Uncategorized

Strategy under uncertainty

Every firm competing in an industry has a competitive strategy, either explicit or implicit…

The emphasis being placed on strategic planning today…reflects the proposition that there are significant benefits to gain through an explicit process of formulating strategy, to insure that at least the policies (if not the actions) of functional departments are coordinated and directed at some common set of goals.

Michael Porter, Competitive Strategy ((Porter, M., Competitive Strategy, 1980, New York: The Free Press, pp. ix and xiii.))

 

Strategy

Every company has to have a strategy. And if you need to create one, Porter and a thousand other business strategists tell you how. But every founder of a high-growth potential startup who has studied mainstream business strategy, including Porter, has realized at some point that’s there’s something missing. That Porter isn’t really talking to them.

Creating a strategy for a startup is fundamentally different than creating a strategy for a large, established company in a relatively stable market. The uncertainty that startups must have as their moat, the uncertainty that dominates so many aspects of starting, running, and planning in a startup makes the strategy taught in business schools around the world almost entirely beside the point. If we want to talk about strategy for startups we need to talk about a different kind of strategy than Porter’s. We need to pull apart the very idea of business strategy, get to its core, and then rebuild it with the presence of uncertainty baked in.

All of mainstream business strategy can be categorized as either:

  • Creating an objective picture of the world so the company can plan1 or position itself2;
  • Interpreting the world, since objectivity is not available3 or not feasible4; or
  • Deciding what the strategist wants to happen or can cause to happen, because even an interpretation of the world in any sense is impossible5 or can only be formed as the company learns over time6.

At the core of each of these types of strategy is an assumption about what you can know or control about the future. If you think that the current state of the world is objectively knowable then you build a strategy by determining what you want and picking the best path to get it. If you don’t think you can know the entire state of the world but believe you can infer the state of the world from what you do know, then you build a strategy based on what you predict, and may the best predictor win. If you don’t think you can infer the state of the world at all then you make a strategy based on what you can cause to happen.

These schools of thought, often wildly different in the conclusions they draw and processes the prescribe, all rely on the axiom that the results of your actions can be predicted. Perfect knowledge of the world implies that you can find a strategy through analysis of the data, however difficult this may be. A model of the world implies that a good model describes the world closely enough that you can use it much as you would use perfect knowledge. And while we tend to think of causing something to happen as being rather different than predicting something, things that you can cause to happen are, by definition, predictable.

Mainstream business strategy does not guarantee certainty, it acknowledges risk. But risky things are still predictable things. As we learned in Startups and Uncertainty, the difference between risk and certainty is nothing more than an insurance premium. Uncertainty, however, is different.

Peter Drucker, in The Practice of Management, says:

Management has no choice but to anticipate the future, to attempt to mold it and to balance short-range and long-range goals.

He follows this by saying

It is not given to mortals to do either of these two things well.7

This is the problem. Uncertainty means being unable to predict the future. Basing strategy formation on prediction means forgoing uncertainty, and uncertainty is one of the key things that enables entrepreneurs to succeed with a high-growth potential startup. William Gaddis in his “Strategy under attack”8 says

[I]nherent instability–‘chaos’–within complex systems influences the behaviour of the systems…applied to the modern corporation, [it tends] to cast considerable doubt upon the notion that any corporate system can be successfully guided toward a pre-conceived future status. Strategic managers, practising in the real world, have long known from first-hand experience that we all share a frail understanding about which causes lead to which effects. Yet these managers have persevered because of their inherited acceptance of western culture’s long-standing belief in the philosopher’s ‘chain of causality’, stemming from an earlier heritage. Now they are being told that random small events in the corporate system (when they ‘cumulate’ instead of ‘offset’) may have significantly more impact upon the future than planned events.

That is, because of complexity “It is no longer merely the methods and processes of strategic planning which are being critically assessed; now, clearly, the very concept of future-oriented management is under attack.” And while complexity is only one source of uncertainty, the basic insight holds: if the future can not be predicted, our received corporate strategy-making processes do not work.9

Google or Apple–companies that are established in their industries, and whose markets evolve in gradual steps–can predict most scenarios they will face. Uncertainty is a small part of their business. This is the audience Porter et al are writing for: MBAs and managers of established companies trying to carve out a sustainable niche for themselves in crowded, competitive industries. They have plenty of data and plenty of knowledge: they can use both induction and deduction to somewhat accurately predict the odds of any given future. Strategic planning of the traditional sort will work for them.

