Effective Startup Ideation

October 30, 2022


The purpose of this essay is to provide principles and processes for discovering unicorn-trajectory startup ideas. These thoughts are based on substantial observation, ample reason, and occasional mathematics. I hope you will find them useful for your journey.

First: why this essay? There are plenty of other posts (notably, Paul Graham's and Sam Altman's) which emphasize the silicon-valley style of entrepreneurial ideation: focus on demand and iterate quickly and empirically in ways that don't scale. All of these points are right. Yet, my experience as an undergraduate was that they were often unactionable. Consequently, this essay is meant as a supplement to provide more specific and actionable methods and criteria. Due to their specificity, the points made here cannot be quite as universal. This essay therefore diverges from the more popular and brief treatises.

At their core, startups build solutions to problems. We use these terms here because they are deliberately vague. Your company's solution might consist of a wide range of products or services, and the problem your solution addresses may span a similarly wide range. In general, profitable companies are built around great solutions solving big problems. This essay is concerned with collectively developing the problem and solution, whose collection I will term as the company's idea. The idea's constituents retain a degree of independence in terms of the forces which act on them but are nonetheless closely linked. For example, change the inefficiency and you will certainly need to change the problem and solution. Similarly, feedback on the solution -- in terms of both feasibility and desirability -- may well affect the choice of problem and inefficiency.

The game we will explore here is how to find a sufficiently good idea as fast as possible. You will only need one because, at the start, your company will only be able to do one thing competently at a time. How good is, good enough? At the end of the day, that's up to you. I'd like to encourage you to set your standards for good enough to be extremely high. There are a few reasons this is important.

  1. Finding a good idea makes the work bearable. Building a startup is a long-term and intense endeavor. It is not a few-month sprint: if it goes well, expect at least five years of golden handcuffs. You need to be in it for the long haul, and you should therefore work on a vision that you'll be excited to pour your soul into for the next five or ten years. Otherwise, it is likely you will eventually become unhappy and lose your motivation.
  2. Finding a good idea increases the probability of success. It's most important for your happiness and your career that you succeed; how big you succeed is secondary. First, regarding happiness: in all odds, you will be happy if you build either a successful company or a very successful company, but running a failing company is contrastingly a categorically stressful and miserable experience. Second, with respect to your career, you should remember that the company you build now need not be the last company you ever build, and even a small success now helps beget larger ones later. For both of these reasons, you should therefore optimize strongly for the probability of success, and a good idea helps here too. If you have only a decent idea, then when something goes wrong -- perhaps some assumptions that you made don't hold up under closer scrutiny, or your users are not excited, or something similar -- it will be difficult for your company to weather that bump. However, if you have a really great idea, where it's plainly obvious that what you're doing is ten times better than all the competition, your company will be far more likely to survive the shocks to which all startups are eventually subjected. Furthermore, as a first-time founder, you will especially benefit from seizing every tailwind and avoiding every headwind. Your choice of idea will significantly determine from where the wind blows.
  3. Finding a good idea makes success much bigger. There are of course valid personal reasons to found a startup, such as autonomy, drive, and the development of skills, but by going down this path you should necessarily expect a huge amount of work and stress over a long period of time. If you succeed, the reward -- measured in impact, money, or prestige -- will be significantly determined by your choice of idea, so you might as well get that right and reap the maximum benefits of your hard work when all is said and done.

Now that I have hopefully convinced you that your choice of idea will be among the most consequential decisions for your company and perhaps also your career, and that you should therefore be particularly discerning, we turn to how to find this idea.

Some expectation setting is in order. To the best of my knowledge, there is no deterministic, time-bounded process to find a superb idea. After all, if such a process existed, it would be relentlessly exploited until all its value became depleted. (As is mostly the case with public markets.) Instead, great ideas have an element of chance to them: inspiration may strike like lightning on the first day, or you may wander in the desert for forty years before finding your oasis. Nor is it easy going. Budding founders often believe that ideation is a fun and relaxing affair before the supposed real work of the founder begins. We will dispel this myth.

Yet this is not to say that this fickle, unpredictable process is entirely random, either. While I cannot guarantee you'll strike gold in any particular place or at any particular time, there is a great deal you can do to raise the odds, and this is where skill, art, and disciplined effort come into it. Certainly, I hope that by the end of this essay you come to understand that effective ideation is among the most difficult and intensive efforts a founder can undertake. However, if the methods to be described are properly applied, I believe that you can raise the odds to the point that you can have a very good chance of finding a superb idea within a few months of disciplined work.

My personal analogy for this process is that finding a great startup idea is rather like making friends with a cat. The occasional cat is very extroverted and will allow you to approach and pet it without any concern. However, most cats are more skittish. If you wish to pet the cat, you can do the work of coming near to it, sitting down, and perhaps leaving it some food, but the cat must do the final approach on its own terms. Analogously, while some good ideas do lend themselves to being simply thought of, most do not. Instead, work can only get you near the idea; the idea itself must come to you. Y-Combinator terms these latter ideas organic ideas (and prefers them) -- the ideas which are not thought of but noticed. That the idea must come to you helps explain why startup ideation is often frustrating.

This essay is divided into two main sections. Part one is on doing: concrete actions and criteria for finding your idea. Part two is on being: qualities embodying the entrepreneurial spirit. While the bulk of the essay is spent in the first section, this is mostly because actions are easier to enumerate. If anything, I consider the second section more important, even if less approachable.

Regarding doing, the framework has five parts. First, we bias ourselves for success by choosing fertile ground. Second, we gather the right data, quickly -- and lots of it. Third, we think in principled ways so as to increase the chance we stumble on ideas others haven't. Fourth, we consider some heuristics which can lead to better ideas. Fifth, we constantly improve our strategy and processes.

Regarding being, we consider eight personal aspects of great founders likely to find great ideas. Unlike the components of the first section, which can be carried out, these can only be practiced.

Finally, through all this do remember that the process remains meandering. We are playing a stochastic game, and one should expect in general to bounce between all of these at least a few times before finding a truly excellent idea.

Part 1: Doing -- Concrete Actions and Criteria for Start-up Ideation

Finding Fertile Ground

Let's begin with the question of how to choose a broad problem area. This is, as best as I can tell, both beneficial and also necessary. It is beneficial because a better problem area will more likely lead to a great idea. It is necessary because without this focus the set of ideas is so enormous and disorganized that to coherently sift through them all is nearly impossible. Because the options for problem areas are usually countable, the approach I will suggest for doing this well is to essentially enumerate them and consider how they stack up on various meaningful measures. Below is a set of qualities that I think are important. Some of these attributes have a clear worse to better spectrum, whereas others may significantly affect the nature of the company you are likely to build without necessarily affecting the quality of the outcomes.


The first and, in my view, most important criterion is your genuine interest in the problem area. Every new company or organization will have to overcome obstacles that look insurmountable in the moment. It is far easier to stay focused, take a deep breath, pull the all-nighter if necessary, and find a solution if you are genuinely attached to the mission of the organization and convinced of its value. From all of my conversations with founders within Prod and without, I have become convinced that some of the most important and differentiated labor that founders and leaders put into their organizations is actually emotional in nature: maintaining confidence and composure even when things don't go your way, and believing in your vision and your team even when others don't. Surviving the ordeal of being a founder is as much a test of internal strength and perseverance as it is of any technical or social skill. Successful startups come when the founder is both optimistic and also correct. Every person is driven by their own unique motivations and impulses: some are concerned with making it; others wish to leave behind a broadly positive and lasting legacy; still others might have an unusual obsession with a particular technology or vision. Only you know your own psyche, but I will suggest that almost everyone will benefit from choosing an area in which they have an authentic interest.

The immediate challenge of the interest criterion is that it can be difficult to actually know your own interest in certain areas. First, one's instantaneous interest can often change with feelings of success or failure. When our companies are going well, it is easy to be excited and optimistic about them, but a sense of failure can just as easily cause us to lose interest and become pessimistic. To bring in the language of signal processing, it may therefore be beneficial in understanding your feelings to try to filter out the high-frequency noise away from your low-frequency underlying interest in the problem. I don't think it's necessary to consider only the zero-frequency mean of your interest. Other low-frequency components are probably positively correlated with your skills, underlying market sentiments, or other information which is genuinely relevant to the success of your company. Still, on a given day your level of interest may be primarily governed by a great deal of high-frequency noise -- the last email you received from a prospective customer, or perhaps even how much sleep you got -- and differentiating what is transient from what is real is necessary to actually know your interest.

