the analysis.

Or, you might suggest a completely different framework for ranking markets altogether.

Depending on your seniority, you could even peel back another layer and argue that the true underlying problem is that the company is trying to figure out how to unlock additional growth, and entering a new market is just one of the options on the table.

But: While digging into the “why” is helpful, you should be careful not to overdo it and waste time by trying to reopen issues that have already been decided.

For example, if leadership has already decided that you’ll launch a new market, and the only question is which one, there’s no point in debating that point.

Identifying and structuring your options

In order to decide on the best path forward, you should first lay out all possible options. In a second step (which we’ll cover in the next section) you’ll then narrow things down.

The most important thing is to make sure the options you’re identifying are MECE (Mutually Exclusive and Collectively Exhaustive). This means you try to find all possible options and define them so that none of them overlap.

For example, let’s say you’re trying to decide whether to buy an AI agent from a certain vendor to automate a task, and these are the options you come up with:

These three vendors are not a MECE set of options. Instead of buying from one of them, you could:

1. Buy from another vendor

2. Build the agent in-house, or

3. Simply do nothing (and continue doing the work manually)

In fact, “do nothing” is an option that’s often forgotten, but it’s an important one to consider. Even if the status quo is suboptimal, it might still be better than any of the alternatives on the table.

And solving a problem always comes with opportunity costs: If you spend time and money on solving it, you’ll have to deprioritize other things you could do instead.

So how do you come up with the possible options?

To get as many alternatives on the table as possible, a brainstorming session with the team can be helpful.

In addition to identifying options you might not have thought of, you’ll make sure that everyone feels heard. This will eventually make it easier to get buy-in.

However, you’ll likely end up with a lot of half-baked and overlapping ideas, so you can’t just put them all on a page and call it a day. Instead, a single person should be in charge of cleaning up and summarizing everything in a MECE way.

For more complex problems, a great way to do this is using an issue tree. Let’s say you’re trying to figure out how to grow marketing leads for your business. This is what the results of a brainstorm and the resulting issue tree could look like:

Evaluating the options

Step 1: Developing an evaluation framework

With your options laid out neatly on paper, you need to figure out how you’re going to compare them against each other so you can eventually frame up a decision.

In the example above, you could simply say you’re going to look at:

1. How many leads you can get from each possible lever, and

2. What the expected financial efficiency of each is.

If you’re concerned about your ability to implement them, you could also consider the operational lift required. Or, if you need the leads fast, you could rate each option with regards to how long it will take to yield results.

There is no right answer here; it ultimately depends on what you care about (or rather, what you think the person who will ultimately make and own the decision will care about).

Step 2: Analyzing the options

Once you have your evaluation framework, you need to analyze each option across the criteria you came up with.

In many cases that means doing data analysis or modeling (e.g. estimating the number of leads and efficiency in the example above); in other cases it means doing research or talking to subject matter experts (e.g. to estimate the implementation effort).

The most important thing is to spend your time effectively. There are two considerations that can help you here:

1. 🎯 Prioritization

You should spend most of your time on the factors that will affect the decision the most.

For example, if you know that the business is struggling to hit growth targets, then the number of leads you can get from a growth lever will be the most important estimate to get right.

2. 🔢 Sequencing

If there are requirements that are non-negotiable, you should start with those.

For example, in the scenario above, you might know that the Marketing team needs to show results next quarter. As a result, any initiative that takes longer than that to show an impact can be immediately disqualified.

Similarly, if you’re selecting between different vendors, first check if any of them are missing must-have features. If so, there’s no need to evaluate them for the rest of the criteria.

Your goal should be to start with the analysis that can eliminate as many options as possible early on, so you can minimize the amount of work you have to do.

Outlook

At this point, we understand the problem, have structured it, identified our options and analyzed them. That means we’ve done most of the work.

However, without a decision on the path forward, you can’t go and implement your solution. And often, that decision isn’t up to you; instead, you need to present the options and your recommendation to an executive and get their sign-off.

In the next post of this series, we’ll cover exactly that.

• How do you present the options and your recommendation to the decision maker?

• How do you make the decision-making process efficient?

• How can you avoid constantly reopening decisions that have already been made?

