Uncertainty

Startups and Uncertainty

I’m trying to write something longer than a blog post and to keep myself motivated I’m breaking it up into smaller pieces and publishing them here. So this should be read in conjunction with Schumpeter on Strategy and A Taxonomy of Moats. I expect that at some point the three will be combined together, I’ll take out about half the words, and it will be called something like ‘The Necessity of Uncertainty’ as chapter 2 of some larger thing. Which is just to say why the first paragraph here is the last paragraph of Taxonomy.


Uncertainty can be seen everywhere in the startup process: in the people, in the technology, in the product, and in the market. This analysis shows something more interesting though: uncertainty is not just a nuisance startup founders can’t avoid, it is an integral part of what allows startups to be successful. Startups that aim to create value can’t have a moat when they begin, uncertainty is what protects them from competition until a proper moat can be built. Uncertainty becomes their moat.

It is 1976. A 20-year old Steve Jobs is pitching his startup, Apple Computers, to Mike Markkula, who would become Apple Computer’s first investor and its new CEO. Markkula, excited by Jobs’ demo of the new Apple II says “We’re going to be a Fortune 500 company in two years.”1 The company had only about $200,000 in revenues, having sold about 175 computers, mainly to hobbyists, up to that point in its existence.2 The 500th largest company on the Fortune 500 in 1976 had just shy of $300 million in revenues.3 Becoming larger than them in two years is a bold prediction.

The mid-1970s were the beginning of the personal computer era. Apple and a few other companies—Tandy Radio Shack, Commodore Inc.—dominated PC sales from 1976 until the beginning of the 1980s. But the 1970s were hardly the beginning of the computer era. IBM had already been one of the ten largest companies in the United States for 20 years based on its computer sales. And while IBM dominated sales, there was a group of strong competitors in the mainframe business (called by the press “the seven dwarfs”; IBM was Snow White.) Minicomputers had also taken a large share of the computer market in the previous ten years. Digital Equipment Corporation was founded in 1957 and had gone public in 1966, growing from $23 million in revenue that year to $710 million in 1976, dominating the minicomputer segment.4

With more than a dozen successful, established, and well-resourced mainframe and minicomputer companies in 1976, why did none of them grab the opportunity Mike Markkula found so appealing? Why did none launch a PC aimed at the consumer and business markets? If any of them had, Apple would not today be a household name. In the five years before IBM finally entered the market, Apple launched the Apple I, the Apple II, the Apple III, grew to $335 million in revenue with net profits of $39 million, and had a highly successful public offering. Without those five years to build a brand and financial cushion, Apple would not have survived the advent of serious competition.

Apple had a couple of competitive advantages—Wozniak’s excellent computer design and Jobs’ understanding of what their customers wanted or, really, needed—but these were not sustainable: the design was there for everyone who understood a circuit board to see and the customer base was evolving so rapidly that any understanding of them had to be constantly renewed. Apple did not start with any sustainable competitive advantages, none of the traditional moats. They had no scale advantages, no valuable patents, and no platforms effects: these cost time or money to build and Apple did not start with either. Without a moat Apple had no ability to prevent incumbents from entering, but for five crucial years none successfully did. Why?

It is 1998. Larry Page and Sergei Brin have developed a simple technique to make Web searches higher quality. They incorporate as Google and enter the fiercely competitive search engine market. Yahoo!, the largest incumbent, already has $200 million in revenue and $482 million of cash after a 1996 IPO. They are a household word. Excite is the second largest, with $154 million in revenue and $61 million of cash, having also gone public in 1996. Others, such as Lycos and Ask Jeeves are significant competitors. There are already winners in this market.