The second challenge of the interest criterion is that it depends to a much greater degree on luck than the other criteria we will examine. If you are lucky, you will happen to be extraordinarily interested in a problem that satisfies well all the criteria to come. If you are unlucky, all of your interests will be in areas that bear the portents of failure. If you believe you may be in this unlucky category, then it may benefit your long-term success as a founder to take some time and further develop your interests. This is not a process that happens overnight. Rather, it requires both significant study and practice to begin to appreciate some area that you did not know interested you. Other than the high-level suggestion that your interests need not be static, I will not go into any more detail on how to find new ones. For now, let us hope that you are reasonably lucky.


Another question you should try to answer is: of all the people in the world, why are you the right person (or team) to build a company in this area? If not, how will you rapidly become the right person?

This is a multi-faceted problem. There are many perfectly good reasons why you might believe yourself to be the right person, but I will not try to rank or prioritize them because these tend to vary in importance quite strongly with the problem space. Your answer is also relatively unlikely to be singular: it is more likely to instead be that you are world-class at this one thing that you think is most important, great at this second thing which is quite relevant, and pretty good at these five other things that you will need at some time or another.

Still, here are some possible categories of reasons you might think yourself to be the right person:

  1. Technical skill. You know more about building things here than just about anyone else. This tends to be more importantly applicable at a graduate or professional level than at an undergraduate level, but can still be highly relevant. It can also be highly specific: what made Google special was not really whether Larry and Sergey were world-class programmers. It was whether or not they had the right ideas about how to build search engines. Having the right ideas is even more important than being broadly skilled. Still, you should approach this with some skepticism, too: how do you know that your ideas are the right ones? Can you test them? Why are the others, who believe that they have the right ideas, wrong?
  2. Social skill. You may have an extraordinary ability to motivate other people and get them to follow you, to communicate your ideas clearly (perhaps in more than one language), or to negotiate and capture value for yourself and your company. The necessity of each of these varies somewhat in terms of both the problem space and also your role as a founder -- a CTO might not need strength in business negotiations but almost all founders need to be able to organize and motivate a team.
  3. Knowledge of the problem space. Maybe you worked for a company, saw how it succeeded in a parallel segment or floundered in the segment of interest, and have learned exactly what is needed. Perhaps you spent years selling a certain kind of enterprise product and learned exactly what the customers want. Or you've done your homework and are confident that your first-principles analytic approach to the problem has led to an understanding of the dynamics nobody else has. In any case, you know what is needed in a way few others do.
  4. Network. You may feel that you're particularly well-positioned to bring together the right team, investors, or customers in a certain space. If you are a student, this may at first seem unlikely to you, but is more common than you'd think in some domains, especially as a student.
  5. Location. For some problems, where you are matters! If you're trying to access some foreign markets, it can help to be from these places. In other cases, you might want to be right at the center of pools of talent. In still other industries, access to manufacturing or cheap real estate might be important.
  6. Personal conditions. For some problem spaces, it helps to be able to take on some more personal risk. For others being risk-averse may even be a good thing. Whether or not you need to pay yourself a salary (let alone pay off loans) may affect the kinds of problems on which you want to work. You might have other constraints or assets too, such as the time you do (or don't) need to spend taking care of family or working.
  7. Mindset. You may also believe yourself to be unusually dedicated to the problem, capable of working extremely hard or challenging the status quo, or have some other relevant personal qualities. As with the first bullet, while these qualities can be real and significant assets, I would be skeptical of having much confidence in their ability to meaningfully differentiate you unless you have substantive and specific evidence to support your claims.

Importantly, remember that the list above is better described as categories of reasons rather than individual reasons (and they have some overlap among them too). You may need to do some additional work to decide both which individual categories and which reasons within them are right.

I would also like to stress that this is not an exercise in persuading yourself that you are the right person to do something. This test will backfire if treated this way. Instead, you need to behave as if you are a disinterested stranger evaluating you to try to understand if you are likely to succeed in this space. Some degree of debiasing is also in order. For example, it is well-established that men are more likely to overestimate their own abilities, and women are more likely to underestimate them. Such trends probably exist in many other axes too. One simple approach which can help is to try to work through these questions with several friends whom you trust to give you honest feedback.

Another approach to de-biasing this question is the following mini-game. Pretend you are an investor looking to get into this problem space. Start with 100 points, and try to divvy them up among the relevant categories or attributes. For example, for a deep tech startup working on fusion, if I were an investor I would probably allocate at least 65 of the 100 points to the technical skill of the founding team and the quality of their physics ideas. By contrast, for a startup working on providing marketing services, I think it is unlikely that I would allocate more than 20 points in the same category. After playing this game -- and you can play it as many times as you want while you work out what you think the right distribution is -- you can then try to assign the same hundred points to yourself and see how well you line up. As with the previous approach to this consideration, it can help de-bias and provide robust answers if you can also have other people who are familiar with the space do this too.

Finally, do also remember that you don't have to satisfy everything perfectly when starting! It's absolutely alright if, for example, you believe that your network is important, but that it isn't yet good enough. You can solve this! Rather, treat this rubric as a tool to help you decide to what extent you are immediately likely to be a successful founder in a given area and what additional assets or skills you would most want to advance.

Why Now?

Many great startup ideas have both a good reason why they make sense now, and, crucially, why they didn't make sense before. After all, if your idea is so great, why hasn't anyone already done it? Here are a few possibilities:

  1. Technical: The technology to do what you want to do simply didn't exist before.
    1. A first category of a technical why nows include the enormous set of companies enabled by the invention and wide-spread adoption of smartphones, ranging from publishing to games to transportation services.
    2. A second, smaller category includes a host of startups that exploit AI to produce copy. Why didn't they exist before? Because large language models, like GPT-3, didn't exist before.
    3. Another deeply technical why now, is that of Commonwealth Fusion Systems (CFS), in which the company became possible only due to the development and commercialization of rare-earth barium copper oxide (REBCO) superconducting tapes.
  2. Structural: Perhaps the ecosystem whose niche you are filling is nascent. Maybe you rely on companies or services that didn't exist. Alternatively, it could be that an existing company is failing and you sense opportunity.
    1. Regarding a nascent niche, Empiric Networks (run by my good friend Oskar) is a good example: it was not possible to build StarkNet's oracle a few years ago because StarkNet did not exist.
    2. For an example of the prerequisite services not existing, consider Dollar Shave Club, which depended on both the scaling of commercial shipping directly to consumers (to allow costs low enough to ship razor blades and still compete with retail!) and the scaling of a large enough online shopping ecosystem so as to be familiar to consumers.
    3. A good case of a startup directly taking on a competitor which had become mediocre is Harry's, which correctly sensed that Dollar Shave Club did not understand the market that they had created and were vulnerable.
  3. Behavioral: Consumers or companies may simply have the willingness to do things they might not have done before.
    1. A fascinating case study of this is that of music. In the early 2000s, consumers were generally skeptical of subscription models for music, preferring to own their collections. iTunes embodied this view, with Steve Jobs declaring that people don't want to rent music... I think you could make available the Second Coming in a subscription model, and it might not be successful. Since then, consumers changed their preferences, and subscription models like Spotify are far more popular.

As with the other suggestions of this essay, not every startup needs to fulfill these attributes in order to be successful. Certainly there exist great startup ideas that have been possible for a while, only no one ever thought of them. Yet these are probably the minority. Most great companies became possible only a little while before they became reality.

Industry Dynamics

A fourth aspect to consider of any problem space is its short-term and long-term market conditions. Is the area growing? Is it expected to grow further? Who are the major players, and why? Sometimes this can be a bit fuzzy, and that's alright too: the point is to get a good sense.

There are a few reasons why this can be a valuable criterion. Of these, the most important is that it gives some advanced insight into a company's potential for economic success. After all, you can raise tons of money, hire a great team, and even build a great product, but if the market itself is rapidly contracting it will still be difficult to achieve meaningful growth and profit. Here, the short-term trends are somewhat more important for two reasons: your company is both most vulnerable and also benefits the most from rapid success in its infancy, and also because the distant future is far more uncertain anyways than the near. Nonetheless, long-term trends are highly relevant too. For example, if you believe that the world is accelerating towards a green future, then a startup selling geology software to oil companies may have poor ten-year prospects even if oil profits are currently at record highs.