We’ll also review some of the most popular decision-making frameworks that leading companies are using.

Stay tuned! I hope to see you then. If you want to make sure you don’t miss it, subscribe below.

💼 Featured jobs

Ready for your next adventure? Here are some of my favorite open roles, brought to you by BizOps.careers (sorted from early to late stage):

Cartesia: Chief of Staff to the CEO | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 3+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Seed | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Lightspeed, General Catalyst

Traba: Strategy & Operations General Manager (Special Projects) | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: New York | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $140k - $180k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 7+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series A | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Khosla Ventures, Founders Fund

OpenAI: Special Projects Lead | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $265k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 5+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Khosla Ventures, Thrive Capital, Microsoft

Faire: Strategy & Analytics, Product Senior Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: NY / SF | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $125k - $172k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 3+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Sequoia, Lightspeed, Founders Fund

Vercel: Chief of Staff, Product & Engineering | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Francisco | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $164k - $246k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: Flexible | 🚀 𝗦𝘁𝗮𝗴𝗲: Series D+ | 🏛️ 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿𝘀: Google Ventures, Accel

Zoox: Strategy Associate | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: Foster City, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $124k - $178k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 3+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Public (acquired by Amazon)

Google: Chief of Staff, YouTube Business | 🏠︎ 𝗟𝗼𝗰𝗮𝘁𝗶𝗼𝗻: San Bruno, CA | 💰 𝗦𝗮𝗹𝗮𝗿𝘆: $174k - $261k | 💼 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲: 8+ YOE | 🚀 𝗦𝘁𝗮𝗴𝗲: Public

📚 What I enjoyed reading this week

• DeepSeek: Frequently Asked Questions by : The most common questions about DeepSeek answered in one spot

• Breaking Into Data: How to Land A Job in a Competitive Market by : A tactical guide on landing data jobs, just in time for the new year

• The Self-Promotion Horror Show by : Tactical tips for creating visibility for your work (a key requirement for getting ahead)

• Stop apologizing for reasonable business decisions by : A good reminder that we tend to apologize too much at work, with serious consequences

• 7 proven mental models for engineering managers by / : A helpful summary of 7 mental models that anyone can use at work, not just engineers

In her newsletter, Tessa shares actionable advice on topics ranging from how to get a job in data to solving business problems with data and communicating more effectively — based on her experience as an Analytics Consultant at McKinsey and Data Science Manager in Tech.

If you haven’t, I highly recommend you check it out here.

As long as humans have existed, they have told stories. For example, the cave paintings in Lascaux, France, are over 30,000 years old and show traces of early graphical storytelling.

And stories still play a central role in our lives today. Ask yourself:

• How much do you remember from the last slide deck you saw at work? Probably not much

• But how much do you remember from The Lord of the Rings or The Office [or whatever book / movie you like]? Chances are, you can quote entire scenes

If you want to get your point across and make a lasting impression, you need to wrap it in a narrative. Sometimes, people who work with data think they’re an exception to this. Data represents cold, hard facts – that should be convincing enough, no?

But this couldn’t be further from the truth. Data by itself doesn’t drive any impact in a company; you can only unlock business value when you get people to care—and ultimately take action—by showing what the data means for them.

In this post, we’ll cover:

• What data storytelling is and isn’t, and why it matters

• How to package data into a compelling narrative

• How to visualize data for clear takeaways

• Best practices and what to avoid when presenting data

Let’s get into it.

What is data storytelling, and why does it matter?

Let’s get the most important thing out of the way:

What data storytelling is: Organizing data into a compelling narrative, putting it into context and deriving a “so what?”

What data storytelling isn’t: Cherry picking or twisting data to support a preconceived narrative

In some organizations, data storytelling has a bad rap because managers use it to put a spin on metrics and hide areas of subpar performance in their department. As a result, some executives only want to look at “unadulterated” raw data.

But that’s an overreaction. Instead of outright banning it, we should make sure to follow best practices to maximize the business impact without misleading our audience.

We will cover those best practices here.

Narratives drive decisions and outcomes

Ultimately, any analysis we do at work is supposed to drive a decision.

• What do we build?

