How do you learn to be a better CEO?
A system for gathering inputs, simulating a problem space, and generating feedback loops from application to become better at my job
This is a long and detailed breakdown of how I think about the challenge that this whole newsletter is about: learning to be a better CEO. I hope this will be interesting for ‘systems thinkers’ or ‘self-improvers’ in the knowledge economy, regardless of the role you’re trying to improve at. I detail below a very thorough way of thinking through ‘active learning’ in this type of role. You might connect with it, and might find it outlandishly over-complicated. Let me know!
(skip to ‘THE SYSTEM’ if you’re a low-context kinda reader…)
The title of this post sounds a bit clickbaity, but I mean to answer it literally; the title of the newsletter is the deliberate CEO because I think a lot about learning to do my job better, and this post answers how I actively learn to be a better CEO. This is not advice, it’s just me exploring and refining part of my process in public.
As the readership of these posts has grown, I’ve sometimes been asked “how do you make time to write your newsletter?”. Each of these newsletters is an open exploration of a topic on my mind as I try to do my job better. They’re long and open-ended because they’re a reflection of my current thinking on a current problem. First Round Review et al. do a brilliant job of giving answers and frameworks - problems found and solved. I do the opposite. The problems I’m tackling are ones most of my readers probably are tackling, or will tackle, written about and explored in near-real time. As such, this newsletter is not me ‘taking time out’ or ‘finding time to write’ or whatever else - this is me learning, which is part of our jobs, right? The newsletter format forces me to crystallise my thinking so I can apply it better at work. Since I wrote Leading with 'what' not 'how' as a startup scales last month, I’ve thought about it, discussed it, and mentally referenced it hundreds of times. I’ve held myself back from the brink of the ‘how’, and pushed myself to set a clearer ‘what’, with a shorter cycle time and more frequently than I would have done had I not sat down and got that piece out of my head.
This newsletter is my public accountability system, but it’s also my version of what Tyler Cowen calls “the knowledge worker’s equivalent of a pianist practising scales”. This newsletter is not the concert, it’s the practise session. You hear the scales and mistakes, not the concerto.
A mental model for active learning
A brief foray into another passion of mine. Learning a language, you start with inputs (learning vocabulary and grammar), and you end with application (travel, communication with natives).
Serious language learners also add a middle stage, simulation, partially alone, but mostly with a teacher or practise partner. The role of this selected interlocutor is to help you map the inputs to the application. My Chinese teacher and I simulate conversations, practise grammar, swap vocabulary in and out, change contexts, test unexpected variations. Good simulation is a safe, playful space before the cold hard reality of application. You should not jump from inputs straight to real-world application. As an adult you can’t learn to speak Spanish on Duolingo alone, because Duolingo doesn’t create a good simulation space. This is an extremely hard, paradigm-shift type problem (…that we at Lingumi are tackling in the pre-school teaching space, as it happens).
The problem set you can tackle in your simulation space needs to be constantly expanding to keep you in your zone of proximal development (ie learning at maximum pace), so you need to add new inputs progressively. Inputs are a mix of new information and error feedback. For me as a Chinese learner, my inputs are 25-50 new Chinese characters a week, and about 250 repeated ones to revise. I then have 2 hours of live simulation (a pretentious way of saying ‘I have lessons’) with my teacher over Zoom each Saturday, working through open conversation and structured exercises, in which we integrate and play with these novel inputs. It’s safe and fun. I then try to apply all of this when I travel to China for work. That application stage happens infrequently, and the stakes are much higher - every Chinese learner has anecdotes about mixing up similar-sounding but entirely different characters (try ordering 睡觉 instead of 水饺 in a restaurant…).
Most people think they’re either ‘bad at languages’, or ‘can’t connect the words to real situations’ as if there’s some magical linguistic gene they lack. I’m not a neuroscientist, but I don’t think that’s true. We just don’t recognise the bottlenecks in our own learning processes. It feels damn scary using newly-learned language in real application situations IF the input set is limited, OR if the simulation space is limited.