Google’s web search is higher quality than Yahoo!’s or Excite’s. It surfaces results that are more relevant than Excite and more tailored than Yahoo! A year after launch and just three months after incorporating as a company Google is rated one of the top five search engines by PC Magazine Online.5 While this is phenomenal adoption by a company still being run by a couple of PhD students using their university’s servers, Google had  no revenue and Page and Brin were anxious to sell the company or just the technology and get back to their studies. But when they offered the technology to Yahoo!, among others, for a reported $1 million, Yahoo! turned them down.6 The market did not find Google’s innovation to be that valuable. So while Google had a moat—the patent—that patent did not account for nearly the value that Google eventually realized. If their value was not in the patent, where was it?7

Both Apple and Google were doing something entirely new, and some vital part of how their innovations would play out–whether a market would exist for Apple’s computers; how Google would make money eschewing the established business model–was completely and fundamentally impossible to predict. Nobody, not their potential competitors and not even the founders, could foresee how much value they would create, or how. The established companies who could have out-competed these startups did not enter because they, like most successful companies, had processes in place to prevent them from investing in projects that had substantial uncertainty. This left the field wide open for Apple and Google.

There’s an old saying: “predictions are hard, especially about the future.”8 While hard, there are two ways to make reliable predictions about the future: deduction and induction. When either of these is available, prediction is (theoretically) possible.

Deduction relies on playing out an inevitable chain of cause and effect. It is possible when the starting state of the environment and all the mechanisms that will cause that state to evolve are  perfectly known. This is a strong condition. (And perhaps too strong. In most cases, knowing not all but just the important starting conditions and transition mechanisms will get you a good approximation of the future, or at least a good estimate of the probability of success or failure. This may be enough.)

If your startup sells dollar bills for 90 cents you know you will lose money and go out of business. Whether you have seen another business do this exact thing, or something that can easily be analogized to it, is irrelevant to your prediction. Many business ideas can be confidently predicted to fail, even complicated ones, because the chain of cause and effect can be analyzed. Their failure can be reliably deduced.

Induction, the second way to predict the future, assumes the future resembles the past. The ancients may not have known why the sun rose every morning but were pretty sure it would, because it had every day previously. Similarly, some business ideas are statistically predictable because they have been tried multiple times. If you go to a new restaurant you may not be certain if that restaurant will succeed or fail, but you know that fewer than half of restaurants survive their first year.9 New restaurants are similar enough to one another in their most important business aspects that their risk of failure can be induced.

Deduction and induction are very useful in established markets and with established businesses. The state of the environment can be well known through market research, and the transition mechanisms thoroughly explored. Strategies and processes have been inevitably tried many times so the probability of their success can be estimated.

But predicting the future of high-growth-potential businesses is much harder, and perhaps impossible, because these businesses:

  • Are often doing something entirely new; and
  • Must create a new system of connections between the company, the rest of their business ecosystem, and society in order to have the resources and support to grow quickly.

Doing something new is usually a condition to the possibility of high growth because if a new company isn’t innovative they end up competing head-to-head with established competitors and it is much, much harder to grow quickly with few resources. The company doesn’t have to be the first to ever do what they’re doing; Google was not the first search engine. But either the category the company is entering must be new enough that no company or group of companies already dominates it, or its product must be innovative enough to take market share from the incumbents.

When businesses do something new there are, by definition, no prior examples of doing it. Without a set of historical data about the outcomes of similar actions, induction won’t work.

New companies must forge ties into a complex network of potential employees, suppliers, competitors, customers, government regulators, financiers, and societal commentators. How each of these will react to the new company’s actions, and react to each others’ reactions, is often also unknowable. A network of independent actors reacting to each other over time is a complex system and complex systems are often inherently unpredictable—you can’t work from cause to effect. This makes deduction impossible.

The outcome of a high growth potential startups—where neither deduction nor induction is feasible—is inherently unpredictable. This unpredictability is more than the everyday unpredictability businesses often face: how will the cost of my raw materials change next year, how much will revenues grow by, etc. These are a tractable sort of unpredictability where we can predict a most likely answer, a best and worst case, and the odds of it all. There is a knowable probability distribution of outcomes. The unpredictability startups face is different. It can’t be quantified; no probability distribution can be formed. To distinguish the two types of unpredictability, we will call the quantifiable kind risk, and the unquantifiable kind uncertainty.