• Which customer segment should we focus on?

• Should we expand internationally, and if so, into which markets?

Any good analysis should start with a business problem that needs to be solved, and should end with a clear narrative that argues what should be done going forward.

If you don’t create a narrative and only present raw data (e.g. “random” sets of charts in a slide deck), you are leaving the work of making sense of the data to your audience.

Best case, your audience arrives at the right conclusions with tons of effort and resents you for the headache.

Worst case, the point is going right over their heads.

Telling a story allows you to take control of the narrative, avoid misinterpretations and get everyone aligned on a path forward. This is needed in a number of common scenarios:

1. Leadership updates

Leaders have limited time and lack the detailed context that the people working on the problem have. So it’s up to you to help them understand what’s going on.

2. Investigations

When you’re doing an investigation (e.g. into a metric drop), you need to get a joint understanding of what has happened so you can move forward.

If you don’t, different teams will be sharing different versions of events and things will get confusing quickly.

3. Influencing stakeholders

Data is great when you’re trying to get others to do things. Instead of just saying “I think you should do this”, you can make the whole thing more objective and say “Based on the data, we should do this because of X and Y”

If someone wants to push back, they now need to disprove your analysis or counter with another data-driven argument.

How to package data into a compelling narrative

Creating a narrative involves three steps:

1. 🔍 Identify your audience and what matters to them

2. 🧵 Tailor your narrative to your audience

3. 💬 Deliver your narrative for maximum impact

Let’s take a closer look at each one.

Step 1: Identify your audience and what matters to them

You need to make people care about your message if you want to get through to them.

And if you want people to care, you need to understand what matters to them:

• What are their big strategic priorities for the year or the quarter?

• What are their goals, and what metrics are they measured on?

• What’s going well, and where are they struggling?

For example, you might find that the Product Director you’re working with is mostly focused on growing engagement for a new experimental bet they launched and that they are behind plan.

If you can establish a connection between the data you’re presenting and engagement growth on their product, you’ll have their undivided attention.

Step 2: Tailor your narrative to your audience

Once you know what your audience cares about, you have to create a narrative that’s specifically tailored to that. This can mean highlighting different data points, or framing the same data points differently.

Ask yourself:

Based on the data you analyzed, what should they be doing differently?

This is helpful for two reasons:

1) Presenting something surprising captures people’s attention; if people walk away thinking “Well, that was obvious”, they likely won’t remember it and it means your analysis just scratched the surface

2) You’re adding immediate value because you’re giving them something they can apply right now

Once you have people’s attention, you need to keep it. That means you should focus only on the parts that are truly relevant to your audience.

Deciding what not to include is just as important as deciding what to include.

Step 3: Deliver your narrative for maximum impact

Even if your narrative is jam-packed with data-driven insights; if you don’t deliver it right, it won’t lead to any real business impact.

You’ve heard us talk about this before (e.g. here and here), but no article on data storytelling can be complete without mentioning the Pyramid Principle. If you already know it, feel free to skip ahead to the next section.

Most data people tend to talk about their analysis chronologically and focus too much on technical details:

By the time you get to the insights, everyone has tuned out. Instead, start with the relevant takeaways for your audience and then go deeper by offering supporting arguments and data:

How to present data for clear takeaways

If you present data the right way, it tells a story all by itself without needing much commentary.

This is your most powerful data storytelling tool: A chart with a single clear takeaway will give your message a massive boost in credibility, while an ambiguous chart will result in debate that can derail your narrative.

Here are some of the most common scenarios you’ll come across with real life examples on how to effectively visualize the data:

Benchmarking

Benchmarking tip #1: How to compare trajectories

Let’s say you’re trying to report on the performance of a newly launched market.

If you simply put the performance data for the new market next to the data of the established ones, people will either think it’s doing terribly, or they won’t know how to interpret the data.

But this isn’t a fair comparison; the established markets had months or even years to improve over time, while the newly launched market is still ramping up.

✅ The fix: Indexing the data by the launch date for each market1. That way, you can compare the trajectories and see if the most recently launched market is developing better or worse post-launch.

This way, you can see clearly that France is growing slightly faster than the other markets, despite having the lowest numbers as of today.