A second reason this criterion is important is that popular perception may impact the difficulty of getting others to buy into your vision. Even if you are right about the long-term value of the problem space and others are wrong, it will be far more difficult to raise money, hire talent, and generate demand than in a problem space that is commonly believed to be on the up-and-up. On the flip side, unpopular spaces may have reduced competition, which can help you get a critical edge. In any case, there is some degree of timing involved in getting this right. Many industries follow cyclical conditions of bust and boom, and the timing of these is individual. A fintech startup focused on bonds may be primarily subject to the decades-long short-term debt cycle, whereas a fintech startup focused on crypto may instead follow a months-long hype cycle. Be careful not to fall into the trap of believing you should start a company in some area just because everyone is excited about it right now. Many startups have failed for being either ahead of their time or at the peak of a bubble.

Third, considering the dynamics of your industry can illuminate the incentives at play. For example, a startup promoting interoperability in a winner-take-most industry may find it difficult to get the companies on board even when it's obviously in the consumers' interest. For the market leader, interoperability only serves to undermine its dominance. (A well-known instance of this is Apple's disinterest in making iMessage broadly compatible with Android devices.) For the other companies which are not market leaders, interoperability is currently in their interest, so the startup may not be infeasible, but only more difficult.

Finally, deeply understanding the large-scale conditions of your problem space can also help you understand some of the external forces to which you will be subject. A startup focused on hiring might have its business model depend entirely on unemployment: at low unemployment, it would expect to be able to capture more value from companies eager for labor, and at high unemployment, there might be more value to capture from people looking for work. A startup interested in transportation should care tremendously about energy prices. A startup focused on retail is probably more interested in consumer demand and supply chain health.

While the attribute of Industry Dynamics might seem operational in nature -- and it is -- it is worth understanding before taking the plunge.

Likely Business Attributes

Understanding some of the attributes of a given problem space can significantly help you understand the kind of company you'll need to build before you build it. There are too many of these to fully enumerate them, but here are some of the particularly important ones.

What are you selling?

Here are a few kinds of things a business might sell and some of their implications.

Software startups tend to have extremely high margins (90%+) and scalability, which makes software an attractive option for rapidly scaling a successful startup into unicorn territory. In current conditions, their greatest (sometimes verging on singular!) expense will almost certainly be programmers, other than a tiny fraction of extremely compute-intensive artificial intelligence startups. Still, hiring programmers often proves even more difficult than paying them because the labor market for software engineering is highly competitive. One advantage of software startups is they need not require much capital for hardware and related fixed costs.

For some software startups, the software can be mostly written by the founders in a few weeks, at least at the beginning. On the other hand, for others, the software might be so complicated as to require many millions of dollars and years of work. One downside of selling software is that differentiating from the competition is increasingly difficult in many industries due to the current volume of software startups. A second is that because software is intangible, it can sometimes be more difficult to generate significant willingness to pay than when selling something physical, though this depends tremendously on the software and to whom it is being sold.

Hardware startups tend to have modest margins (30% is a good starting point, though this can vary considerably) and scalability because the production of additional units requires additional resources and infrastructure. Hardware startups tend to require more diverse expertise, too: they might need mechanical engineering, industrial design, supply chain management, electrical engineering, and software engineering. While these difficulties are real, the necessity of building such a diverse skillset can prove a valuable moat against competitors, which can increase the size of the opportunity if the startup succeeds. Furthermore, a hardware platform sometimes provides opportunities to make software and service upsells. For example, growth in Apple's revenue increasingly depends on software services leveraging the wide distribution of their hardware. Apple's growing focus on services is not a show of vulnerability in their hardware business but rather a key strength: their integration of hardware into their ecosystem gives them substantial control of an installed base of one to two trillion dollars worth of consumer electronics.

Service startups have margins and scalability which vary quite dramatically. In general, services startups find it harder to exploit standardization and automation, thereby having higher costs (and lowering demand) and lessened scalability. Despite this, startups that perform services have several key benefits. First, they can be very lightweight in the beginning and adapt rapidly. The lack of a concrete product to build means that many of the associated costs and delays can be avoided, and if the company gains a better understanding of the problem or solution it is also easier to adapt the service to make use of this new information. In fact, many recommend that product companies first validate and refine their ideas by offering their product as a stripped-down service in what is often known as a concierge test. Finally, there are also some problems that right now cannot be wholly solved by a product and require human labor too. Virtually anything for which humans are currently employed is a reasonable candidate.

To whom are you are selling?

If you are selling to consumers (B2C) you will benefit from relatively short sales cycles and rapid growth potential. Additionally, because you will make many more individual sales, there is somewhat more space to collect aggregate data which can help drive future decision-making. However, effectively selling to consumers is also very challenging. Finally, consumers tend to demand a higher level of design polish on their purchases than most businesses. Consequently, if selling to consumers you should expect to learn or hire significant expertise in both design (e.g., HCI) and marketing.

If you are selling to other businesses (B2B) you will benefit from much larger contracts and less emphasis on marketing and advertising in achieving these contracts. The disadvantage is that sales-related costs will be much higher and sales cycles will be far slower. Consumers tend to make most purchasing decisions on a seconds-to-hours basis, reserving only occasional and expensive purchases for more deliberation. By contrast, businesses are more likely to have sales cycles ranging from months to years.

Additionally, just because advertising and social media are somewhat less important does not make selling to businesses any easier: each business has its own complicated decision-making unit (DMU) which dictates purchasing decisions. Business development and sales are no less a critical role in any B2B company than design and marketing are in a B2C. One final and useful correlation is that sales cycles tend to be somewhat inversely correlated to the size of the company. Selling to very large corporations tends to take a while even if the value proposition is very strong because they are complicated institutions that may require many levels of buy-in. Conversely, selling to startups can be much faster, though the sales will of course be smaller.

There are other entities to whom you could sell which might not neatly fit these two categories. Governments (B2G) have entirely different processes and behaviors. A product targeting ultra-high-net-worth individuals might technically be B2C but might behave completely differently. All this is to say: know your customer.


In summarizing the above, we now consider a quick checklist to evaluate whether a problem space indeed qualifies as fertile ground.

  1. How interested are you in these problems? (Rate 0-10)
  2. How good of a fit are you to solve these problems? (Rate 0-10)
  3. Is now the right time to solve these problems? (Rate 0-10)
  4. How promising is the industry, as a whole? (Rate 0-10)

A problem space need not be 10/10 in all regards in order to yield great ideas. But, the more, the better.

Interlude: Being Right

Almost everyone -- including very smart people -- is wrong about almost everything. Even me.

First, the reason for this is not that they do not know anything about anything. People often know many things about many things.

Nor is it, per se, that the things that they know are wrong. Most of what most people know is actually right.

The issue, instead, comes from the things they know not really mattering, and the things they don't know mattering a great deal more. In aggregate, this leaves them a little bit right but mostly wrong.

This problem is very difficult to remedy in the general sense, but it is made easier for us in the realm of entrepreneurship; This is because starting a successful company only requires being really right about a few things. (One needs to do at least B-grade work on many more.) My overall suggestion for how to be really right will sound extreme, but I believe in it wholeheartedly.

To be right, the very first step is to let go of all of your existing presumptions, ideas, and frameworks, and accept that your understanding of the problem is probably wrong even if most of what you know is right. I believe this to be the single most categorically difficult step in the whole essay. Fortunately, it gets easier with practice.

Before I try to convince you that this deeply uncomfortable approach is the right answer, let me first enumerate the two main ways it might be wrong.

Perhaps, when all is said and done you will come back to your original beliefs and complain, fairly, that I have wasted your time. I have never seen this happen. Much more often, you will discover that the things that you thought mattered, didn't, or else mattered in ways very different from how you thought they would. And, there were other things that mattered a great deal more which you had entirely neglected. This was certainly true for my project Prod.

Another, somewhat more common possibility is that it might be possible to become right by iteratively revising your existing frameworks and presumptions with new data as you collect it. If you have a good reason to believe you are very close to the right answer, this is a good approach -- for example, when tuning an already great idea into a unicorn-trajectory one. Yet this process is unlikely to yield the insights which will guide a new, differentiated, successful startup. Proceed with caution.