Applying this mental model to the CEO role
I maintain absolutely no illusion that I, nor anyone, have some natural ability to be a great CEO. I think it’s essential to perceive our jobs, whatever they are, as things we actively study to be better at. By doing that, we increase our probabilities of success…though without removing the statistically significant risk of failure (as the wonderful book Algorithms to Live By covers well). So going back to Tyler Cowen, how much do you feel you’re practising your scales, rather than just playing concerts unprepared?
Applying this mental model from language learning to how I learn to be a CEO gives me a new way of thinking about my time and focus. If my leverage increases as the company grows, and I make fewer, bigger decisions (by delegating, and pushing strategy down the org), then not studying (inputs) and practising (simulation) would be like Michael Jordan taking free throws in matches without training between seasons. My decisions (application) would degrade over time.
So basically I study HARD! I want to get this right - our mission, the team, the investors, my personal ambition all rely on me scaling myself, being progressively better, even as the game levels up and gets harder. Going back to the graphic from NFX that I pasted in my ‘Builder to Architect CEO’ piece, this learning element feels missing. Perhaps they should add “luck & pluck” —> “deliberate self-improvement”.
What do you do to learn to be a better CEO?
This is a question I’d pose to you: what do you do to learn - or study - for your job as CEO?
In the old days, the answer might have been “do an MBA”. An MBA probably gives a certain flavour of inputs (case studies, theory…the fancy stuff, but not the hair-on-fire stuff?), and some level of simulation, but no application, I think? The learning model I’m proposing works because it’s integrated into a tight loop: you have to continually expand your inputs, continually simulate, continually apply - pulling any part out, or freezing it, breaks the loop.
Nowadays, the trendy thing to say is “I just make tons of mistakes and learn fast”.
I’m not suggesting that I don’t embrace making mistakes. I make lots, all the time, and learn a lot from them. But as CEO leverage grows, mistakes made on the pitch are (a) negatively impactful on the growth of the business, the mission, or your ability to create shareholder returns, and (b) can damage team members’ lives or wellbeing if they’re serious. “That manager was a disaster, what a bad hire!” is a mistake made and learned from in the CEO’s version of the anecdote, but it’s >=5 people’s career growth, mental health, and work/life balance negatively, even irrevocably, impacted in another telling…one that I may never even hear, alas. So I don’t think the de-facto CEO mindset should be “I learn to be better by trying things and making mistakes”. I think that’s an excuse for not making time for active learning. Mistakes are fine, but not the optimum?
So, here’s my system! I am conscious this whole post could be labelled “over-thinking it”. This is just my method. Take from it what you will. Avoid like the plague if you wish.
Gather diverse inputs
Like learning a language, I begin with gathering and absorbing inputs selected to expand my underlying level of thought & knowledge on a topic. I spend a lot of my time (~15-20 hours a week) reading or listening to blogs, podcasts or some business books (recent favourite: The Culture Code recommended by my colleague Brandon. Thanks!) I do that at work, after work, out on runs, while washing up - it’s a fairly constant stream of thought-provoking stuff. As readers of my piece on non-fluffy strategy might remember, I often seek out blog summaries of business books because they shorten time-to-value.
I sometimes talk directly to peers or mentors to get very specific technical inputs, like “what is your % pension match policy”, where sufficiently contemporary or localised public content doesn’t exist on the topic. I am in a founder group run by one of our investors that is superb for this stuff. Beyond that, I think peer discussion / talking to people is quite inefficient for gathering inputs for the same reason most Clubhouse talks are quite rubbish - the signal/noise ratio is very poor. I know this sounds sniffy, but I have three good reasons:
I rarely approach interactions with peers or mentors with sufficient preparation to make it impactful for me, or them. Sending notes, questions, and other inputs you’ve already drawn on ahead of time improves the signal/noise ratio of the conversation, and makes them feel they can be more helpful too.
