Uncertainty

Productive Uncertainty

It’s tempting, as a venture investor, to back companies developing amazing new technologies. But this often doesn’t work out as well as investing in companies that are using existing technologies in a new way. New technologies have a larger long-term societal impact but new markets are better investments.

Because new markets begin small and their impact is measured along a different metric than the one people are used to, they are often derided as “we wanted flying cars, we got 140 characters.” This exhortation disguised as an observation leads venture investors astray. 140 characters–a new market–was a great startup investment; flying cars–a new technology–has been a poor one. Take, for instance, new technologies most of us would much rather have exist than flying cars: cleantech.

In 2007 legendary venture investor John Doerr said “Green technologies…could be the biggest economic opportunity of the 21st century” and he committed Kleiner Perkins to investing $200 million in cleantech. This is a perfect example of wishful thinking: of the $25 billion of investment in the sector between 2006 and 2011, only half was returned; a huge miss. What went wrong?

The MIT Energy Initiative said “cleantech companies commercializing innovative science and engineering were especially unsuited to the VC investment model” because “working out the kinks in new science is time consuming” and “the likely acquirers were…unlikely to acquire risky startups and averse to paying a premium for future growth prospects.”1

The World Economic Forum proclaimed three lessons from this failure:

  1. Energy investments are inherently prone to capital intensity;
  2. Regulation matters;
  3. Not all fields and technologies within the energy sector grow in the same manner.

These three things are (trivially) true. They are also not what the venture investors got wrong. Kleiner Perkins didn’t overlook, for instance, that regulation matters.

What really happened is this: investors bet on startups whose products were better than their competitors’ products. This sounds like a good thing, but it is the wrong strategy: they bet on companies with competitors. Investors need companies that don’t have much competition to start and that can build moats to prevent competition later. Startups whose success is predicated on a new, better technology rarely make the transition from innovator to dominant player. Technology in itself is usually not a moat, and companies that are based on introducing new technologies rarely get to build moats. Investors need to invest, instead, in companies that are entering new markets.

This post tries to answer a riddle I wrote about five years ago.

“Technical risk is horrible for returns, so VCs do not take technical risk…VCs have always waited until the technical risk was mitigated…Market risk, on the other hand, is directly correlated to VC returns.”

Heat Death: Venture Capital in the 1980s

This has bothered me since. Why should there be a difference between technology risk and market risk in determining returns?

My answer, five years later, is that it is not the risk, it is the uncertainty that matters.

Some background, to get you back up to speed after my extended hiatus.

In Schumpeter on Strategy I argued that companies create excess profit through innovation, and keep making this excess profit by protecting the innovation from being copied. In a perfectly competitive market, competition reduces economic profit to zero. For a company to have an excess, or entrepreneurial, profit, it must do something differently than its competitors. The resulting excess profit only lasts until the innovation is imitated by competitors. The sum of the excess profit from innovation through perfect competition I call excess value.2 Companies can lengthen the time between introduction of an innovation and imitation, and thus increase excess value, by creating barriers to entry, or moats.

In A Taxonomy of Moats I argued that the only moat that can create excess value for a new startup is uncertainty. This argument is two-fold. First: if a moat exists prior to the startup being founded (say, a patent) then, absent uncertainty, this patent could be sold for at least as much as the startup could garner from it. In this case, no excess value is created by the startup: it already existed, inherent in the patent. Second: if a startup can’t start with a moat, then it must build a moat over time. Uncertainty keeps competition at bay long enough for a moat to be built.

By uncertainty I mean something different than risk. I mean Knightian Uncertainty: the inability to predict, even probabilistically, what will happen. I talk a lot about this concept in Startups and Uncertainty. This inability to predict what will happen when a startup pursues an innovation keeps other companies from entering to compete: they will look at the opportunity and say things like “it looks like a toy” or “there’s no market for that.” Once the startup begins to succeed they will re-evaluate, but a smart startup will have built a moat by then. The uncertainty gives the startup time and competitive space.

Companies that have no excess profit still have value, but it is more or less equal to the market return for any similarly risky asset; it is all beta, no alpha. If a venture capitalist wants alpha, they must invest in uncertain opportunities.3

All excess profit is created through innovation, so VCs have to invest in innovative companies. And for a VC to make a good return there must be substantial excess value, so VCs have to invest in companies facing uncertainty.4 But the converse is not true: not every company that faces uncertainty is a good investment. VCs seem to have a poor track record of making money in certain types of substantially uncertain companies: it seems VCs are far more likely to make money in companies serving new markets than companies introducing new technologies.