Benchmarking tip #2: Don’t compare apples to oranges

If you pick the wrong metrics or don’t add enough context, you risk comparing apples to oranges.

For example, take a look at this chart:

If you present this without any explanation, your audience will likely think “We should just scale Organic; it’s super efficient!”.

But this comparison is misleading.

Your paid channels like LinkedIn and SEM can be scaled somewhat linearly by investing more money in ads. Organic, on the other hand, is much harder to influence (depending on what your attribution counts into “Organic”, you will have to invest in SEO, content, and your brand efforts, which can take months to show results).

Takeaway: Pick the right comparisons, and clearly communicate any caveats.

Benchmarking tip #3: How to compare actuals vs. plan

If you have ever worked in an analytics role, you’ve likely gotten questions from stakeholders on how a product / team / business unit is performing.

How you present the data here matters a lot. If you simply present actuals vs. the forecast as a snapshot, it will always look like the current period is not going well:

And even if your audience is aware that the current period is incomplete, they can’t easily tell if you’ll eventually hit the goal or not.

To address this, you can show how you’re tracking towards the goal throughout the period. And to make things even clearer, you can add an extrapolation of where you’ll likely end up based on the partial data you have:

If your company regularly updates forecasts, it can be helpful to visualize how they have evolved over time. This allows you to see if you tend to be overly conservative or optimistic.

Here’s an example from Wall Street:

Highlighting a mix shift

Line charts are great when you’re trying to show a change over time or benchmark relative performance like in the market launch example above.

But sometimes, what really matters aren’t absolute numbers, but the changing composition of a whole; this is called a mix shift. In that case, stacked bar charts tell a much clearer story.

Let’s look at an example. You’re checking a dashboard and see that your overall conversion rate from signup to active user is declining. When you plot the conversion rate by user acquisition channel, it’s not clear what’s going on; the conversion rates seem fine.

A line chart of signup volume gives you a first hint, though: Some channels like Organic are declining, while paid channels like Paid Social and SEM are increasing.

Switching to a stacked 100% bar chart gives additional insights: You can now see that over the last 12 months, the share of Organic signups has dropped from 40% to 25%, while Paid Social and SEM have increased by a similar amount.

And when you put it side-by-side with the overview of conversion rates by channel, the story is crystal clear: Volume is shifting from the highest-conversion channels to the ones with the lowest conversion. As a result, the blended average conversion rate is plummeting:

Highlighting inflection points and step changes

Everyone loves a good elbow or step change in the data. It’s not always there, but when it is, you should highlight it.

There are three ways you will encounter this in practice:

1. An inflection point in a trajectory. The trajectory permanently changes at a certain point; this represents a change in the growth rate

2. An elbow in a plot between two variables. You’ll often find that relationships aren’t linear; instead, there is often a point of diminishing returns beyond which a change in variable A leads to little change in variable B

3. A step change. Instead of a change in the trajectory, you’ll sometimes find one-off changes that bump a metric up or down

Often, these drastic changes can be tied to meaningful changes in the business and once you’re aware of them, it becomes much easier to make sense of the data.

For example, here is the subscriber growth chart of :

You can see a stark inflection around May 2022. That’s because Substack launched its Recommendations feature and thousands of other newsletters started recommending Lenny.

If you want to make sense of Lenny’s subscriber data (e.g. forecast what growth he can expect in 2025), you need to make sure you split the data into two phases: “Before Recommendations” and “After Recommendations”.

Note: Just because two things happened around the same time, doesn’t mean there is a causal relationship. In this case, we could validate our hypothesis by looking at a breakdown of Lenny’s subscriber growth by channel.

Showing how to get from A to B

Showing what it takes to get from A to B is one of the most powerful data stories you can tell. The need for this comes up at work all the time:

• You want to show how you plan to turn around an underperforming segment

• You want to break down your company’s profit or loss, i.e. show where the money goes

• You’re trying to illustrate how your startup can ultimately become profitable when it’s currently losing money on every transaction

WeWork & SoftBank taught us what not to do:

Instead, create a waterfall chart that breaks down the path from A to B. That way, you can quickly see 1) which factors play a role and 2) how much each one contributes (example from The Information’s reporting on OpenAI):

Highlighting underlying drivers

Data in isolation is difficult to interpret. So if you want to tell a clear story with data, you need to put it into context.