If you perceive your time as being valuable, and you can absorb written information faster than spoken, it is rational to gear your inputs for, say, 3/4s written, and to select carefully for source. I evangelise First Round Review partly because the signal is so consistently high-quality.
A peer/mentor group is likely to be quite homogenous unless you design for diversity of input, and I’m not calculating enough as a “networker” to do that very well.
When I write ‘diverse’ here, what I mean is I aim to gather different views on the same topic: the for and against, Methods A, B and C to achieve a goal. I also try to push myself to read stuff I’m not that interested in, but is from a totally different world, so might offer later opportunities for cross-pollination of thought or just ‘arm me for the future’. For example, while I’m not very interested by the mechanics of IPOs or the hype around SPACs, I pushed myself to stay on top of the recent wave of writing on the topic to ensure I’m equipped for the day, if and when it comes, where a business I am part of can begin to evaluate and debate instruments for listing.
Capture, synthesise & refine inputs with tools
It is extremely inefficient to absorb so much content if you’re not capturing, synthesising, and refining it in a manner that suits your brain. This is the most “work” (scary, right?!) in the process, but is critical. Nicolas Colin, who writes the brilliant European Straits newsletter, has a voluminous and well-tagged Evernote of seemingly everything he’s ever read, and remixes it constantly in his newsletters, surfacing quality signal on anything and everything.
For me, my tool of choice is Roam Research. Roam is a much-hyped note-taking tool, and with good reason - its deceptively simple concept is that any note can be linked to, or referenced by, any other note. Instead of having fixed pages, collections of ideas are assembled based on shared tags or words. It’s inspired by the Zettelkasten technique, and I love it because it’s a reflection of the brain’s natural associative tendencies.
I’ve been on Roam for 7 months now, and my intersecting ‘graph’ of notes is growing reasonably large, with innumerable interlinking topics:
Previously, I used a mashup of Evernote, Bear, and Notion, and none of them quite fitted how I thought. Roam has stuck.
A rank of my topics and mentions gives you a good insight into the topics I look for inputs on, the people I have recently talked to or read about, and the extent to which those are connected to other thoughts. For example, 'I reference ‘strategy’ in 42 notes or pages, and often turn to the Strategy page to re-read all the assorted notes I’ve tagged with that word. More recently I’ve been gathering inputs on ‘high performance’ and ‘culture’ a lot, so those concepts are being referenced all over my notes. This feels like just pasting an X-ray of part of my mind for you all, but here goes:
My approach to capturing the inputs is to copy in snippets from my reading or listening (manually transcribing where needed), link to the source, and tag them as appropriate. Here’s an example if everything I added on Feb 26th:
As I go, I’m already inter-linking new notes to existing ones, like referencing a pre-existing collection of my favourite [[culture decks]]. When in future I add anything on the topic of [[high performance]], it’ll join the notes from discussions with my colleague Ken, or from reading Tyler Cowen, in an associated list of disparate inputs for processing.
A good inputs management system also lets you find tiny moments of leverage day to day. For example, when a team member during a skip checkin-in brought up a problem, but without suggesting a solution, I thought “ah, that feels quite Level 1”, went to Roam, found Ken’s page (my mental ‘tag’ of the article was the person who sent it here), and shared them the link. I’ve never got along with browser bookmarks for all sorts of reasons, and this is helping me solve that problem.
Roam might not be the right input tool for everyone - you know your brain - but lacking a perfect memory palace, I find it the next best thing.
So, the inputs are gathered. Next up is the critical intermediary phase: creating safe, playful simulation space to practise and test the application of the inputs before applying them in real life. Here’s how I think about that bit.
I think of this in two parts: private synthesis, and collaborative simulation.
Synthesis: which combination of inputs feels right for our situation?
The first step of simulation for me is to do it on my own private synthesis. This feels like drawing football tactics on a chalkboard in an empty classroom, long before I arrive at the pitch or meet the team. It lets me test and expand my own perspectives.