More often, these sorts of situations won't apply, and then we will need to determine how best to start from scratch. The first and principal pillar will be good and sufficient data.

Gathering Data

We want to be scientific about our approach to entrepreneurship. Logic and analysis certainly have an important role to play in this discussion, and I will cover our perspective on them soon, but to be good scientists, the foundation of our approach must be empirical. One thing I'd like to emphasize is that there is a real -- and perhaps irreversible -- cost to doing analysis and reaching conclusions before acquiring the right data. The first and obvious drawback is that if you collect data later that disagrees, you will have to redo your analysis, and this may take additional time and effort beyond if you had acquired the data in the beginning. And yet, there is an even greater cost.

The true cost is that coming up with partial answers too soon will permanently impair your ability to come to the right answers even once you have all the data. As with a building, our ideas and frameworks about the world acquire real inertia and are not easily dislodged. If our building (collection of ideas and frameworks) is built on an uneven foundation (insufficient, incorrect, or misapplied data), it is no easy task to try to shore up the foundations with the building still standing because the building may have acquired the grade of the slope on which it was built. Nor can we easily tear down the building, because we will be forced to navigate the rubble as we build our new building. Of course, having a single wrong idea about a problem space is probably a reversible process -- the point is not to be afraid to think any thoughts -- but having a large collection of wrong frameworks and ideas about a problem space is not so easily resolved. Better, then, to ground our analysis in truth from the beginning.

How does one ground their analysis in truth? I will divide among the three key types of data founders tend to rely on: primary sources, secondary sources, and personal experiences.

Primary Sources

Of the three types outlined above, if I could only have one, I would (without hesitation) choose primary sources: user interviews, feedback, and field conversations. As written by Raymond Wolfinger, the plural of anecdote is data. Our approach will be similar to his.

There are many flaws of primary sources. Doing research with primary sources is extremely time-intensive, and as a result, your survey will probably be smaller than that of published research, leading to bias from small sample sizes. Coherently sifting through the masses of writing you will produce in an organized way challenges even the experienced. Knowing which users' ideas and feedback to prioritize is a difficult art. You must also learn to use all feedback, not just the feedback that you want.

Despite these challenges, primary sources have key advantages that make them absolutely instrumental in finding your idea. First, primary sources are removed of all of the biases that emerge in the condensing of this vast data for secondary publication: they are raw and unfiltered. Second, primary sources can help capture data that would be hard to understand otherwise. Knowing the specific feeling of a user, evoked through tone as well as words and bolstered by the ability to ask questions and clarifications, is quite possibly the most useful information a founder can receive regarding the needs and wants of that user. Hearing a user rage about the status quo in words not fit for publication is a powerful indication that you are in a promising space.

However, there is great skill required to capture the full value of primary sources. For example, one must be very careful when conducting interviews to not become a salesperson, but instead to remain a neutral listener striving for understanding rather than validation. There is also an art to coaxing out the true wants of the user. Users can neglect to tell you their problems because they believe them to be impossible to solve, or because they think their boss would never pay. As a reporter, it is your job to nonetheless discover these problems -- even if the user turns out to be right -- in pursuit of a complete and holistic understanding of the problem.

Secondary Sources

Secondary sources provide another valuable fount of data. These constitute industry outlooks, investment reports and analyses, government and NGO research, and more. Secondary sources are useful in three main ways. First, because they tend to aggregate much larger pools of data in a condensed way, they can allow you to consume representative perspectives on large problems in a time-efficient way. This in turn can help validate or invalidate business assumptions based on ideas generated in primary market research. For example, you might have sampled a set of influencers and discovered a tremendous need, only for an industry report to demonstrate that the overall market is much smaller than you had realized. A second benefit of secondary sources is that they can tell you the prevailing opinions in a given area. If you're reading a government report, there's a good chance that others who care about your area are too, and as a result, you can gather insight into what the status quo actually is in terms of what others believe. Third, secondary sources provide a bird's-eye perspective on complicated problem spaces. If one were constricted so as to only use primary sources, one might imagine missing the forest for the trees because the people on the ground cannot always see the full picture.

Despite these benefits, secondary sources come with a major cost: the very nature of condensing down a large amount of data into a consumable report means that the author made choices about what to keep, what to cut, and which story to tell. Consuming secondary research biases you towards these narratives, ideas, and frameworks, which makes you less likely to challenge the status quo in a differentiated way. As a result, I would be cautious to read too much before you have learned directly from the source yourself and formed some interesting opinions of your own because the clashing boundary of your data-driven opinions with those of other experts is the most promising place to derive real insight.

Personal Experiences

Personal experiences are among the most important but also unreliable sources of data. The trouble is not that you are wrong about your experiences, but that you may not know the ways in which your experiences themselves are normal and the ways that they are not. For example, building developer tools based on one's own experiences as a software engineer requires extreme care, because software developers operate in very heterogeneous ways: different languages, editors, version control, targets, customers, management styles, and more. As a result, even if you find a particular feature of some product to be highly irritating, it can be hard to know if others feel the same way. My suggestion would be to lay out your experience-based beliefs and try to classify them as truly representative, modestly representative, or unrepresentative based on data corroborated by other sources. Those which you feel are representative you should keep and even encourage -- building a product for yourself has many benefits over building it for others, not least because it is easier to rapidly achieve both function and polish. Inversely, those experiences that you feel are not representative you should try to purge from your consideration and instead refer to more reliable sources. A functioning, polished, and wrong product is still the wrong product. As with other aspects of ideation, there is an element of luck to this. If you are lucky, all of your experiences and beliefs will turn out to be right and useful. If you are unlucky you will find that you will need to exclusively draw on data from elsewhere. Fortunately, one can do very well even with just that.

On Action

Usually, the best, differentiated data is the data collected in creative and active ways. For example, one approach to gauging consumer interest in an idea -- detailed earlier -- is to find a bunch of consumers and ask them. Yet this approach has many challenges. Finding representative users is time-consuming. Asking their opinions in an unbiased way requires care. And in any case, the sample sizes are unlikely to be very big.

Sometimes, a better approach is to just build a mock-up of your proposed idea and see who bites. For example, one could build a website that looks nice, describes what the product will do, and then both publicize it on appropriate subreddits and also spend $50 advertising it on Facebook, which will buy thousands of impressions.

If your pitch is good, you'll end up with a few hundred people on your waitlist. The benefits of this are several-fold. First, if you want to find very qualified leads for user research, you now have them. Indeed, they will most probably be excited to give you their feedback and ideas for the product. Second, these users are great leads to be your first customers once you build the product. Finally, you'll have also gotten a head start on designing, building, and marketing your product if you follow through.

On the flip side, if there is very little interest in your mockup, then you know fairly quickly that one of your assumptions was wrong, and you can immediately begin to figure out which and why.

Finally, one major advantage of this approach is that the only real bottleneck is you. If you're reaching out to talk to people, you have to adapt to their schedule, and they may not be able to speak for a week or two -- an eternity in entrepreneur time. By contrast, it may be possible to build out a mockup and market it within a day or two or three.

To conclude, in gathering your data, always bias towards action and away from bottlenecks. The more you only need to rely on yourself to make progress, the better. And, the more scalable your approach to generating high-quality data, the better.

Final Thoughts on Data

Just as it applies in computer science, so does the principle of Garbage In, Garbage Out apply in entrepreneurship and ideation. Thus the able entrepreneur is discerning in the sources to which they pay attention, so as to avoid coming to misleading conclusions and throwing good effort after bad. However, this alone does not constitute a silver bullet in developing a sophisticated understanding of a problem space. If anything, the adjective sophisticated connotes not just having a few correct ideas but also knowing many details and the interplay between them. Unfortunately, getting that much information takes time, as the maximum rate at which new information can be downloaded into our brains' long-term storage is probably fairly small. As a result, collecting data in the right ways is necessary but not sufficient.

You also simply need to put in the work to collect lots of data. My own hunch is that the average founder probably collects about an order of magnitude too little data, basing their company off of ~10 user interviews rather than ~100. In other words, there are no true shortcuts to developing this sophisticated understanding. All we can do is avoid the detours so often accidentally taken.

This leads us naturally to the second pillar of finding great ideas: great thinking.