It won’t work for startups, companies doing something entirely new or in an entirely new market. But this obviously doesn’t mean startups can’t have a strategy. Startups do succeed, and they succeed partly based on their strategies. Apple’s success with the iPod and Facebook’s success in social networking were not accidental, they were the result of strategy. It just wasn’t the type of strategy Porter was thinking about.

If we reject mainstream business strategy, it is our obligation to replace it with something else, a theory of business strategy under uncertainty. To create this theory we need to break the concept of strategy itself down to its underlying meaning and then rebuild it with uncertainty in mind.

What Does ‘Strategy’ Mean?

Strategy seems to be one of those things that we can’t describe, though we know it when we see it. Every domain of endeavor has a strategy, and the concept remains intelligible even as it changes form from domain to domain:

  • Chess: One of the first taught strategies in chess is “Try to control the center of the board throughout the game.”
  • War: Sun Tzu in his Art of War, said “The good fighters of old first put themselves beyond the possibility of defeat, and then waited for an opportunity of defeating the enemy.”10 In other words, wait until your opponent gives you an opening.
  • Golf: “Start by looking at where a double-bogey is most likely to come from…identify the danger and adopt a strategy based on how best to avoid it.”11

These are all strategies, but differ from what we think of as business strategy in instructive ways:

  • Unlike business strategy, none of these strategies involves attempting to predict the future, so predicting the future can’t be axiomatic.
  • Strategy sometimes tells us how to anticipate and counter our opponents. War, business, and chess have opponents, and a large part of their strategy literature is dedicated to figuring out how they will respond to any given action so further actions can be planned. This often results in a game-theoretic approach.12 But golf, for one, does not have an opponent that is actively trying to thwart you.13
  • War and business require delegating action, so a large part of their strategy requires effectively communicating the plan, trusting it to be executed or modified it in a sensible way, receiving feedback and information from the front lines, and expecting that what you want to happen will inevitably not be exactly what happens. But delegation is not required in chess or golf.
  • War, business, and golf strategies also take into account factors outside of the strategist’s immediate control. A strategy of waiting in war requires a knowledge of the timing and amount of supplies coming in, for instance; businesses must think of the political or social ramifications of their actions; and the dangers golfers face change with the weather. Chess, on the other hand, is a closed world: two players and a chess set.

Etc. Most of the things people say strategy is are not part of strategy in at least some domains. These things, then, can’t be a prerequisite to the concept of strategy itself, they are things bolted on to the core idea of strategy to tailor it to the domain. The core idea, the one thing each of these strategies has in common is that they prescribe how to make decisions in situations that are not yet known. This is what strategy means. These strategies in golf, war, and chess tell you how to make decisions, but they can’t tell you what decision to make ahead of time. They are decision-making frameworks, to be applied when their required inputs becomes available.

Business strategy is the same, it is a decision-making framework for future decisions. Mainstream business strategy takes the core idea of strategy and tailors it with the assumption of predictability to make it useful for predictable businesses. We will have to take the core idea of strategy and tailor it with the assumption of uncertainty to make it useful for startups.

This work is, as social scientists say, normative, not descriptive. That is, it will talk about how decisions under uncertainty should be made, not how they generally are made. The distinction is important, and not just because most of the work done on decision making under uncertainty is descriptive.

People make decisions under uncertainty all of the time, and people who study decision making have documented how they do. Some of their strategies, like satisficing, aim to limit the time spent making an unmakeable decision by stopping the decision process not when the decider has reached the best solution, but when they have reached a good enough solution.14 Others, like sensemaking, aim to justify a decision by organizing the available facts in a way that supports it. Decisions are also often negotiated among different groups in an organization in an effort to move from talk to action (with the negotiated action being left just vague enough that all sides can think they have prevailed.) Some decisions are even made by leaders to reinforce the ethos of an organization: rather than making decisions based on what is, they make decisions based on what is supposed to be.

You’ve probably seen some of these decision making strategies in use, and may have even thought the decision maker was making obviously bad decisions. But these strategies are not irrational, they just prioritize goals other than making the objectively ‘optimal’ decision. They might even be better decisions than the ‘optimal’ given the totality of the decision making environment.