Let’s say you’re presenting a chart on the number of deals prospected by the Sales team, and it’s lower than the plan. Most people’s first assumption would be that the Sales team is not performing well and you need to increase productivity.

In reality, though, it could be a very different issue, and it’s our job to figure out what’s going on and then present it clearly. Breaking a high-level metric like “number of deals” into its drivers can be a great way to find the underlying trend that explains what we’re seeing.

For example, it could be that you didn’t hire fast enough (or had attrition on the team) and that productivity per Sales rep is actually above plan:

Best practices when presenting data

If you read the article up to this point, you have seen some of these best practices in action already.

Here’s a summary of the most important ones for your reference:

1. Make titles meaningful

How many times have you opened a presentation and looked at a chart only to wonder what it’s supposed to tell you?

There’s an easy trick to avoid that: Put the takeaway in the title of the chart (or slide). This is the Pyramid Principle in action: The reader first sees the insight in the title, and then looks at the chart for supporting data:

2. Create the storyline before you create the content

If you want your presentation to tell a coherent story, you need to sketch out the storyline before you write the actual content.

If you follow the tip above and use slide titles to communicate takeaways, doing this is easy. Just open an empty presentation and write the key points you want to make in the titles—one point per slide—so that they form a complete story end-to-end.

Only once you’re happy with the flow, you should start adding the actual content (text, graphs, tables etc.) to the slides.

Besides helping the presentation flow better, there’s another benefit: Many people that open your deck later on won’t have time to read all of it. If your titles tell a clear story, they can get the gist by just flipping through it.

3. Visualize deltas and growth rates

If the point of your chart is to show the difference between two numbers and you just show the numbers side by side, you’re leaving the work of quantifying and interpreting the delta to your audience.

Instead, explicitly highlight the delta as a percent change (e.g. “+22% month-over-month”) or in terms of goal attainment (e.g. “80% of goal”).

wrote a great piece that dives deep on this topic.

The same can be done for compound growth rates; instead of letting people calculate the growth rate over time, you can show it as a CAGR (Compound Annual Growth Rate) or the monthly equivalent directly in the chart:

4. Make it clean

A lot of data stories don’t have the desired impact because of sloppy delivery.

This is not about making charts “fancy” or matching corporate colors or fonts. The important part is that the chart is easy to parse. Some best practices:

• Clearly label your axes

• Always include a legend

• Smooth out granular data (e.g. daily trends)

• Remove excessive detail

Here’s what that looks like in action:

5. Keep it simple

As a general rule: Avoid complex charts.

Most charts should have a single message, and usually a single axis. Only add more complexity (e.g. a second axis) if it really helps to have it in the same chart; otherwise, just create a separate chart for the other point you want to make.

The same applies to slides; a good slide should have one key message that goes in the title and is supported by data in the slide. More data != more impact.

6. Provide insights, don’t just recap

Approximately 9 out of 10 data people do not write insightful commentary.

The point of comments is to add additional context or derive a “so what?”: What does this data mean for the business? In practice, though, most of the time it’s just restating what you can see in the data:

As you get more senior, it will be more and more common that your team is preparing the analysis and creating the charts, and you are only writing the commentary.

In other words, adding insightful comments will become a core skill, and it will pay off to invest in this early on.

It’s one of the easiest ways to stand out and get noticed by executives.

🔔 Data storytellers to follow

The best way to learn data storytelling is to see it in practice. Once you’ve seen enough good examples, you will automatically start applying the best practices yourself.

Here are some of our favorite data storytellers to check out:

• Peter Walker from Carta (LinkedIn, Newsletter)

• Jason Saltzman and Ethan Elias from Live Data Technologies

• of

• of

• The Data is Beautiful community on Reddit

Who did we miss? Let us know in the comments.

If you enjoyed the post, consider sharing it with a friend or colleague 📩; it would mean a lot to me. If you didn’t like it, feel free to send me hate mail.

It’s 2025! If you’re like me, you’re starting into the new year with big plans and tons of motivation. So many things we’re gonna get done!