The Graph of Knowledge and First-Principles Thinking

Within Prod, we put a great deal of emphasis on so-called first-principles thinking, which we believe represents both an ideal and a practice to which virtually everyone claims to ascribe but few do well. This naturally leads to the question: what is first-principles thinking? This question is made somewhat more difficult by the varying choices of definitions that exist. We will produce our own definitions which are at least precise, consistent, and reasonably intuitive, so that we may better understand both what it is and how one might actually put it into practice.

At its core, first-principles thinking is the method of reasoning upwards from deeper truths. The world is made out of all sorts of behaviors and phenomena, of which some seem to be fundamental and others emergent, and others still emergent from emergent phenomena (and so on). Which phenomena are most fundamental is a question for metaphysicists, but for our purposes, the deepest will be physics, mathematics, and logic. From these simple starting points emerge chemistry, biology, psychology, economics, and business. Then, the purest first-principles thinking would be the re-derivation of everything about a problem from just physics, mathematics, and logic, in pursuit of both better understanding and better solutions.

This level of purity is, to say the least, infeasible. Not only is it just too complicated to rederive economics from the Schroedinger equation and a careful analysis of the cosmic microwave background, but there is too much in the middle that still isn't very well understood. So, we will need to settle for a lower level of purity instead, and in return, we will get a much more efficient algorithm. To begin this, let us consider the graph of knowledge.

Imagine, for the sake of discussion, the set of all ideas which describe the world reasonably well. These ideas might range from the straightforward and physical, like water is denser than ice, to the somewhat subjective and psychological -- perhaps pediatricians are generally happy with their careers. We can then start to link the ideas in this set together with edges. Some edges will be directed and causal: the second, at least in part, follows from the first. Other edges are undirected, representing some commonality, difference, or analogy between ideas; these edges represent an unordered relation.

When we think about an idea to better understand it, we are usually exploring some local neighborhood around the idea. We draw in other ideas we associate with the main one, and perhaps we further associate others with those neighbors, in the hope that one of these related ideas will answer a question (often implicit) about the idea. For example: Why is this idea true? Under what circumstances? What would need to change for it to not be true? What other ideas depend on it? Etc. I will even suggest that the quality and density of one's internal index of this local neighborhood is a pretty good starting point for how we might formalize our earlier notion of a sophisticated understanding.

The assumption of first-principles thinking is that some kinds of edges and relations are more useful than others, and that causal ones are especially good. Furthermore, some relations are better than others: as a trivial example, a direct causal relationship is, under almost all circumstances, more useful than the relationship that two ideas rhyme when stated in English. Then, first-principles thinking is really a strong bias towards a deep exploration of causal relationships between ideas. Nor is this arbitrary. The preference for depth comes from the fact that not all edges are equally robust. Usually, ideas relate very strongly to just a few other ideas, somewhat strongly to somewhat more ideas, and very weakly to many more after that. As a result, if we want to best understand an idea, we need to prioritize understanding the ideas which relate most strongly -- causally -- which in turn requires understanding their own dependencies, and so on. In other words, under the prior that the strengths of relationships are a highly skewed distribution, a preference for deep analysis emerges as a natural result of optimizing understanding for a given amount of time.

These sorts of deep analyses have two major benefits and one major drawback. The first advantage is that first-principles thinking is less likely to mislead you than other kinds of thinking which rely on analogy since the causal links of first-principles thinking are usually more robust than links of analogy. The second benefit is that because these links are more robust, we can safely access ideas further away from the one we are most interested in, which makes us more likely to stumble upon novel and creative ideas. After all, if one is only exploring a very small local region around an idea, it becomes very likely that many others have also considered these ideas, which makes it correspondingly less likely that your own ideas will be differentiated and better. However, as you can expand to see more of the neighborhood of ideas, it becomes increasingly likely that you will consider ideas that nobody has ever thought of before in their analysis.

Yet despite my genuine belief in the value of these causal analyses, entirely limiting oneself to them would be similarly unwise. The graph of knowledge is much richer and denser than the subset which considers only the causal links, and while many paths through it are mysterious, shadowed, and even misleading, our associative abilities remain one of our brains' core strengths. We should utilize them well, too, because the right analogies can also quickly take us to far away parts of the graph of knowledge, similarly helping us find new ideas not considered by others. The game is dangerous -- we might end up in territory truly irrelevant -- but the payoffs can be great, too.

In the future, one could imagine an analysis of the graph of knowledge much richer than the one I have put forth here. In it, we would consider a taxonomy of the many kinds of edges and when they are useful. Perhaps in some parts of the graph of knowledge, certain kinds of analogies are especially useful but in others, it is only the causal links that are worth considering. I would imagine that when writing songs or poems, the rhyming relation I had dismissed earlier becomes especially useful even though it has little value for the budding founder.

Taken together, we can be just as strategic in how we think about a problem and explore its nooks as we can be when choosing a problem of interest or gathering data about it. In this section, we have seen how a preference for venturing deep down the causal relationships underlying an idea in the graph of knowledge makes the able entrepreneur more likely to find creative and useful new ideas too. At the same time, we need not completely neglect our ability to reason by analogy, which requires great care and intuition to do well but can proportionately accelerate our creativity.

While I do believe the above theory on how to think is important, of even greater importance is their practice. Theory is nothing without execution and adherence trumps optimality.

Ideation Strategy

There are a few separate challenges of ideation strategy which I will address in this section.

How does one know if an idea is good enough?

This question -- how do we know if the idea is excellent? -- is pivotal. In a sense, it's a catch-all proxy for what we actually care about in the long-term: of whether and to what degree a startup based on the idea will turn out to be successful. It's also intractable. No founder, investor, or algorithm has yet been able to convincingly and consistently foresee the future. Thus, to know whether or not an idea really is good enough requires a tremendous investment of labor, capital, and perhaps even reputation to find out. This is not terribly encouraging, and I have no real solutions to this problem. All I have are two workarounds and one band-aid.

One workaround is to make a very large effort to disprove an idea before actually pursuing it, because this is easier than comprehensively investigating whether an idea is good. Negative results for an idea are easier and generally more reliable than positive ones. Consequently, the faster you can eliminate ideas and cease effort, the better. Do be aware that great ideas often hide close to terrible ones, so you need to be targeted in what you eliminate, so as to not cause too much collateral damage. Just because one particular approach does not work does not mean that there are no others.

Another workaround comes from the fact that the processes by which great ideas are generated also tend to yield larger numbers of merely good ideas close by. As a result, one good heuristic approach is to not take the first seemingly good enough idea that your process has generated, but instead wait for a few more to come by as well. (The mathematics underpinning this particular approach is related to that of the well-known secretary problem.)

A final band-aid is to be Bayesian: while pursuing an idea, be constantly identifying the key remaining assumptions which will eventually determine whether the idea is great or not, and then make a continual and concentrated effort to gather more information on these areas. Find clever, fast, and lightweight ways to prove your views right or wrong. If the idea is found not to be great, cease work. After all, you've already lost upon choosing the lesser of two evils, for there are usually good answers if you expand your scope and think harder.

We have also come to prefer that founders take a long time to circle before landing, rather than stumbling around later. Experience within Prod's cohort suggests that once companies find their way, they can make astonishing progress astonishingly rapidly. By contrast, the fastest way to burn time and money is to work on something mediocre.

Increasing SNR

Ideation often feels cacophonous because there is too little signal and too much noise, and so the signal-to-noise ratio (SNR) is too low to make meaningful inferences. I argue that this is due primarily to two key obstacles: the cost of gathering data, and the randomness of the results.

First, many of the actions of ideation are difficult, costly, or time-consuming, which makes it difficult to collect much data. For example, talking to a potential user might be easy -- such as asking a friend for feedback -- but data acquired that way is likely to be skewed by all sorts of biases. More often, talking to a plausible user requires cold outreach, scheduling, and then at least fifteen minutes to talk and some more time to digest. Building a prototype to put in front of users might take weeks of work. Consequently, founders are limited in the number of these actions they can take -- and therefore in the amount of data they can collect and opportunities for inspiration they will receive.

Second, the process of a good idea coming to you from a given action is random, complicated, and relatively unlikely, and as a result, it can be very difficult to know whether that action was productive or not. In other words, the reward signal is very weak and assigning credit is very hard. Knowledge that might not have seemed important can turn out to be vital in the ideation process, and vice-versa. If your first ten user interviews have yielded truly nothing of interest, perhaps user interviews are the wrong way to approach the idea, or perhaps you're conducting them badly -- or perhaps you just got unlucky!