This isn’t to say that everything about every startup is uncertain. It is just that the nature of high-growth potential tech businesses means that the most important drivers of their value creation must be uncertain. Competitors are attracted to markets where there is value being created, hoping to reap some of it; but individual companies in that market must keep competition to a minimum if they don’t want to lose share and margin. They can do this by digging moats if they have the time or money. But in the time before any moat has been built only uncertainty will deter an onslaught of competitors, leaving decent odds for the few that enter. To succeed, startups must choose to brave uncertainty because most companies, especially established ones, won’t.10

Observers have long noticed that uncertainty is integral to startups. As far back as 1755 the economist Richard Cantillon wrote that an entrepreneur is “someone who exercises judgement in the face of uncertainty”11, though he was vague about both what he meant by uncertainty and why it should be prevalent in startups.

Keynes was more precise about what uncertainty is (in 1937):

By “uncertain” knowledge, let me explain, I do not mean merely to distinguish what is known for certain from what is only probable. The game of roulette is not subject, in this sense, to uncertainty; nor is the prospect of a Victory bond being drawn. Or, again, the expectation of life is only slightly uncertain. Even the weather is only moderately uncertain. The sense in which I am using the term is that in which the prospect of a European war is uncertain, or the price of copper and the rate of interest twenty years hence, or the obsolescence of a new invention, or the position of private wealth-owners in the social system in 1970. About these matters there is no scientific basis on which to form any calculable probability whatever. We simply do not know.12

But it was economist Frank Knight in his 1921 book Risk, Uncertainty, and Profit who addressed entrepreneurial uncertainty directly. (Note: Knight uses ‘profit’ in the same way as Schumpeter, to mean entrepreneurial or excess profit.)

Profit arises out of the inherent, absolute unpredictability of things, out of the sheer brute fact that the results of human activity cannot be anticipated.13

Knight described uncertainty as something not knowable in advance as a measurable probability, distinguishing between uncertainty and mere risk.

[A] measurable uncertainty, or “risk” proper, as we shall use the term, is so far different from an unmeasurable one that it is not in effect an uncertainty at all. We shall accordingly restrict the term “uncertainty” to cases of the non-quantitative type. It is this “true” uncertainty, and not risk, as has been argued, which forms the basis for a valid theory of profit.14

Risk can be quantified, uncertainty can not. Both of these lead to unpredictability but, as noted, they are qualitatively different.

Like Schumpeter, Knight thinks that “profits are…the result exclusively of dynamic change.”15 But he thinks that only change linked to uncertainty can create entrepreneurial profit because while established businesses are comfortable taking risk, they are not comfortable with uncertainty. If it makes business sense to take risk, many businesses will take it, and the risky innovation will have competition from the start.

No a priori argument is necessary to prove that with general foreknowledge of progressive changes no losses and no chance to make profits will arise out of them…It cannot, then, be change which is the cause of profit, since if the law of change is known, as is in fact largely the case, no profits can arise…Change may cause a situation out of which profit will be made, if it brings about ignorance of the future. Without change of some sort there would, it is true, be no profits, for if everything moved along in an absolutely uniform way, the future would be completely foreknown in the present and competition would certainly adjust things to the ideal state where all prices would equal costs.16

Of course, established businesses often decide to take risks despite the potential competition. They can do this because they have a moat that gives them the competitive advantage to reap profits. Because startups do not have a moat, they can’t.

Another way to think of this is to consider risk not an unknown but just a cost. A measurable risk can easily be transformed into a known cost by buying insurance. And this cost is the same for anyone taking that risk. This works because aggregating many measurable risks often results in a predictable outcome for the aggregate.

Imagine a game of dice. You throw the dice multiple times but you only win if you roll a seven. How often will you win? There are six ways of throwing a seven, out of 36 possible ways for the dice to come up. So seven shows up 1/6th of the time. But if you throw the dice once, you either win or lose…you don’t 1/6th win. If you throw them twice you either win-win, win-lose, lose-win, or lose-lose…your win frequency is 100%, 50%, or 0%, with 50% twice as likely as 100% or 0%. The more times you throw, the more likely you are to win very close to 1/6th of the total throws.

Below are several runs of a computer ‘throwing’ the dice 200 times. You can see the win percentage over time converging on 1/6th.

200 rolls of the dice

After throwing the dice 200 times, there is a 95% probability that your win percentage will be within 5% of 1/6th.17 Throwing the dice once yields a surprise, throwing them 200 times yields none, in aggregate.