Unfortunately, soon, reality is going to slap us in the face. Days in 2025 have the same number of hours as those in 2024, and once that initial “New Year’s Resolution Motivation Boost™” fades, we’ll be back to our old patterns and productivity levels.

Time is the ultimate bottleneck—both in our personal life and at work—and the only way to drive more impact is to use it more effectively. In other words, you need to be more intentional about what you work on and when.

There are two fundamental considerations:

1. Bandwidth: How do you decide if you should take on a new task?

2. Prioritization: What should you be working on right now?

Now, you might think “Ugh, yet another productivity post”, and I’m with you — the topic has been beaten to death. But I’m not going to cover any general-purpose productivity techniques like the Pomodoro method here.

Instead, I am going to share concrete things that, in my experience, differentiate people who get stuff done in high-growth companies from those that frequently miss deadlines or drop the ball.

As always, I am going to get very tactical with concrete examples so you can apply this directly on the job.

As a reminder, this is Part 3 of a multi-part series on nailing the basics:

Part 1:

Part 2:

👉 Part 3 (this post!):

Part 4:

Let’s get into it.

Tracking what to do and when to do it

Before we get into the specifics of managing your bandwidth and priorities, let’s briefly talk about logistics.

Many productivity systems tell you to either use a to-do list or a calendar, but I personally find it helpful to use both.

Why do you need both? Isn’t that duplicative effort?

It might seem that way, but I think of them as complementary rather than substitutes. Let’s start with why I think you need a to-do list (whether hand-written or in a tool):

• It serves as a “brain dump”. By putting everything you need to do in one place, your brain doesn’t have to constantly worry about forgetting something

• In addition, there are many tasks that can be knocked out quickly and where it’s overkill to schedule time blocks in the calendar

So, why do you then need a calendar in addition? Blocking time for specific tasks on the calendar serves two main purposes:

1. 🛡️ That time is protected. When I didn’t do this, the day usually filled up with other people’s meetings and random things and I ended up doing most of the important work after 6PM

2. 👀 Time-blocking visualizes your bandwidth. For example, if you can’t find a block in your calendar to work on a task before date X, it’s not going to get done before that (unless you reprioritize)

Note: You don’t have to actually stick to the exact time blocks you schedule. What’s more important is that all important tasks are accounted for. You can (and should) move things around as your priorities change. We’ll get more into that in the next sections.

And don’t forget about routine tasks: We rarely think about them, but they take up a lot of time collectively.

For example, if you are running a weekly metrics review and you need to prepare a deck, reserve a block a day or two before. If you need to check data pipelines or dashboards at the beginning of the week, schedule time for that.

You’ll be surprised how little time is left for project work after meetings and these routine tasks are accounted for.

Managing your bandwidth: How do you decide if you should take on a new task?

Whether your team runs a formal process or your manager yells requests at you in the hallway: You need to manage your bandwidth.

Otherwise, you can’t confidently commit to taking on new projects.

This consists of four things:

1. Understanding what you would need to do for the new project

2. Deciding if you should do it

3. Assessing your current situation

4. Working with your manager when you’re hitting full utilization and can’t take on any new work

Step 1: Understanding what needs to be done

Managers often ask vague questions like “Do you have bandwidth to work with Sales on the lead scoring work stream?”. It may seem like a simple question, but you can’t really answer that without more information.

• 📅 What’s the deadline for the project?

• ⏳ What stage is it currently at? Do you need to kick off the project from scratch or are you supposed to help bring it over the finish line?

• 🙋‍♂️ What is your role supposed to be vs. what will others own? E.g. in the example above, is Sales owning the model development and you’re just advising?

If your manager doesn’t have this information, you can offer to do this scoping for them (without committing to owning the work).

This information is needed whether you or someone else will end up being staffed on the project, so either way you’re helping your manager.

Note: NEVER trust other people’s estimate of how long something will take. Stakeholders (and your manager) always underestimate things because they likely have never done it themselves.

Always collect the information above and form your own opinion on how much time you’ll need to do a good job.

Step 2: Deciding if you should be the one to do it

Developing a reputation for getting stuff done is great.