In Roam, I run comparisons of ideas, or remix several ideas together into new pages full of references (retrieved in Roam with ((double bracket)) notation). This process creates a natural filtering effect - testing inputs against each other, against your instictive principles, against company culture. Moving from inputs to application without simulation would be dangerous: if I implemented everything I read and liked in the moment, the culture would yo-yo in a volatile fashion. If I felt stressed about revenue growth one month we’d swing to a sales-heavy culture lacking innovation and long-term thinking, while the next we’d become blue sky creatives dreaming of new features, because we’d had a bumper revenue month. Moreover, If I wholesale implemented one idea I liked from one input, we’d be over-fitting someone else’s solution, rather than remixing to refine a solution that works for our culture, team, and mission. Synthesis is the critical, private first stage of my Simulation.
Let me give an example. My [[High Performance]] inputs lately have been deliberately divergent. After reading the provocative ‘Amp it Up’ by Frank Slootman, I was buzzing with ideas about introducing more intensity to our culture. My notes reflect that:
Thought-provoking! But to effectively simulate - to see if this would work at Lingumi - I needed diverse inputs. So I went in search of them, and collected a bunch of thoughts as counter-points, many from ‘The Culture Code’:
Suddenly, this stuff was speaking to me louder - it felt more nuanced, less American, less enterprise sales-y, closer to our DNA.
The true synthesis begins when I try to bring these together to solve the challenge I perceive. Here is an example of the sort of sheet I might write in Roam, laying out the empirical truths I’ve observed or been told, the goals, relevant inputs, principles/guidelines we hold, and lastly, the unanswered questions raised by the synthesis.
This might feel long-winded, but it feels worth the time when I dive deep on something like this, because it gives clear answers to some of the existential cultural questions a team faces, like:
What should our cultural norm / behaviour be when we miss a big target?
What is our cultural expectation when an individual or squad is under-performing?
Given a thought exercise where a leader can focus 100% of their energy on ‘building psychological safety’, OR on ‘amping up pace and intensity’ (assuming some mutual exclusivity), which should a Lingumi leader do?
Good solo synthesis is absolutely critical to get to a basic set of things that I believe to be desirable and optimal before I am ready to make or give input to a final decision. I have never built an organisation before, and culture seems to be absolutely critical to the success of a company, hard to iterate fast on, and very ‘fluffy’ if not well-defined early and constantly upheld. These are not easily reversible things to implement.
The next stage of Simulation is like what I do with my Chinese teacher: simulation with a peer, mentor or colleague.
Simulation through structured conversations: test the ideas with confidants
My approach to Simulation depends on who I’m talking to. If my coach, a team member, or a board member, all of whom have intimate knowledge of the company, I am often using the session to test ideas - “If we changed x to y, would it solve the stated problem without conflicting with our principles?”, “What do you think we change about x?”
If an external peer or friend, my approach is usually to ask them to share experience of solving problems similar to these ones in their context, capturing new inputs, and then propose my ideas for my context, and ask them to critique.
It’s never quite as structured and formal as I’m making it sound, but I’m increasingly trying to be more structured and rigid with these sessions, because that’s how I find the signal/noise ratio improves fastest, as mechanical as it sounds.
The times where I have got the most valuable input on a question from conversations with peers, mentors or colleagues are when I have armed them well ahead of time, and then structured the conversation carefully around what I am hoping to gain from it. “Structured” is the key - the conversation HAS to be prepared for well if you want to use it for simulation. This means doing hard work, not letting the meeting just creep up on you. I try to send my notes - often pasted from Roam - ahead of time, and some topics I want to discuss.
I won’t collapse all the bullets, but here’s an example of a conversation with Matt Clifford, an investor in Lingumi and great thinker on organisational and technology topics, where thoughts I’d sent ahead of time prompted us to discuss performance, hiring, and organisational systems at a deeper level than we might otherwise have done:
Again, this might sound laborious, but it directly impacted high-leverage decisions. It led to a change in one role in my team, and became a critical component of my continued synthesis on culture. As new inputs arise in these conversations, I add those to a page marked as belonging to that conversation, and tag the relevant topics, synthesising in those new inputs to update the model. I feel like a really slow-moving, manual machine learning algorithm doing this sometimes.