Even if you only have the opportunity to throw the dice once, you can lay off the potential surprise by buying insurance: if many others are also throwing the dice, the various risks can be pooled until they’re not really risk at all. No one knows the day and hour of their death. But your life insurance company can sell policies to thousands of people and know very closely how much they will pay out every year. By either aggregating a bunch of risks internally or externally insuring single risks, an established business can manage the potentially surprising downside of risky projects and bad predictions. “If the actuarial gain or loss in any transaction is ascertainable…the burden of bearing the risk can be avoided by the payment of a small fixed cost limited to the administrative expense of providing insurance.”18

Risk is insurable but uncertainty is not, because risk is only locally unpredictable while uncertainty is globally unpredictable. Startup uncertainty may be of the sort where the dice are only thrown once, ever, because once the experiment is run the results are there for all to see. Or it may be of the sort where aggregating many risks does not actually make the aggregate more predictable, as is the case with many complex systems.19 This is why established businesses are fine taking risks: it is equivalent to just another cost, so it can be accounted for. But uncertainty can’t be reduced to a cost, so it can’t be included in projections or plans.

Many mainstream economists do not believe uncertainty, as distinct from risk, really makes any difference. They believe that people will assign probabilities regardless of whether or not they have the basis to do so.20 That is, you can find someone to sell you insurance on anything, no matter their lack of actuarial data. This view, often attributed to statistician L.J. Savage, began to lose support when one of Savage’s colleagues, the noted decision theorist Daniel Ellsberg, used a thought experiment to show that people are ‘ambiguity averse’: they shy away from uncertainty.

The Ellsberg Paradox, as it’s called, imagines there are two urns (statisticians love urns), the first with 50 black balls and 50 red balls and the second with 100 balls that are either red or black, but you don’t know how many of each. You are asked to pick a ball from an urn. If it is red, you win $100, if it is black you get nothing. Which urn would you choose from?

Most people prefer to pick from the first urn, the one with the known proportion of red and black balls. They prefer a known probability of winning to an unknown probability. Even more, if they are then given a second chance but told they will win the $100 if they pick a black ball this time, they will still choose from the first urn. They are not reacting to some preconception of how many of each type of ball is in the second urn, because then they would switch on the second bet. They simply prefer the devil they know to the devil they don’t.21

An experiment that doesn’t involve urns comes from Richard Zeckhauser.  He asked a group of people their estimate of the probability that an asteroid had passed within some distance of the earth. The group believed there was a 3% chance that a 10,000 ton asteroid had passed within 40,000 miles of the earth in the previous decade. He then offered, to a different group of people, the choice of winning $2000 if a ball numbered 17 was drawn from 100 consecutively numbered balls or $1000 if a 10,000 ton asteroid had passed within 40,000 miles of the earth in the previous decade. While the expected value of the first option is obviously $20 and the expected value of the second should be $30, most people picked the first option because they preferred the quantifiable risk to the uncertainty.22

It is easy to see examples of uncertainty-averse decisions in real life, especially in the tech industry. These decisions are not irrational, nor are they the result of bad information, they are the rational result of not being able to know the future. This not knowing does not usually take the outward form of trying and failing to form a probability distribution, it usually never gets that far.23

  • Lawyers at Bell Labs did not want to patent the laser, invented there, because they thought it wasn’t relevant to the telephone business;
  • Western Union withdrew from the telephone business in 1879 in exchange for Bell’s promise not to enter telegraphy–they were more worried about a competing Bell telegraph business than being made obsolete by the telephone;
  • The inventor of the radio did not see how broadcasting could be a viable business;
  • The inventors of the cell phone (Bell Labs again) thought it would be primarily used to reach inaccessible places, not as a replacement for the landline;
  • Ken Olsen, the founder of Digital Equipment Corporation, said in 1977, that “he couldn’t see any need or any use for a computer in someone’s home.” 24

Etcetera. We could catalog the inability to predict the future of technologies and the businesses built to commercialize them for thousands of pages.