This is similar to something Clayton Christensen noticed in The Innovator’s Dilemma. Christensen looked at survival rates of new disk drive companies and found that startups going into new markets are far more likely to be successful than startups just using new technology.

Data from Christensen, C., The Innovator’s Dilemma, Harvard Business School Press, 1997, p. 131.

He reasoned that successful companies cater to the needs of their existing customers. They downplay what non-customers ask for, and new markets are composed of non-customers. On the other hand, if a new technology improves the product for their existing customers, those customers will ask for it and the incumbent will quickly adopt it. Startups can’t survive direct competition with better-resourced incumbents, so those that offer new technology to improve on existing products rarely survive. They must instead offer something to a class of customers the incumbent did not serve. Sometimes it is new technology that allows servicing these new customers to be profitable, but it is the new market, not the new technology that is the key.

This is an astute observation, but there are both striking counter-examples and theoretical questions. When Genentech introduced a synthetic human insulin manufactured by genetically engineered bacteria, the startup successfully entered an existing market (human insulin) with a new technology (genetic engineering). On the other hand, existing customers of Amazon (online shoppers) were not asking for cloud computing, a new market. Regardless, Amazon became one of the main competitors. In the former case, a new company succeeded in an existing market with a new technology, and in the latter an incumbent succeeded in a new market. Both seem to defy Christensen’s precept.

Why do incumbents immediately copy new technology when they see it is working but don’t immediately enter new markets when they see they are working? The Christensenian answer that new markets start small and seem inconsequential has a symmetrical argument vis a vis new technology: new technologies seem like toys. The disruption argument rings true, but it’s vague. Why are incumbents adaptive in one way but not the other?

The answer is that to adapt they have to feel comfortable with how uncertain the new technology or new market is. Incumbents strongly dislike uncertainty so they wait for it to be mitigated. But startups can build moats in new markets while they are still uncertain where they usually can’t with new technologies.


Productive Uncertainty

Uncertainty, generally, is something to be avoided. If you can’t predict the outcomes of your actions you will have a hard time planning and managing. And if others see that your business proposition is uncertain they will shy away from including your product in their plans. But uncertainty can also shield against competition, allowing you to create excess value. If it does, it is productive uncertainty. Innovations, because they are new, usually come with uncertainties of one sort or another. Founders have to choose the subset of innovations where the uncertainty is productive to have the best chance of succeeding.

What makes some uncertainties productive and others not? Every business must eventually mitigate the uncertainty it started with. How and when this mitigation unfolds determines whether the uncertainty is productive or not. At a high level, there are two basic sources of uncertainty a high growth potential technology startup faces, and they are mitigated differently. These two types are novelty uncertainty and complexity uncertainty.

Novelty Uncertainty

When something has not been done before, it may be that no one can predict the outcome. Prediction relies on either inductive or deductive reasoning: the first requires data and the second requires an understanding of the process that produces the result. Novelty uncertainty results when we have neither. For example, with no well-understood theory of aerodynamics the Wright Brothers could not know if their 1903 Flyer would leave the ground until they tried it. This is novelty uncertainty.

Novelty uncertainty raises questions like:

  • Will the technology work?
  • How long will it take and how much will it cost to prove it will work for the product we are trying to build?
  • How long will it take and how much will it cost to prove it can be scaled to commercial levels of production?
  • What will the quality of the resulting product be?
  • Will we be able to improve that quality over time?
  • What level of quality do we need before it is useful?

Etc.

If there is substantial uncertainty, many of these questions can not be answered. When new technologies are uncertain, it is generally novelty uncertainty.

Complexity Uncertainty

It is impossible to predict what many complex systems will do. Systems are composed of many interacting agents, each making decisions by their own hard-to-know-for-sure rules, and some of the inputs to their decisions are the results of other agents’ decisions. Both the opacity and the feedback loops can make outcomes impossible to predict; startups interact with systems that have both.5 For example, during the ‘War of the Currents‘ in the late 19th century, whether Edison’s direct current or Westinghouse’s alternating current would become the eventual standard depended on competing technical, social, and economic interests. Because the decision-making of a system like this is iterated and path-dependent, it was not clear there was a best objective result, nor even if there were it would win out. Both Edison and Westinghouse faced complexity uncertainty.