In this section, we'll explore five methods to increase the signal and reduce the noise.

Answer 1: Just Do More

In order to improve your ability to improve your strategy, the first approach I will suggest is the Just Do More method: simply work harder at ideation. This might seem obvious, but I mean it in a specific way: working harder is the only common way to genuinely surmount the twin obstacles of cost and noise. Regarding the first obstacle, more effort yields more information about what works and what doesn't, which helps you refine your strategy. Regarding the second obstacle, this sort of problem is well-studied in statistics, and while there are intelligent methods to better discover what's really going on, the simplest way to increase your statistical power is to just do more!

A common mistake that founders make is to think that their investigations should follow normal product development cycles -- perhaps two months to build the product, and one month to test it, before iterating. This is entirely wrong. Your probability of eventual success is closely related to the number of shots on goal you take. Instead ask: how can I test my assumptions in the next three days? In this way, you can try thirty times as many ideas as a founder with lower standards. This yields a far greater likelihood of finding a great one, and much faster learning along the way.

To reiterate a core emphasis of this essay: stumbling onto a great idea is, by definition, easy, in the same way that winning the lottery is easy. But inventing a great idea is tremendous work and, if successful, tremendously rewarding work. These two aspects are inseparable.

Answer 2: Reduce Costs

You can always aspire to be a good deal more thoughtful and clever about how you go about the various motions of ideation to make it a truly efficient process. While efficiency is relevant in many domains, like money, time is the key resource of ideation and should be treated as such. I think this is best illustrated by several examples. (They also illustrate that the magnitude of the gains can vary dramatically.)

[Scenario 1]: you've spent tons of time carefully crafting cold outreach on LinkedIn, but are getting few responses, and are having trouble managing them all to schedule time to chat. You're thinking of doubling down and just grinding out tons more. Don't. Here's what you might do instead.

  1. Immediately message your friend who is best at this and ask for help. (You might be surprised how broadly effective this strategy is across the board!)
  2. First, determine the problem. Are the messages not good enough? Or are they just not being read? You do a quick google and realize to stop doing outreach directly on LinkedIn -- too much is spam and thus few read it. But now you need some other way to get in contact! Solution: just pony up and subscribe to a service such as Apollo to grab personal and professional emails going forward.
  3. You also decide you're spending far too much time on each individual message, personalizing it. You revise your approach to have a common template with just one job title and a few adjectives missing, which you can fill in on-the-fly. Or, if you're feeling really scalable, you could use a service like GMass.
  4. You should make sure you're sending from your institutional email, which will almost certainly have higher response rates.
  5. You should subscribe to a service like Calendly or Cron and simply link to it in your template. That way, no more back-and-forth for scheduling.
  6. Try to send it to people more like the ones that positively responded to you. Try to find a common pattern of job title, demographics, education, or similar.
  7. If you want the structure, add a CRM too!

While this scenario is made up, I think it is plausible that this could go from 10 minutes per message plus 10 minutes of scheduling overhead to 1 minute per email plus 1 minute of scheduling overhead, while increasing response rates from 15% to 40%. A factor of 25x in productivity is nothing to sneeze at, so it is worth tuning these systems carefully before grinding away.

[Scenario 2]: You are trying to understand a new technical area relevant to your problem interests, but are having trouble establishing the necessary background to effectively read and learn more. You also feel unmotivated each time which makes it harder to put in the work. What to do?

  1. First, find someone who knows more about the space than you and ask how they learned. If possible, get them to recount how difficult it was for them, too. Read what they recommend.
  2. Block out consistent time to do just this. Turn off all alarms and notifications.
  3. If necessary, print on paper and do not bring electronics. The best learning is undistracted.
  4. Immediately start scheduling meetings with experts whose opinion of you matters to you. There are three reasons to do this.
    1. Pipelining the process of scheduling means you'll be able to learn something more advanced right as soon as you're ready, instead of having to wait to schedule once you actually are ready.
    2. It will motivate you to work harder and learn quickly in preparation for these meetings.
    3. In the event that you're still really stuck even by then, you might embarrass yourself but you'll be glad to have someone to explain what's what at that point.

This process is subjective and individual, so you alone know what would help you most, but the above bullets would probably help most people.

[Scenario 3]: You think there might be demand for your idea -- lots of positive feedback from interviews -- but you're not sure. You're thinking of building an MVP to validate demand.

  1. Before you start building, quickly make a mock-up website that describes the product and what it will do. (And hopefully looks and feels polished.) Send it to any relevant lists -- people you've already talked to for interviews, strangers on LinkedIn, dormspam, whatever. Provide a box to leave an email for the waitlist.
  2. Run advertisements too, if you can afford to blow $50. Measure effectiveness.
  3. Also just run Google Analytics.
  4. Then build.
  5. This way, as soon as the MVP is ready, you don't need to spend a bunch more time figuring out how to sell it, because you've successfully pipelined the user acquisition.

One can also be infinitely more clever for each of these problems, but I leave that as an exercise to the reader (and a good one at that!)

Answer 3: Create Proxy Objectives

Another approach is to create your own metrics to judge your success along the way in ideation. This approach can be very helpful by grounding you in something more predictable and measurable than the difficult-to-evaluate quality of your idea. But, it is also necessary that your objective be meaningfully related to the eventual success of your startup. As a result, the proper choice of proxy objective will be very dependent on the sort of startup you intend to build. Nonetheless, here are a few examples to get you started.

A surprisingly good proxy objective is: Do I know much more about this space than everyone else I talk to? In other words, if your efforts are incidentally giving you a truly sophisticated understanding of the area of interest, then you're probably on the right track.

Another proxy objective worth considering is how hard you find yourself working. If you're really excited and things are going well, you're likely to work better and harder. This virtuous cycle is worth attention and encouragement. Tracking this metric is also broadly useful, too.

A third proxy objective is that of engagement. While the precise application of this is very area-dependent -- engagement means something very different for consumer software versus deep tech -- the question is how excited can you get relevant people to be about your work. If your ideas are very good, there is a good chance that you can get others to be consistently excited about them. It also forms a reasonably objective metric of your ability to build momentum by selling your vision -- be it in selling, hiring, or raising.

There are plenty other proxy objectives one might consider, too. You may want to come up with several, track them, and ask yourself each week, day, and moment how you can follow their collective gradient upwards.

Answer 4: Solve the Exploration-Exploitation Tradeoff

The kernel idea underpinning this answer is to reframe the principal question of this essay (that is, how to find a good start-up idea) in slightly more quotidian terms. After all, nobody has infinite time to ideate their startup, so the question of finding a great idea as fast as possible is somewhat artificial. Perhaps we should instead ask how to maximize the probability that we find a great idea within a specified timeframe.

Most of the time, these two questions lead to the same answers, and we usually focus on the first because it's simpler. Occasionally, the difference is important, and one such place is in considering how your strategy should change with time.

So, at the beginning you should focus more on exploring better ideation processes. By the end of your window, you should be simply looking for ideas and not worry too much about new processes. I hope this sounds obvious when stated in these terms. Still, it is common for founders to waste lots of time finding mediocre ideas with mediocre processes before realizing they need to up their game.

Nor should one presume that investing in better ideation processes is divorced from doing ideation. The only way to find these better processes is to simultaneously ideate, reflect, and improve.

Answer 5: Estimate Uncertainty

In addition to gathering more data, we can also be intelligent in understanding how much confidence we should have in our beliefs. As a simple example, if we ask ten users' opinions and they all say exactly the same thing, we can have pretty high confidence that other such users will have similar opinions. By contrast, if all ten users have completely different opinions, we intuitively understand that we would probably need to gather more data to figure out what's what.

Heuristic Transformations

One final, interesting, and (I believe) novel approach to generating great startup ideas is to start with okay ideas and stochastically transform them into better ones. This section is concerned with a few approaches to generate these better ideas. Because they are only heuristics, they often won't work -- in fact, they may not even be applicable to your idea. Yet these approaches are very fast and easy to try and have a reasonable chance of providing some benefit, which makes them worth discussing.

Transformation 1: One Step Earlier

One common reason why startup ideas turn out not to be good is that they try to do too much at once, which makes them infeasible to scale. Often, a better idea is to first figure out the hardest part of building the startup and then build that startup instead.