Ellsberg and Zeckhauser talked about individuals being uncertainty-averse, but large businesses are even more so. Aside from being managed by individuals, successful and well-run businesses usually have processes in place that are precisely and knowingly tailored to exclude uncertainty. And where there aren’t formal processes, informal norms do the job. While these are usually meant either as a control mechanism in large hierarchies or as a way to force decision makers to confront the likely outcome of their decisions so they don’t go off half-cocked, they have the side-effect of preventing plans whose outcome can’t be adequately predicted. Clayton Christensen wrote

Companies whose investment processes demand quantification of market sizes and financial returns before they can enter a market get paralyzed or make serious mistakes when faced with disruptive technologies. They demand market data when none exists and make judgments based upon financial projections when neither revenues or costs can, in fact, be known. Using planning and marketing techniques that were developed to manage sustaining technologies in the very different context of disruptive ones is an exercise in flapping wings.25

Christensen says this about disruptive innovations, but it is true of any innovation subject to substantial uncertainty. No company can make decisions about uncertainty because the information their processes demand does not yet exist. Kenneth Arrow considered information to come as feedback from markets, by which he meant actual commercial activity.

[T]he info needed by the optimizer is not provided by an existing market. It will be provided by a market that will exist in the future, but that is a bit too late to help in decision made today…In short, the absence of the market implies that the optimizer faces a world of uncertainty.26

But then, if a company is making decisions about a new commercial activity, it can’t have the required information to make the decision until after the commercial activity starts. Someone has to create that info, and it won’t be the ‘optimizer’, the process-driven incumbent.

Companies install these rigid processes because without them bad decisions are sometimes made. Stefan Thomke wrote

When it comes to innovation…most managers must operate in a world where they lack sufficient data to inform their decisions. Consequently they rely on their experience or intuition. But ideas that are truly innovative—that is, those that can reshape industries—typically go against the grain of executive experience and conventional wisdom.27

Large companies learn to instead wait for more information to become available rather than face uncertainty, even if they strongly suspect this will cost them money in the end. The tension in large hierarchical organization between where decisions are made and where accountability is held naturally push organizations into needing rational explanations for their actions to get them approved. Experiments have found that “when subjects anticipate the need to be able to explain (or even expose) to others [their choices in uncertain situations], ambiguity becomes even more undesirable than it is in isolated personal choice settings.”28 People working at established companies must anticipate that they will be asked to explain their decisions, especially their bad decisions. These explanations need to follow procedures or frameworks the company adopts to try and avoid bad decisions before they are made. Investment memos, financial models, payback periods, discounted cash flow models, etc. As Christensen says, these all demand quantification of market sizes, financial projections, and financial returns. Decisions made outside of these models and without these quantifications are very difficult to justify as rational after the fact, especially if they go bad.

Imagine you work for a large company and propose a plan for a project that has a quantifiable chance of producing a specific multiple of its outlay within a set period of time. That is, the project is risky, but measurably risky. The sort of thing where the recommendation memo says something like “The project has a 50% chance of a 3x return on investment within two years.” Rational managers can decide whether to proceed with the project based on its calculated expected outcome. Insurance companies and casinos do this as a matter of course and every business does it to some extent. This is a quantifiable risk with a positive expected outcome and the decision can be defended no matter how it turns out.

Now imagine walking into your boss’s office and presenting an investment rife with uncertainty. “How likely is this to succeed?” your boss asks. “I don’t know.” you say. “How big will it be if it works?” your boss asks. “I don’t know.” you say. “Why don’t you know?” “Because the customers may be different than who we think; because the customers may want a somewhat different product; because the other companies we need to produce complementary products may decide not to.” Etc. “Well,” your boss says, “we’ll just have to wait until we know those things before we can make a decision.”

Unfortunately, these things and many others may be uncertain—meaning they can’t be known beforehand, the information does not yet exist. Your boss will never approve the project. Many entrepreneurs find themselves in the same situation: facing these substantial uncertainties. But without the boss they can go ahead with the project anyway. The entrepreneur may decide to act, while the boss will not, because the entrepreneur does not need to explain themselves to anyone.