Getting from descriptive to normative can be done in one of two ways. The first way is inductive: show that the observed strategies are successful. This is difficult to do. Even if successful decisions are made in a certain way, and people who make decisions that way are often successful, there is no control group of purely rational decision makers to compare them to, so the evidence is subject to every cognitive bias under the sun. This is, nevertheless, the foundation of most business books: here is how I have framed how a successful leader made decisions, so this must be the right way to make decisions.

The other way is to start with a purely rational decision mechanism and build in the possibility of uncertainty at each step to come up with a deductive theory of how to make decisions under uncertainty. Then we can not only understand the above ‘irrational’ decision processes, and many others of their ilk, but understand when and how they should be used. This is what I am going to attempt to do. I warn you, it will be a long road.

Rational Decision Making

From James March’s Primer on Decision Making:

A rational [decision-making process] is one that pursues a logic of consequence. It makes a choice conditional on the answers to the four basic questions:

  1. The question of alternatives: What actions are possible?
  2. The question of expectations: What future consequences might follow from each alternative? How likely is each possible consequence, assuming that alternative is chosen?
  3. The question of preferences: How valuable (to the decision maker) are the consequences associated with each of the alternatives?
  4. The question of the decision rule: How is the choice to be made among the alternatives in terms of the value of their consequences?15

We’ve all seen this procedure in the guise of maximizing expected utility. (Utility meaning the value of that result to you. I will use value and utility interchangeably here.) For each possible action, you figure out what the possible consequences, or results, could be. You then place a numeric value on each result. You then determine the probabilities of each action leading to each of its possible results. This allows you to calculate the expected value of each action. You then take the action with the highest expected value. Here is a graphic representation, with three possible actions, and six possible results. The highest expected value comes from taking Action 1.

March’s more generalized framework is really just this. The rational decision making process is

  1. List all the actions you could take and, for each,
  2. Figure out what might happen and how valuable it is to you,
  3. How likely it is that each of these will happen if you take the action, and
  4. Decide how you will choose among the actions.

I assumed above the decision rule was ‘take the action with the highest expected value.’ This is not the only possible decision rule. Action 1, for instance, has not only the highest expected value but the highest variance in outcomes (there is a ten percent change of a utility 100 outcome but also a 20% chance of a utility 0 outcome) and you may decide to take a less risky action, like Action 2 which, while having a lower expected value, has no chance of resulting in zero utility.

Building a company involves decision after decision and this rational decision making process involves just one. But if you think of each ‘result’ being the starting point for the next decision, you can link these into a tree of decisions.

If you could predict the future–if you knew which branch you would take at each step–you would have more than a strategy, you would have an airtight plan, from first action to final result. Of course, in a risky world you can’t know ahead of time which end result you will get so airtight plans aren’t available. But even taking risk into account the strategy of making rational decisions will get you the best possible outcome on average. This is a perfectly viable strategy.

As an aside, I am using the words decision and action interchangeably and this is wrong. As Mintzberg and Waters have noted16: “we made the implicit assumption that decisions inevitably preceded actions, that if an organization did something, then it must have previously decided to do so…another interpretation is possible: that decisions are difficult to uncover because sometimes they don’t exist, in other words, that the relationship between decision and action can be far more tenuous than almost all the literature of organization theory suggests.” That is, a decision can be made without acting and an action can be taken without deciding (or at least, without a formal decision.) In this context, though, where we are trying to come up with optimal decision rules, we are going to assume that decisions lead to actions. The organizational dysfunction that makes that sometimes untrue has been the subject of study since at least Herbert Simon’s Administrative Behavior, and we have enough on our plate without taking on that additional complication.

Uncertainty Screws Up Your Perfectly Viable Strategy

The rational decision making model works when you know what is going to happen, and it works when there is risk. But when there is uncertainty it starts to fall apart. Every piece of information we use to construct and compute the model is subject to uncertainty, from the actions to the outcomes to the probabilities to the utilities. The more uncertainty there is, the less we know about each of these things. This is especially acute when planning strategy for high growth potential startups.

Startups often can’t know what the results of a specific action will be. When Bill Gates started Microsoft in Albuquerque in 1975 to develop a BASIC interpreter for the Altair, he could not have known that a result of his actions would be Microsoft becoming the sole supplier of operating systems to IBM several years later. Microsoft had no intention of designing operating systems, IBM was not in the personal computer business, and even if they were, it would have been inconceivable that they would outsource the operating system to a small, unknown company. “What will happen if we make this decision?” is often unanswerable in a startup not just because you may not know which result is likely but because you may end up with an outcome you never even imagined. How can a startup set goals if it doesn’t even know what is possible?