For example, imagine a startup that depends on signing up individual restaurants for its more cost-effective online advertising. A key challenge with this startup is that this is a very expensive process for relatively little revenue per startup, and it takes a long time to get each individual restaurant up to speed. In fact, you might imagine that building out online advertising is pretty easy, and the startup spends most of its time building better processes to sign up restaurants -- both expensive and daunting.

A good approach is to lean in. Rather than selling advertising services themselves, this startup should be building the common interface into restaurants for other startups. That way, it can focus all its efforts on this hard part -- and reap the benefit of opening up an industry much larger than it had originally planned.

Often, this type of pivot happens by trying to build the startup, failing, and tacking to this new idea instead. Indeed, the benefit is you learn more about what actually is hard, but it takes much more time and money. Instead, by considering this common transformation as operating on the idea instead of the company, you can often achieve the same results earlier.

Transformation 2: The Only Thing that Matters

Another, less dramatic transformation is the act of distillation. Some startup ideas suffer from being unfocused. Specifically, they have a fairly clear idea of how they want the world to look at the end, but to get there they want to do several things at once.

This causes two main problems. First, doing one thing well is hard enough. Great companies can eventually do many things well, but only because they build up over time. Second, trying to do more than one thing muddles the message, which makes all aspects of the business harder. It becomes harder to explain to prospective customers just what the company actually does. It becomes harder to convince new hires to be part of your vision. And, it becomes harder to sell investors on that vision.

A red flag is when your elevator pitch keeps trying to escape its time bounds. I experienced this personally while building Prod. In fact, I almost refused to create a short pitch because I felt it left too much out. And I can indeed talk about the philosophy and experiences underpinning Prod for hours on end, and I think they do make the pitch stronger. But being unable to craft a short pitch was not a sign that my understanding was so deep as to transcend these bounds. Instead, it represented the fact that although I knew many details, I didn't know what was really important. Learning that took longer.

So, try imagining your company if it could only do one thing. If the value proposition seems too small, then either modify the idea so that it is larger, or, completely failing, add one more thing back. But remember: each idea you add back costs ten seconds of work, now, and six months of work, later. By starting with the absolute minimum and adding complexity back with great caution, one more easily constructs a feasible company and a compelling story.

Transformation 3: Concierge-ify

Another good transformation is to consider how you would build your company if the hardest technical part of the company simply turned out to be impossible. In all likelihood, you would need to replace that part with humans. In other words, it would make the startup look a bit more like a concierge version of itself.

This may sound like a bad thing. It is, in fact, often a very good thing: many valuable platform companies are simply the concierge versions of idealized companies. For example, an ideal transportation company might involve a fleet of self-driving cars roaming the streets. But building self-driving cars is a long, difficult, capital-intensive effort. The concierge version is Uber, which would still be a great company even if self-driving cars never come to pass.

Concierge testing of startups is an idea already long discussed -- for example, in Bill Aulet's fabulous Disciplined Entrepreneurship. This transformation goes one step further, asserting that concierge startups are often better, too, by applying at even greater scale the advantages of concierge testing.

First, concierge startups can build and iterate faster than normal startups. Whereas a normal startup would need to build out new systems, software, and structures in order to change its approach, people-glue is far more malleable, letting you build the easy parts and pay for the hard parts -- at least to begin.

Second, concierge startups are able to aggregate demand in winner-take-all industries before their slower-moving competitors, giving them both revenue and also access to the investments which will allow them to eventually realize their idealized selves. This increases both their short and long-term value.

Finally, this speed allows concierge startups to access certain flywheels that can help lead to future success. For example, Tesla is able to use data from the sensors it designs and provides (but the customer pays for) to train its own self-driving functionality. If Tesla had tried to come at self-driving cars head-on, it would have had to pay for and maintain its cars, design and build its own sensor suite, and then pay people to drive the cars to generate data. Instead, Tesla's data flywheel is a natural extension of its people-centric business.

So, consider turning the idea for your startup -- partially -- into a concierge version of itself. You may find that the resulting idea is not only easier but better too.

Transformation 4: You First, and You Alone

The best startups occur when there is both product-market fit and product-founder(s) fit. While it is common to spend most of our ideation efforts on the former, the latter is important too!

So, try to imagine how your idea would change if you had to build the company by yourself. How can you make it so it is as easy and fun for you as possible, and plays to your strengths?

There are two good reasons for this transformation. First, as we had discussed earlier, both your personal interest in and alignment with the startup you intend to build are important criteria for success. Yet these aspects of ourselves are hard to change predictably and quickly. Therefore, it makes more sense to project our ideas onto ourselves rather than ourselves onto our idea.

Second, by making the idea easier for you to execute, you help yourself at the point where you need it most: the beginning. If the founding team can find phenomenal traction before hiring, the company will be on far better footing to continue to sell, hire, and raise into the future.

Of course, if one goes too far with this, one might badly warp an idea with otherwise great product-market fit, but small steps in this direction are often worthwhile.

Final Thoughts on Heuristic Transformations

These transformations, despite their simplicity, can be deceptively powerful. A good example of them in action is the story of how Scale AI found its idea.

Scale's original idea was to make it easier for people to book doctor's appointments online. Anyone who has tried to book a doctor's appointment knows it can be a tremendous pain: calling, being put on hold, explaining insurance, compromising on schedule, and more. Scale wanted to make this easier and automate much of the process. However, they quickly discovered that interfacing with hospitals' outdated information systems would be a tremendous barrier.

So, they revised their idea to put people on the back-end to interface to hospitals. Now, the idea was that the user would go to Scale's website and enter their insurance information, availability, location, and whatever other information was needed. Then, Scale would have an agent take this information, call the hospital, and book the appointment. Once this was done, the agent would fill in the details and Scale would email them to the user. Yet they discovered that hiring contractors to do this was not as easy as they had imagined.

So, their following iteration focused on that problem instead. Scale was going to become the API for all human labor. If you have a simple task that needs to be done by a human, you could write it into their API, and they'd have their contractors on the back-end go and do it for you. They suggested a half-dozen applications of their API, the last of which was labeling data to train machine learning models.

Eventually, Scale discovered that AI data labeling was tremendously in-demand and all of their other applications didn't really matter, and they distilled their company into the AI data labeling company. That idea turned out to be a unicorn.

One can trace their iteration directly through these transformations. They started with, frankly, a pretty mediocre idea. Next, they concierge-ified it. Then, they built the company which was one step earlier. Finally, they distilled even that to its most profitable element. Their start and end ideas were wildly different, but their path through maps surprisingly cleanly onto these transformations laid out above:

  1. Concierge-ify
  2. One Step Earlier
  3. Distill
At each stage, Scale's idea became monotonically better.

Of course, this anecdote is not meant to suggest that these heuristics are guaranteed to convert a terrible idea into an excellent one. Rather, it serves as an example to build intuition. These heuristics are probabilistic at best, and their successful application also requires art. Still, they provide concrete starting directions when trying to improve an idea.

Part 2: Being

The Entrepreneurial Spirit

All of the above writing considers how to actively pursue ideation -- to make it your principal work. Yet to suggest that this is the full story would be... misleading. In Zen and the Art of Motorcycle Maintenance, Robert Pirsig writes,

You want to know how to paint a perfect painting? It's easy.
Make yourself perfect and then just paint naturally.

Analogously, the easy way to find a perfect startup idea is to first become a perfect entrepreneur.

Unfortunately, I don't know how to become the perfect entrepreneur. But, I now at least understand some of the qualities which make for founders who find great ideas. And, while some of us might have a more natural affinity for some of these qualities, I believe all of them can be learned.

Freedom of Thought

Great founders completely compartmentalize the status quo, meaning they understand the world as it is but can prevent it from intruding on their ability to create a vision. This is easy in theory but extremely hard in practice.

Assumptions, explicitly stated, are straightforward to question. The algorithm is as follows.

  1. Consider the world with the assumption as stated.
  2. Consider the world with the assumption rejected.
  3. See which you like better.

Unfortunately, this algorithm is generally insufficient for effective startup ideation. The reason is that it still implicitly predicates the analysis on the rest of the status quo. And, by challenging only one or two things at a time, one is very unlikely to find ideas that have not already been considered.