Complexity uncertainty raises questions like:

  • Who and how many people will want this product?
  • What will they use it for?
  • Not knowing what they will use it for, what design is best?
  • How do we convince people to buy it, not knowing what they will use it for?
  • What price will people be willing to pay?
  • Who will partner with us to make needed ancillary products or integrate our product into customers’ workflows?
  • Will our suppliers take us seriously enough to customize our inputs for us?
  • Will established companies in the field decide to compete with us?
  • How will the media, the government, and society at large react?

Etc.

Complexity is a barrier when predicting the evolution of new markets, whether they use new technology or not.

Working with Uncertainty

Uncertainty must be mitigated over time for startups to successfully introduce and sell their products. Uncertainty scares away customers, employees, and suppliers. It increases the cost of financing the business. And not least, it makes planning impossible and management hair-raising. One of the goals of any startup has to be to work to mitigate the uncertainty it faces when it starts.

The uncertainty resulting from novelty can be mitigated by action. That is, you answer questions like “will it work?” by building it and seeing if it will work. What you learn by resolving novelty uncertainty is stable; it is a foundation you can build further learning on. Having flown, the Wright brothers could be confident they could fly again, if only by doing the exact same thing.

The uncertainty resulting from complexity can’t be entirely mitigated by learning because you can’t ‘learn’ what a system will do when it does something different each time. (I say not entirely because real systems tend to have bounds they act within most of the time, and these can be learned, even though relying on them leaves you open to ‘black swans’.) Mitigation has to take the form of waiting for the system to reach an equilibrium on the issue you’re uncertain about or of modifying the system itself, by, say, creating a narrative about what you are doing. Because your uncertainty about what the system will do is mirrored by every other agents’ uncertainty about what you will do, mitigating their perceived uncertainty can cause them to be more predictable, resulting in an actual decrease in uncertainty. Other ways of modifying complex systems include reducing the variance of feedback among agents by standardizing interactions through, for instance, contracts, standards, or common understandings. Modifying the system to suit your company rather than modifying your company to suit the system is “creating the future you want to see happen.”6 We’ll explore these things further in another post.

Note that neither all novelty nor all systems generate uncertainty. The entrepreneur needs to find those that do.

From Uncertainty to Moat

Since uncertainty is what is keeping unbridled competition away, once the startup has mitigated the uncertainty they will face competition unless they have built some other moat. A startup’s strategy must have two pieces: managing through and mitigating the uncertainty it started with, and building a moat. Which moats are available depends greatly on whether the uncertainty the startup is resolving is novelty or complexity uncertainty.

For simplicity, this analysis will generally identify new technology companies as having primarily novelty uncertainty and new market companies as having primarily complexity uncertainty. While this is not always true, the analysis is easily extended to the corner cases.

New Technology Company Moats

New technology is hard to protect once it is commercialized. Technology innovators often hope to protect their ideas with patents, but patents are only valuable when competitors can’t easily find a substitute innovation.

In a society that is capable of generating rapid technical progress, no single innovation is indispensable. However, the reason for this is not that individual innovations do not matter, in some absolute sense, but rather that such a society can readily generate substitute innovations. It is precisely the capacity to generate many possible innovations that renders any single innovation expendable.

Nathan Rosenberg, Inside the Black Box: Technology and Economics, p. 29.

For companies staking their future on a single technical innovation, this is bad news.

If the moat must be a patent, the technologies that can be protected are limited. Sometimes, at the beginning of technological waves, there are ideas so fundamental and simple that the patent can be broad and vague enough to preclude almost any competitor. Edison’s patent on the telephone, for instance, or James Watt’s patents on steam engines. Patents might also be effective on things discovered through costly trial-and-error, like many pharmaceuticals. Trial-and-error has to be paid for up front, and pharmaceuticals tend to have a low marginal cost to produce, so economies of scale kick in once the product is commercialized. This lowers the incentive for competitors to create a substitute. And because a newly introduced pharmaceutical can generate substantial cashflow very soon after commercialization, the company introducing it can afford costly patent defense. (Of course, because trial-and-error is costly up front, it is often funded by government grants, universities, or established companies. Any startup coming from these sources has created much of their value before being founded.)

Without a patent new technology can’t be so easily protected. So when a new-tech startup resolves the uncertainties and introduces the technology to customers, incumbents take note…especially when the new company is selling to the incumbents’ own customers. If these customers find the new technology valuable, the incumbent will quickly either copy the innovation or find a substitute. Then the startup faces a better resourced competitor, limiting its opportunity to create excess value. This imitation can happen quickly enough that the startup does not have the time to build any other moat.