Even if you have a goal, it can be impossible to know the likelihood that it will result from your actions. Apple’s goal in 1976 was to have a personal computer in every household in America. How likely was this? Any probability they guessed was just a number pulled out of the air. There was simply no way of knowing.

Startups might not know all of the actions they could possibly take. Decision-makers usually lay out their options before making a decision by surveying the means at hand, their resources. They wrack their brains to brainstorm more options, and come up with new ways to use the means they have. But there are always possible actions they don’t know are available to them. In the early years of Google, the founders and management and their investors tried every thing they could think of as a business model for their wildly popular search engine. They rejected traditional banner advertising, they tried to license the technology, they even tried to sell the technology outright. It wasn’t until Bill Gross launched GoTo.com that they found the action they had simply not thought of before: pay-per-click advertising, which neatly aligned the incentives of the searcher (to find relevant content) and the business (to show relevant content.) Many businesses succeed by doing something others had not thought of, and many probably fail because they could not come up with the actions that would have saved them.

The last way uncertainty enters the decision process is harder to see on the diagrams above, because these static diagrams don’t capture change: even if you could know the best decision to make, the factors that lead to that decision constantly change. Startups operate as part of a complex system that encompasses not just their internal operations, but their customers, their suppliers, other companies that might compete or cooperate with them, financiers, the media, the government, and society at large. Each of these other entities also makes decisions, and the results of their decisions must factor into the startup’s decision model. The changes most likely to affect a startup are the ones that happen as a result of the decisions the startup itself makes, a complex feedback loop. Uber, for instance, decided to use some morally questionable practices in secret to advance their company. When these practices came to the attention of the media, the legal and social pushback from the public and governments changed the consequences of these decisions, and many others Uber made, in ways no one there seems to have predicted.

The rational decision model is predicated on having good inputs, and uncertainty makes this impossible. And if uncertainty makes rational decision making impossible, it also makes strategy based on rational decisions impossible. Now obviously, startups do have strategies, and these strategies often succeed. Some of them are related to the ‘irrational’ strategies we described above–satisficing, negotiation, sensemaking, etc.–and some of them are the ones founders have come to know and love over the years–scenario planning, iterative development, ‘lean’, customer development, etc. Some of them we can bring over from other disciplines that also face severe uncertainty. All of these strategies rely on a decision making framework somewhat different than the one we decribed. This new framework assumes that means, goals, and causality are all subject to substantial uncertainty and unpredictable change. It shows ways to mitigate, compensate for, or work around this. And understanding it will explain why and under what circumstances each of these strategies work so an entrepreneur can decide which of them make sense for their company.

That will be the work of the next few posts.


  1. Ansoff, Rumelt. 

  2. Porter, op cit, and even the ‘vision’ strategists, who think about high-growth potential startups but seem to ignore the Knightian conception of entrepreneurship. 

  3. Simon 

  4. Tversky, Kahneman and the behavioral economics school 

  5. Mintzberg, any strategy book with Power in the title 

  6. Weick, Pralahad & Hamel 

  7. Drucker, Peter, The Practice of Management, … 

  8. Gaddis, PO. “Strategy under Attack.” Long Range Planning, vol. 30, no. 1, 1997, pp. 38–45, doi:10.1016/S0024-6301(96)00094-5. 

  9. And even Gaddis’s attempt to transcend the strategic thinking of his time is mired in the idea that strategy tells you what to do. 

  10. Sun Tzu, trans. Giles, Lionel, The Art of War, London: Luzac & Co., 1910, p. 26. 

  11. Tappin, Neil, “Ten Ways to Improve Your Golf Strategy”, Golf Monthly, 6/15/2018, https://www.golf-monthly.co.uk/tips/10-ways-improve-golf-strategy-158462 

  12. Reflected in the title of a classic work on game theory: Strategy in Poker, Business, and War by John McDonald. 

  13. It just seems like that. 

  14. Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2), 129–138. https://doi.org/10.1037/h0042769 

  15. March, JG, A Primer on Decision Making, Free Press: New York, 1994, pp. 2-3. 

  16. Mintzberg, Waters, “Studying Deciding: An Exchange of Views Between Mintzberg and Waters, Pettigrew, and Butler”, Organization Studies, 1990, 11/1, pp. 1-16.