Depending on the impact of the decision I’m making and the existing knowledge base I have, this could be a very quick, same-day process with 1-2 close team members / investors / peers to build confidence in something, and never enter my written notes, or it could be more protracted, but the loop is the same. Lately, I’ve been thinking about culture a lot, so have tried to maximise the number of simulation sessions I’m having with people I rely on inside and beyond my team.
I’ve mentioned adding fresh inputs or updating the synthesis several times. This is the feedback loop that is so essential to the active learning process, both in language learning, and learning to improve in one’s role. Some feedback loops are rapid and visible: an outrageously high company sales target would get instant pushback. A non-compliant sick pay policy would have team members up in arms.
Unfortunately, most forms of feedback are slower and harder to see: I may be failing to establish clear cultural principles, or setting the hiring bar too low, or setting fluffly strategy, and not realise it until I reflect on why some star team members are demotivated, or growth is slowing, six months down the road. This is the thing I am most neurotic about in my job.
I don’t know how to “best” counteract this slowness, but my instinct tells me to hunt for more diverse inputs: how many team members am I speaking to outside my direct reports? What leading indicator metrics might be indicative, but disregarded? How many peers have I asked to critique our culture deck? Have I hired experienced enough execs to bring rigour to areas where we lack it?
Sound tough for a CEO? Imagine being a VC where you have to wait ~10 years to close your feedback loops. Ouch.
The moments of application of all of this are infrequent but critical, and they become less frequent and more critical as the company - and so the impact - get bigger.
When a team member says something like “I don’t think we have a culture of high performance”, there are two things I can do:
React instinctively in the moment, informed by how I feel, what springs to mind, and what I had for breakfast.
Respond with socratic questions, narrowing down to the drivers and analysis behind the comment to make sure we’re on the same starting line, and then respond, informed by many hours of reading diverse inputs, synthesising and simulating alone and with peers, and really actively thinking about the topic.
This is the moment of application. I wish I did (2) more than (1), though I don’t think I’m there yet. When I meet with one of my mentors, John, I feel through our discussion how he was forged in the fires of this second mode of thinking after years in a CEO seat, and I often find myself thinking “What Would John Do?”.
Maybe what I say in that high-leverage moment will lead to a change in our company values, or our Q2 strategy, or team formulations, or team goals. Maybe it’ll change the hiring plan. Maybe someone will get hired or fired. All of those are serious, ‘one-way door’, hard to reverse changes. If what I wrote earlier is true - that over time a CEO’s leverage increases as the company grows, and they make fewer, bigger decisions - then for me, I feel it is my responsibility that each important conversation or decision - the moment of application - should be informed and strengthened by the foundation of active learning, with rigorous inputs and simulation undergirding it.
It’s so difficult to feel confident as a CEO. We are ultimately responsible for the happiness, fulfilment and salaries of a big group of people who have believed what we’ve promised, who have put part of their life into something we control. We are ultimately responsible for millions of pounds of investor capital that we feel a fiduciary duty to return many times over. And we are constantly looking ahead at the unknowns - will we hit our projections? Will our cashflow inflect in the way we predict given hire X or launch Y? Is person X the highest impact individual on the market for job Y?
I feel reasonably comfortable projecting confidence when it’s called for, but I feel even better when I am confident that I’m acting in the right way, making the best calls I can, and so on. That emerges from active learning, when I get it right. For a second-time or tenured CEO, I suspect that the experience feedback loop does a lot of the heavy lifting here sub-consciously. The process may not need to be so deliberate. I think a lot about this active learning process partly because this is my first cycle doing this. Every decision and every stage of growth is a new level to learn to master, but without a trial run or respawn point. I need to be deliberate!