Given time a startup can build economies of scale around a new technology, or special know-how–either closely held knowledge or tacit knowledge. Special know-how works as a moat when the product being sold is not the technology itself but the result of using the technology, as at Genentech. Genentech’s eventual moat was not patents; the patents they held did not protect them from competition. The real moat came in building the tacit knowledge to keep creating new products using their technology. Genentech had the time to build this knowledge because the initial uncertainty about whether the technology could be scaled kept many of the large, incumbent pharmaceutical companies from immediately competing with them. Regardless, in most cases for most new technology companies to succeed, they must be able to create the time to build one of these moats.

New Market Company Moats

Companies addressing a new market primarily face complexity uncertainty, and because complexity uncertainty is mitigated either by surviving through the time it takes for the system to reach some equilibrium or by working with the system to reduce its complexity, these startups have the opportunity to build many sorts of moats.

If the introduction of a new technology product means most of its uncertainties have been resolved, its launch is a starting gun for competitors. In a new market, a startup introducing its product does not necessarily resolve the uncertainty: it may last for some time afterwards as customer, supplier, and societal reactions try to find equilibrium. During this time the startup can build moats. By beginning to make and sell their product, the startup can start to build a brand, economies of scale, or network effects before competitors enter. By the time the market has matured to the point that it seems manageable to incumbents, the startup may already be secure.

When Amazon started it was unclear if people wanted to buy books without seeing them in person, flipping through them, and getting the expert curation of bookstores. Even after launch when it became clear there was a market, how big the market would become was still uncertain. This prevented incumbents like Barnes & Noble from wholeheartedly embracing the new market and gave Amazon time to build their brand.

Even before entering the market, if the startup works to make the system less uncertain it can do so by making it less uncertain for itself and not others. Contracts, standards, narratives, etc. can all be customized to suit the startup better than potential competitors. System rigidity is one of the primary sources of sustainable competitive advantage: since making the system more rigid is what removes complexity uncertainty, startups can take advantage of this to build these moats.

For instance, Netflix faced complexity uncertainty when it began streaming video: what would the reaction of content owners be? Would they agree to allow Netflix to license their content? The uncertainty was resolved by signing contracts for streaming rights. These contracts incidentally made it hard for other potential streaming companies to compete. This strategy can also involve several startups cooperating to introduce a new way of doing business that obsoletes an older way.

These two strategies are not exclusive, many companies in new markets use both.

Creating Excess Value

If the point of a startup is to create excess value then the startup needs an innovation that can be defended over time. This means the team must not only find an innovation and use it to create a product customers want, they must have a strategy to defend their position once they have mitigated the uncertainties. How they defend depends on what they can do to build a moat while they are resolving the uncertainty and how much time they have to build one afterwards.

But whether the uncertainty is from novelty or systemic complexity, a strategy to defend the innovation once the uncertainty is mitigated must be in place for the startup to create much excess value. This strategy must ask:

  • What are the productive uncertainties this business faces?
  • How will these uncertainties be mitigated?
  • When will they be mitigated, relative to product launch?
  • Will competitors be able to see that they have been mitigated (and/or how we mitigated them)?

Questions asked about things that are uncertain can not be answered, but these questions can be asked even if the eventual product and market are not entirely clear because the two types of uncertainties give such different answers. These answers should guide product and market strategy and be updated as these strategies progress. The goal of the strategy-making process is to determine the most protectable manifestation of the company’s innovation, which moats can be put in place, when they need to be in place, what needs to be done to build the moat, and how these activities can be intertwined with the designing, building, and commercializing of the product.

Many startups are founded based on a great innovation but end up failing because they never build a moat. Because entrepreneurs and investors in uncertain ideas can’t know for sure what the eventual product or market will be, they often resort to a “build it and they will come” mindset. This leads to valuing new technology and new market ideas equally. The real question is, “if we build it will they stay?” This analysis can be done in a rough sense before the exact product or market is known. It will create a much smaller set of possible successful outcomes, and this set can be used not only to rule out some ideas, but to shape ideas as they become companies so they are more likely to reach one of these goals.

While this process may favor new market companies over new technology companies, it won’t rule out all new technology companies! For those new tech companies it allows, it will help determine exactly what entrepreneurs need to do before product launch to make sure they can defend their company from imitation.