But it also comes with a major drawback: Everyone will just come directly to you with their requests. Why go through your manager or an official intake process if they can go straight to the expert?

Luckily, just because a stakeholder comes to you with a request doesn’t automatically mean you have to do it.

Often, you can refer them to someone else who is better positioned to help them, or you can point them to self-serve resources (e.g. a wiki article, a query they can tweak, a raw data pull etc.) so they can solve their own problem.

Step 3: Assessing your current situation

Once you understand what the project entails and you decide it makes sense for you to work on it, you need to figure out if you have bandwidth to take it on.

Nobody can forecast accurately how much time a project will take, especially if it takes weeks or months to complete. Luckily, you don't really need to do that.

99% of projects or tasks that come up outside of planning cycles need to be handled urgently. That means in order to decide if you can do it, you just need to understand your other commitments in the next few weeks.

That’s where a well-maintained, time-blocked calendar comes in.

If you blocked off time for everything you already need to work on, and based on the information you gathered about the project, you’ll quickly get a sense of whether you can fit it in.

This is not an accurate science, and it doesn’t have to be. All you need is a rough gut check:

• You have a few hours of spare time in your calendar and your manager wants you to give a presentation to new hires? ✅ No problem, you’ll make it work

• You’re almost booked out for the next two weeks but you’re asked to lead a cross-functional metrics standardization work stream? 🚫 No way that will happen without reprioritization

The goal is to make sure you don’t put yourself in an impossible situation. Adjusting on the margins is easy: If you need a few more hours for the task than you anticipated, you’ll be able to find them by skipping meetings or pushing things to next week.

Not all work is created equal

💡 Pro tip: Don’t just look at the amount of time you have available, but also what kind of work you have on your plate

Squeezing in a few more hours of routine work is easy; if you have the time, you can do it.

But in my experience, you can only really manage one or two complex projects at a given time. The more you have to context-switch between topics, the more likely you are to miss something important and drop the ball.

In addition, most of the solutions to difficult problems come to me not when I sit down to solve them at work, but during evening walks and other activities where my brain is working in the background. And that subconscious problem solving works best when I don’t overload my brain with too many competing topics.

Step 4: What to do when you don’t have time to take on new work

You shouldn’t commit to additional work if you don’t have enough time to do it.

But here’s the thing:

❌ You can’t just say “No”. If you did, your manager would be wondering:

• “Are they truly stretched or just not motivated enough?”

• “What’s more important than this? Can’t they just make time for it?”

That’s nothing personal. It’s simply because they don’t know exactly what you’re up to all day (and you should be glad that’s the case, otherwise you’d feel micromanaged).

And there’s another problem as well:

If you reject any new project because your plate is full, you’re implicitly saying that just because a project came earlier, it’s more important. That makes no sense.

Here’s what I recommend you do instead (this has worked well for me at any level, even when working with C-level executives):

Stack-rank your priorities, including the new ask, and present your manager with a trade-off

In one of my last roles, I already had a lot on my plate when my manager asked me to create a monthly business review for our function. I agreed that it would be a valuable thing to do, but it was a meaty project and I didn’t have any spare time to work on it.

But instead of just saying “No”, I looked at my calendar and figured out how I could make room for it if I had to. This allowed me to give my manager three options:

This approach comes with a few key benefits:

1. Your manager gets a reminder of your priorities (you’d be surprised how often managers forget what each team member is working on)

2. Instead of arguing over whether you can do it, you’re reframing the discussion as “Here’s what it will cost you to do it”

3. Your making a recommendation (which is expected of any senior employee), but they still get the final say

Try it. Every time I did this, the discussion immediately became much more productive. Instead of being antagonists in a negotiation, we were suddenly partners in solving a problem.

Prioritization: What should you be working on right now?

What can you work on?

Before you can decide what you should work on, you need to figure out what you can actually work on.

A few years ago, my routine typically looked something like this:

• I opened up my to do list

• After briefly scanning through all items, I decided I should work on a task related to quarterly planning; it was important, and planning was due soon

• After spending 20 minutes digging into the materials, I remembered why that item had been sitting on my to do list for a while: I was blocked and waiting for input from Finance

Going through this every couple of days is massively inefficient and feels like Groundhog Day.