Established businesses are uncertainty-averse far more than they are risk-averse. Risk can be managed by building a portfolio of projects. But managers can’t mitigate uncertainty ahead of time. How can you plan for something that is inherently and irreducibly uncertain? Most businesses will simply decide not to embark on uncertain projects. This doesn’t mean established companies never purposely do things that have some measure of uncertainty, but they are far less likely to. And the more obvious the uncertainty is, the less likely an established company will take it on.

That startups can do something without having to compete with better-resourced companies is not a side-effect of entrepreneurship, it is a prerequisite. Startups may open new markets or may use innovations to create a valuable new product, but if these new markets can be exploited or the new product can be imitated by other companies the chances for success are far lower. Uncertainty about the prospects for the new market or product will cause most other companies to spend neither time nor money on it. At their best they will give it lip service, more probably they will ridicule it, but most likely they will simply ignore it or quash it internally.

When Mike Markkula said Apple would be a Fortune 500 company in two years, he was wrong: it took six years. But it was still a crazy thing to say. First off, there was no way to know how big the PC industry would get. At that time fewer than 50,000 personal computers had been sold, probably entirely to electronics enthusiasts since they required some degree of self-assembly and a large degree of willingness to put up with the vagaries of a half-baked and poorly documented product. There was also for years very little commercial software for personal computers and no idea what they would be used for by the new audience of mainstream computer owners. The spreadsheet was a couple of years in the future, word processing was better done on dedicated machines from IBM and Wang Laboratories, and games were better played on the purpose-built cabinet-enclosed upright video games found in arcades, or even on the dedicated home video game consoles like the Atari 2600, released that same year. The personal computer did nothing best.

Even if Apple could have predicted the various ways people would use the personal computer, Markkula couldn’t have predicted the timing. The many pieces that had to fall into place, from the coevolution of PC hardware and software, to the pace of hardware innovation, to the willingness of mainstream customers to believe that the advantages of adopting the new technology would outweigh the costs of changing longtime routines. This massive uncertainty is what kept companies like IBM from entering the market for years. It wasn’t until 1980 that Bill Lowe of IBM managed to convince corporate management that a special division outside of the company’s bureaucracy should be set up to develop a personal computer (previous tries had been doomed by design by committee.) He did this by showing the executives how big the PC market had already become.29 Once the uncertainty had been mitigated, established companies could do the analyses they needed to approve entry.30

When Google tried to sell their technology to an established search engine company they were turned down or thrown out: no one wanted better search results because the result of better search results would be searchers clicking on the results and leaving the site, lowering advertising revenue. The CEO of Excite said “he wanted Excite’s search engine to be 80 percent as good as the other search engines.”31 Google did not have a viable revenue plan at that time, and there was no way to know ahead of time what one would be or how customers would eventually react to it if one were found.

Google eventually adopted the new business model of pay-per-click advertising, where advertisers only pay if the viewer of an ad clicks on it. This neatly aligned the incentives of the searcher, the advertiser, and the search engine because if the search results were good then the searcher would be more likely to click on one of them, leading to revenue for the search engine and a potential customer for the advertiser. It was not a technological insight nor difficult to implement, but it was an idea that had been nay-sayed by critics across the board when first introduced by Bill Gross at Goto.com. It may look obvious in retrospect, as uncertain things often do, but if it had been anything other than uncertain, incumbents would have quickly adopted it and then been free to develop better search technology. If that had happened, Google would never have been able to get traction.

Apple and Google’s respective uncertainties gave them time to establish themselves. But once the world could see their strategies working the uncertainty was gone; competitors could enter the market and compete. Actions to decrease uncertainty have the unfortunate side-effect of making a market more attractive to others. Apple and Google survived the entry of competitors because by this point, years after they started and with plentiful cash, they had each established a traditional moat. Uncertainty can create a no-go zone around a new market, allowing a startup to build without competition for some period of time, but the startup’s management has to remain aware that if they are successful, competition will eventually arrive. Uncertainty creates space and time, but only a limited amount. Wise startup management will make use of what time they have to build a moat.

Managing a high-growth-potential startup is fundamentally different than managing any other type of company. To be successful a founder must seek out uncertainty and then must manage the company through it and emerge from the other end with a moat. Uncertainty creates a difficult trade-off for entrepreneurs. Without uncertainty they will immediately face competition from many others, including some who are better resourced. But a business subject to high levels of uncertainty has seemingly intractable management problems.