To illustrate why this algorithm doesn't work, let's take an example: perhaps you want to reinvent the car. So (1), I'll suggest the assumption that cars are driven with a steering wheel, and I hope (2) a flood of alternatives will come to your mind. Maybe it should be a joystick! Or perhaps it should just track your gaze. Or, technology permitting, you should just tell it where you want to go, and the car will go there. Some of these ideas might be worse than the status quo, and might be better (3). Yet none could even plausibly be differentiated. Given the trillions of dollars spent on cars over the last century, it is certain that all of these ideas -- along with the variants of a thousand other assumptions ranging from economics to materials to design and more -- have already been well-considered.

So to find truly innovative ideas, we need to challenge many assumptions simultaneously. To achieve this, we need to go deeper. So we take a step back. What is a car? Why do people actually buy them? One possible answer is that car owners want to transport both people and things cheaply and with low friction. So we should attack this want directly, rather than restricting ourselves to start with the car as-is. Maybe we can partially solve this problem by reducing the amount of transportation that car buyers need to do by delivering or transporting goods as a service on demand. (Such a premise might yield diverse companies ranging from Instacart to a school bus service.) Or maybe we can bypass the problem entirely, at least in cities, by better coordinating public transportation: maybe an app that will automatically rent a scooter and have it waiting when you get off the subway to finish your journey. Though we began considering how to build a better car, we found completely different ideas which both get closer to the core of the problem and are more inventive.

Great founders learn to do this intuitively. Rather than having to stop and think every time, they eventually become adept at getting to the core of problems on-the-fly and by default.

Bias Toward Action

Great founders bias towards doing. Of course, sometimes they sit and think; this is a good thing. But they don't only sit and think. They sit and think about what to do, and then they go do it.

Great founders find delays excruciating. They prefer strategies that incur no delays. And, since some delays are unavoidable, they carefully pipeline their projects to prevent blockages.

This quality is quite difficult to adopt because it requires truly changing your behavior. So, let's consider some examples:

  1. The inactive founder spends a week reading papers. The active founder spends a week building and launching an MVP.
  2. The inactive founder is dragged along by external pressures, staying just ahead of failure. The active founder barrels ahead of deadlines and then makes up new ones, too, to stay busy.
  3. The inactive founder spends a day discussing their ideas with their friends. The active founder spends that day discussing their ideas with their customers.
  4. The inactive founder finds a reason why they couldn't make more progress. The active founder stays up late and gets it done.
  5. The inactive founder laments that they missed an ongoing trade show. The active founder buys a plane ticket at 2 am and shows up the next morning.
  6. The inactive founder carves out time in their day to work on their company. The active founder is also working on their company whenever they have five minutes free.

TThis quality is most powerful when it is combined with freedom of thought. Founders who not only do something, but who do something creative, counterintuitive, and scalable, are likely to find success. The feeling of doing this for the first time is worth briefly characterizing: it usually ranges from am I really allowed to do this? to what if somebody I know sees me?! Double-check your analysis, ensure what you are planning is legal and ethical, take a deep breath, count to ten, and do it. Fortunately, it gets easier with practice.


A common backdrop to tremendous impact is tremendous drive. In complex disciplines -- and I think there are few things more complicated than starting an excellent company -- even tremendously smart people cannot always intuit their way to success. Exceptional results are driven by exceptional ability, and exceptional ability requires practice and attention.

Such focus is rarely achieved when one is simultaneously distracted by other pursuits. The best founders are entirely committed. They go to bed thinking about their companies and wake up thinking about them, too.


Great founders have the highest standards. They know the bar for great work, and that most work is mediocre at best. To some degree, knowing great work requires great taste, and not everyone has great taste. But the barrier is more often delusion or over-reliance on bad measures. Just because one could in school get an A by submitting last-minute writing does not really mean that their writing was great. Yet young founders similarly often assume that such last-minute work will suffice for their companies. It does not. Furthermore, it deprives them of the ability to learn to do great work. The path to competence requires striving.

Hunger and Persistence

Great founders have a need to succeed which verges on pathological. They are strategic, but they also know the value of a dollar. They want the customer more than any of their competitors.

In this aim, they are relentless. They only give up when they have proven that what they want is impossible. Setbacks actually increase their determination, yet still they keep their emotions separate from their analysis.


Just as great founders compartmentalize the status quo of the world, so do they compartmentalize their own status quo. They don't get attached to their plans or ideas. They constantly ask themselves if what they are doing is the very best thing to be doing. And if not, they immediately stop and do something else.

There is one necessary word of caution. Because switching usually incurs costs, one must be careful not to be too optimistic about the unknown. Overeager application of the principle of flexibility undermines the principle of persistence.

Bias Towards Yes

Great founders are appreciably more optimistic than the average person, to the point that they are often wrong. What follows is a brief theoretical argument to describe why this is actually a good thing in early-stage companies.

First, because markets are pretty efficient, most ideas are pretty lousy. Suppose 99% are not worth pursuing. Now imagine that we are trying to classify ideas as good or bad. The simple algorithm labeling all ideas as bad will have 99% accuracy, which seems pretty good! Yet this algorithm will never add value. By contrast, the algorithm labeling all ideas as good will have 1% accuracy, which feels terrible, and yet will add incalculably more value. Thus, founders who want to create value should be more like the optimistic algorithm than the pessimistic one.

Of course, we can do better than issuing a blanket yes or no. But in entrepreneurship, one should not confuse accuracy for value. As we saw in the introduction, successful startups come when the founder is both optimistic and also correct. Neither is a substitute for the other.


One of the important mechanisms by which great founders find the best ideas is that they work with the best people. They therefore make sure that the best people will want to work with them.

Not all great founders need to be charismatic leaders. Not all great founders need to be technically gifted. Not all great founders need to be clever strategists. But all great founders must conduct themselves with honor and integrity. When they make mistakes, they correct themselves quickly and learn.

I believe that fair behavior, beyond its deontological justifications, is also simply in the economic interests of most founders. Founders shouldn't be too greedy about their own share of the pie; as rewards in entrepreneurship are exponentially distributed, it is far better to have a modest slice of a large and growing pie than to wholly own a small and decaying one.


We began this essay by contemplating the enormous value and difficulty of finding great ideas. Finding great ideas makes one's work more fun, more valuable, and more likely to succeed. Yet the process is random and ideas are elusive. Despite these challenges, diligent work, strategically applied, has a good chance of yielding success.

In part one, we considered concrete actions and metrics for startup ideation. First, we enumerated five criteria and considerations for productive problem spaces: interest, alignment, timing, industry dynamics, and business attributes. Second, we emphasized the intense, deliberate, and creative acquisition of data, stressing primary market research but also making careful use of secondary research and personal experiences. Third, we explored the theoretical underpinnings of first-principles thinking through the graph of all knowledge. Fourth, we contemplated ideation strategy: how to know if your idea is good enough, and how to get better at ideation over time. For the former, we suggested focusing on disproving ideas instead of proving them, building a zoo of good ideas before choosing the best, and being Bayesian. For the latter, we presented five solutions: working harder, reducing costs, creating proxy objectives, solving the exploration-exploitation tradeoff, and estimating uncertainty. Fifth and finally, we introduced the heuristic idea transformations of building the preceding company, the distilled company, the concierge company, and the self-reliant company.

In part two, we considered eight qualities of great ideators. First, they don't just challenge the status quo; they completely compartmentalize it. Second, they are always in motion. Third, they focus completely on their work. Fourth, they have the highest possible standards. Fifth, they are hungry for the customer and are dogged in their pursuit. Sixth, they are flexible, and learn to ignore sunk costs. Seventh, they are optimistic, because they understand that their role is to befriend the new. Eighth, they operate with honor and integrity. While these eight attributes do not provide direct processes, they serve as a north star for one's progress as a founder and ideator.

Finally, although this essay emphasizes ideation in the context of high-growth startups, many of the principles and techniques outlined apply to other kinds of creative work. Indeed, the entrepreneurial spirit -- which applies more broadly than startups alone -- cannot be learned overnight, and it can only be learned by doing. Go practice!


I'd like to start by thanking Bill Aulet and Richard Dahan, who have both been instrumental in teaching me new perspectives on entrepreneurship. I see this essay principally as a fusion of what I've learned from the two of them.

Finally, I'd also like to thank Kevin Jiang for his insightful feedback and good ideas, Alfred Spector for his wisdom in both thought and writing, and my friend Oskar for his help in revisions.