To avoid this, I now mark all items where I’m blocked with a simple status (and move them into a separate section in my to-do-list):

The key is to make sure you don’t stay blocked forever. So if you’re waiting for someone to get back to you, set yourself a reminder to follow up with them by a certain date.

But be careful not to use “I’m blocked” as an easy excuse not to work on something.

Based on experience, I’d say that roughly 50% of the time when people think they are blocked they are not, and simply need to think more creatively about how they can move forward.

Here’s an example from a few years ago:

A direct report on my team was in charge of competitive intelligence. We were standing up a new program where we’d purchase data from a third-party vendor and then create reports and dashboards on top of that.

We were expecting to get the data in a few weeks, so my report wanted to focus on other tasks in the meantime. After all, you need the data to build the analysis, right?

But there was a problem: Leadership was eagerly waiting for the competitive intelligence reports. We knew with 100% certainty that once we got the data (which we’d paid a lot of money for), we’d get daily pings asking us whether the insights were ready.

And: As part of the due diligence process, the vendor had given us sample data to review. So we already knew what the data looked like!

As a result, there was no reason to wait until we got the full dataset to start working on the reports. We could build the entire thing based on the sample data and then just swap it out for the final dataset once we got it.

The moral of the story: Even if you’re blocked on some part of a project, there are usually other pieces of work you can pull forward to not put the timeline at risk

What should you work on?

Now that you understand what you could work on, let’s figure out what you should tackle right now.

The Eisenhower Matrix is a good starting point:

You should have a good idea of what’s important based on your company’s and team’s priorities and what your manager told you. But figuring out what’s urgent at a given point in time requires a bit of work.

Here’s what I like to do to make sure I work on the right things at the right time:

Step 1: Housekeeping

• I add deadlines for all important projects to my calendar as all-day events (many to-do list and project management tools allow you to do this via a sync)

Step 2: Planning

• Once a week, I look at my calendar for the next 2 - 3 weeks to see what’s due; this includes project deadlines as well as meetings that need preparation. I have a time block for this to make sure I do it even when I’m busy (which is always)

• I then work backwards from these deadlines and make sure that I have sufficient time blocked off to work on these tasks

Step 3: Readjusting

• Every evening, I look at the next few days and my to-do list to see if anything changed and I need to adjust my plan (e.g. somebody might have scheduled a new important meeting, or I might have gotten a new ad-hoc request)

This is a good start, but there’s one weakness: It only accounts for work where a clear deadline has already been set, or a meeting has been scheduled.

If you want to stand out and take things to the next level, I recommend you look beyond your own calendar:

• Last year, quarterly planning kicked off in early March? It’s probably going to happen again this year even if Finance hasn’t reached out yet

• There are a few new hires starting on your team next month? Even if your manager hasn’t reached out yet, you might be asked to help with onboarding

Getting ahead of these things will 1) prevent you from having to reshuffle your entire calendar when they do kick off and 2) make you look thoughtful and highly organized.

The above process works if you have enough time to comfortably get everything done before the respective deadlines. But 99% of the time, that’s not the case: You’re stretched and you need to prioritize. So, what can you do?

Let’s clean up first. Anything that’s neither important nor urgent shouldn’t have been on your list to begin with, so delete those tasks. Out of the remaining tasks, those that aren’t important can be delegated out (if possible) and those that aren’t urgent can be pushed to later.

That leaves you with the urgent and important tasks. How do you prioritize among these?

Core principle: Maximize the ROI of the marginal hour

Most people tell you to rank your projects by importance and work on the most important one until it’s done, then move on to the next one and so on.

I think that’s a mistake.

No project is ever going to be perfect, so with the approach above, you’ll spend way too much time trying to perfect your most important task and then run out of time for everything else.

Instead, try to maximize the Return on Investment (ROI) of the next hour you’re going to spend.

• Should you spend another hour tweaking the executive dashboard you’ve been working on for a week, or should you take something else from 0 to 1?

• Will the forecast get materially better if you keep changing the assumptions, or is your time better spent moving on and debugging those SQL queries you need to push to production?