Fools rush in where angels fear to tread, and founders start companies where experienced managers fear to go. Experience has a point: aside, entirely, from not being able to know the chances of success, uncertainty creates a more immediate problem. How do you, as a manager in an uncertain environment, choose a strategic path? How do you choose a goal when you do not know all the possible goals? How do you make decisions that will move you towards that goal when you do not know if you know all the possible decisions you could make or their chances of success? This is not just the classic fog of war, it is the fog of war that obscures not just the situation on the ground but also who you are fighting against and even who you are fighting for.

You may be uncertain about whether your new technology will work as hoped, how long it will take before it works well enough to please its users, and even who will use it and for what. You may be uncertain about how many customers will want it, who will pay for it, and how much, and how soon. You may be uncertain about how others will react—your competitors, companies whose product you are replacing, the government, and society as a whole. You will have to convince financiers, customers, employees, and yourself that your idea is a good one, even though you can’t really know yet. Meanwhile, you need industry incumbents and other potential startup founders to continue believing your idea is too uncertain to want to compete with you.

Of course, these can’t really be intractable problems, because startups succeed in solving them every day. Startup strategies like lean, customer development, design thinking, and many others were all created to manage uncertainty. These strategies were created out of necessity by hands-on practitioners. Like similarly developed ways of working, they can be made more useful by connecting them to underlying theory. If you understand where uncertainty comes from and how it effects decision processes you can craft the right strategy for your company’s situation.

Uncertainty is no longer a primary protector of Apple or Google’s business. They have become the incumbents they once had to avoid. Their decision making can be made using traditional business strategies, which were developed explicitly for big companies like them. They are no longer startups. But while they were startups, before the time they could develop moats, while uncertainty was still a primary factor in how they ran their businesses, they still needed to answer the question “how can I make decisions when I can have no idea what they will lead to?” A startup’s primary driver of strategy has to be answering this question, figuring out how to manage uncertainty.


  1. Isaacson, W., Steve Jobs, New York: Simon & Schuster, 2011, p.76. 

  2. The stats on the number of PCs sold come from Reiner, J., “Total share: 30 years of personal computer market share figures”, Ars Technica, 12/15/2005. Retrieved from https://arstechnica.com/features/2005/12/total-share/ on 11/11/2019. The revenue number is partly from Isaacson and partly an estimate based on the sales price of the Apple I. 

  3. Forbes, “Fortune 500”. Retrieved from https://archive.fortune.com/magazines/fortune/fortune500_archive/full/1976/ on November 25, 2019. 

  4. Wiseman, T., “EDP/IR: 1976 Marked Springboard Year for Industry”, Computerworld, June 13, 1977, pp. 107-108. 

  5. PC Magazine Online, “Top 100 Web Sites, Search Engines: Google!”, from the Internet Archive, December 1998. Retrieved from http://web.archive.org/web/19990508042436/www.zdnet.com/pcmag/special/web100/search2.html on 1/24/2014. 

  6. Sherman, Chris, “How Blunders And Myopia Helped Fuel Google’s Rise To Dominance”, Search Engine Land, 4/4/11. Retrieved from http://searchengineland.com/how-blunders-and-myopia-helped-fuel-googles-rise-to-dominance-71448 on 1/24/2014. 

  7. Interestingly, Apple had also tried to sell its technology and designs before raising money: Wozniak had offered the initial circuit board design to HP, where he was an employee; Jobs had tried to convince Atari, where he had worked, to license the design; and Jobs had pitched Commodore, who later entered the market, on buying the company. They had all declined. (Isaacson, 2011, pp. 64-65, 72.)  

  8. Often attributed to Niels Bohr, but it predates him. 

  9. https://www.cnbc.com/2016/01/20/heres-the-real-reason-why-most-restaurants-fail.html 

  10. Some startups succeed because they are lucky, and some because the founders are business masterminds. But, this being a work on strategy, we will ignore the former. And the latter isn’t a source of excess profit: the business mastermind should, again in theory, get paid the same amount their genius would have earned in entrepreneurship working for someone else. 