Cleantech

Let’s go back to cleantech. Where does the uncertainty come from? The MITEF report says “cleantech companies developing new materials, hardware, chemicals, or processes were poorly suited for VC investment” while “by contrast, cleantech companies developing software solutions were a better bet.”7 This distinction should sound familiar by now: the former are new technologies while the latter are probably primarily new markets.

The failed cleantech investments of the last wave were victims of novelty uncertainty. The lesson to learn is not that these companies were capital intensive, regulated, and in risky markets (though they probably were) but that they were competing with existing solutions. The report mentions that even those that navigated these uncertainties successfully still found no buyers willing to pay more than the money invested. When looking at a build vs. buy decision, these buyers certainly noticed that development is costly, imitation is cheap.

This does not mean that all cleantech (now rebranded climate tech) companies are bad venture investments. Celine Herweijer and Azeem Azhar’s paper for PWC, “The State of Climate Tech 2020” show that VC investments in this segment are growing again. How should investors avoid the mistakes of the past? “The ultimate solutions will require a broad range of approaches and novel business models to trigger behavioural changes in consumers and enterprises alike” and notes success stories like Tesla, Nest, and Beyond Meat.8 In other words, pursuing new markets and complexity uncertainty can pay off.

Tesla and Beyond Meat had some novelty uncertainty, but it is the complexity uncertainty that allowed them to weather competition. Compare Tesla’s ability to sell a car into the new market for fully-electric vehicles, where customers are judging the product on a different metric than they judge internal combustion engine cars, to Solyndra’s attempt to sell a new kind of solar panel into an already existing solar panel market. Tesla faced no real competition in their new market while Solyndra faced intense competition as soon as they introduced their new technology. Many successful companies had both novelty and uncertainty complexity to deal with, but the ones with complexity uncertainty had a better chance of succeeding because they had little competition before they built their moat. Tesla, like Apple Computer in the 1970s, built a great brand while incumbents waited for the uncertainty around the market to resolve.

New markets versus new technologies looks like the distinction here, but it’s a difficult and slippery distinction to make. The right way to look at this is to think about where the biggest uncertainties come from. New technologies seem to map to novelty uncertainty and new markets seem to map to complexity uncertainty, but instead of thinking about this distinction, think about the uncertainties themselves.

If the company has new technology ask the four questions a few paragraphs back, and judge whether there is a way for the startup to build a moat around their idea before competition comes to bear. Crucially, don’t assume a patent will protect them! Ask yourself if the tech fits into one of the few buckets that deters imitation.

If the company is creating or entering a new market, again ask the four questions. Figure out how uncertain the new market is, how uncertain it will remain after product launch, and whether the startup will have time to build a moat. Then figure out if the startup can shape the evolution of the new market to leave themselves with a moat when they launch.

You might be surprised when you go down this path at how useful the analysis is. It’s worth doing because having a moat when you transition to a fully competitive environment determines if you will have excess value or not.


  1. Gaddy, B., et al, “Venture Capital and Cleantech: the Wrong Model for Clean Energy Innovation”, MIT Energy Initiative, July 2016, MITEI-WP-2016-06, p. 9. 

  2. There’s probably already a term of art for this in the economics literature, but I don’t recall it. 

  3. Private equity invests in a spectrum of companies ranging from ‘someone with a bright idea’ to already large and profitable companies. People who call themselves Venture Capitalists invest along this spectrum, but the character of the problems investors face changes as they move from one end to the other. As always in my writing, when I talk about venture capital I am thinking of what others qualify as ‘early stage’ venture capital because that’s where the problems I think about are most acute. As you move along the spectrum, this analysis becomes less and less applicable. So if you’re Warren Buffet none of this probably applies to you at all, I wouldn’t know. 

  4. I should note that as with all theories on social phenomena, including business, this is not always true. There are plenty of ways to make money in finance. This theory only applies to making money by helping entrepreneurs create excess value. 

  5. For a discussion of complex systems see, for instance, Miller, John H. and Scott E. Page, Complex Adaptive Systems, Princeton University Press, 2007. 

  6. This sounds similar to effectuation, but effectuation takes the idea of system modification and tries to universalize it. I am instead taking uncertainty as the universal principle and saying system modification is one way to productively mitigate certain types of uncertainty. 

  7. MITEF, p. 2 

  8. Herweijer, C and A Azhar, “The State of Climate Tech 2020”, PWC, 2020, p.7