  11. Hebert, R.F., A.N. Link, The Entrepreneur: Mainstream Views and Radical Critiques, New York, NY: Praeger, 1982. 

  12. Keynes, J.M., “The General Theory of Employment”, The Quarterly Journal of Economics (1937) 51 (2): 209-223. doi:10.2307/1882087. 

  13. Knight, Risk, Uncertainty, and Profit, Boston MA: Hart, Schaffner and Marx; Houghton Mifflin, 1921, p.168. Retrieved from https://oll.libertyfund.org/titles/306 

  14. Knight, Risk, Uncertainty, and Profit, p.24. 

  15. Knight, Risk, Uncertainty, and Profit, p.38 

  16. Knight, Risk, Uncertainty, and Profit, pp. 40-41. 

  17. This is just the binomial distribution at work. 

  18. Knight, Risk, Uncertainty, and Profit, p. 24. 

  19. For instance, many complex systems result in power law distributed outcomes (cf. West, G., Scale: the Universal Laws of Growth, Penguin Press, 2017.) When you aggregate many long-tail events, described by power-law distributions, you do not necessarily lower the variance of the aggregate. 

  20. “I have not referred to this distinction because I do not believe it is valid. I follow L.J. Savage in his view of personal probability, which denies any valid distinction along these lines. We may treat people as if they assigned numerical probabilities to every conceivable event.” Leroy, S., & Singell, L., “Knight on Risk and Uncertainty”, Journal of Political Economy (1987) 95(2), pp. 394–406. 

  21. Ellsberg, D., “Risk, Ambiguity, and the Savage Axioms.” The Quarterly Journal of Economics, Vol. 75, No. 4 (Nov. 1961), pp. 643-669. Ellsberg uses the word ambiguity instead of uncertainty because, I suppose, it is less ambiguous. We will continue to use the word uncertainty, but remember we are using it in its strong form. 

  22. Zeckhauser, R., “Investing in the Unknown and Unknowable”, Capitalism and Society (2006) 1(2). 

  23. This draws heavily from Rosenberg, N., (1996), “Uncertainty and technological change“, Conference Series ; [Proceedings], 40, issue Jun, p. 91-125, https://EconPapers.repec.org/RePEc:fip:fedbcp:y:1996:i:jun:p:91-125:n:40

  24. Ahl, D., “Interview with Gordon Bell”, Creative Computing, Volume 6, Number 4, April 1980, p. 89. Retrieved from https://archive.org/stream/creativecomputing-1980-04/Creative_Computing_v06_n04_1980_Apr_djvu.txt on 11/13/2019. 

  25. Christensen, Clayton, The Innovator’s Dilemma, Boston MA: Harvard Business School Press, 1997, pp. xxi-xxii. 

  26. Arrow, K.J., “Limited Knowledge and Economic Analysis”, American Economic Review (1974) Vol. 64, Issue 1, pp. 1-10. Available at SSRN: https://ssrn.com/abstract=1506319 

  27. Thomke, S. & J. Manzi, “The Discipline of Business Experimentation”, Harvard Business Review, December 2014. 

  28. Kleindorfer, Paul R., “Reflections on Decision Making Under Uncertainty” (December 2, 2008). INSEAD Working Paper No. 2008/73/TOM/ISIC, p.14. Available at SSRN: https://ssrn.com/abstract=1310239 

  29. Chposky, J., and T. Leonsis, Blue Magic, London: Grafton Books, 1988. 

  30. Before that, even the established computer companies that did enter could not resist tailoring their machines to the less-uncertain audiences they knew. Hewlett Packard, soon after declining Wozniak’s offer of the initial Apple technology, decided to build a personal computer. But their “Capricorn” project launched a machine aimed only at the scientific and professional users that larger machines already catered to: they could not be sure there were any other customer segments. Xerox, likewise, entered the PC market before IBM, but their machine was far too expensive (at more than $16 thousand) to be affordable to other than the professional market. Apple had staked its future on the yet-unproven consumer and business markets, the markets that turned out to be the fastest-growing, and eventually the largest. It was only after these markets were proven that IBM decided to enter. 

  31. Levy, S., In the Plex, Simon and Schuster, 2011, p. 30.