The terrain sets the rules (why not every decision is decided the same way)

Moats and speed

May 2018. On an earnings call, Elon Musk dismisses one of the most venerated concepts in business strategy: "moats are lame." What protects a company, he argues, isn't the barrier it built yesterday but "the pace of innovation": if your competitor innovates faster, the moat won't save you. Days later, at the Berkshire Hathaway annual meeting, Warren Buffett —who built one of the largest fortunes in history buying companies with moats— answers dryly: "I don't think he'd want to take us on in candy." He meant See's Candies, the confectioner Berkshire bought in 1972 and which has spent half a century defending its position without innovating much of anything.

The exchange went viral as a clash of egos. Read more carefully, it is something else: an argument about which world each business is standing in. In the world of premium candy —stable demand, a century-old brand, inherited loyalty— the moat works: yesterday's rules still hold tomorrow, and strategy consists of defending a position. In the world of electric vehicles in 2018 —technology shifting quarter by quarter, entrants on every side, regulation in motion— today's position guarantees nothing, and the only defense is moving faster than everyone else. Musk and Buffett don't disagree about strategy: they operate in different terrains, and each describes his own with precision.

That is the thesis of this second installment. In Part 1 we looked at the factory defects of the decision-making apparatus —biases that work exactly the same way in the analyst and in the CEO. But there is an error costlier than any bias, and it comes first: applying the right method in the wrong terrain. Analyzing exhaustively where analysis cannot produce the answer. Improvising where proven best practice already existed. Demanding the precise forecast where only the experiment teaches. The first question in any decision isn't "what do we do?" but "which terrain are we in?".

Four terrains, four ways to decide

The most useful map for that question was published by David Snowden and Mary Boone in 2007, in one of the most awarded articles ever to appear in the Harvard Business Review ("A Leader's Framework for Decision Making"). It distinguishes four terrains by the relationship between cause and effect —and therefore by what kind of answer exists at all.

The clear. Cause and effect are obvious to anyone: order processing, payroll, the checklist for opening a branch. There is the best practice —singular: one correct, known, documented way. You decide by sensing, categorizing and responding: see the situation, recognize which case it is, apply the rule. Management here is standardization, and delegation can be total.

The complicated. Cause and effect exist and are stable, but they aren't in plain sight: they have to be dug out with analysis or expertise. Designing a bridge, valuing a company, structuring the tax treatment of a regional transaction. There is no single right answer but good practices —several defensible solutions, and experts who can find them. You decide by sensing, analyzing and responding. This is the realm of the expert, the model and the plan: analysis, done well, pays.

The complex. Cause and effect are only visible in hindsight. The system has too many pieces adapting to one another —customers, competitors, employees, regulators— for any analysis to anticipate the outcome: merging two cultures, launching in a new market, a deep organizational change. There are no right answers waiting to be found; there are patterns that emerge. You decide the other way around: experiment first —small probes whose failure does no damage—, sense which pattern appears, and amplify what works. The detailed five-year plan isn't ambition here: it's a category error.

The chaotic. There is no discernible relationship between cause and effect: the fire, the bank run, the collapse of the system. Looking for the right answer is a luxury that doesn't exist; first you act to stop the bleeding, then you sense where there is some stability, and only then do you respond. Leadership is directive by necessity —and the danger is falling in love with that command mode once the crisis has passed.

Snowden and Boone add two warnings that are worth the whole article. The first: the dangerous edge isn't between the complex and the chaotic, but between the clear and the chaotic —complacency. Whoever believes his business is simple and orderly stops looking, and from there the cliff is a single step away (it happened to manufacturers who "knew" what their industry was like until it stopped being that way). The second: the expert's typical failure is entrained thinking: solving the new problem with the solution that made him successful on the old one —treating every terrain as if it were the one where he learned to win.

Complicated is not complex (and confusing them is expensive)

Of the four boundaries, one accounts for the majority of strategic errors: the one separating the complicated from the complex. Gökçe Sargut and Rita McGrath devoted their own article to it ("Learning to Live with Complexity", HBR, 2011), with an image worth memorizing: a commercial aircraft is complicated —thousands of parts, but fixed, predictable interactions: it can be modeled, tested and flown by checklist—; air traffic is complex —the same aircraft plus weather, controllers, passengers and airlines adapting to one another in real time: nobody predicts the state of the whole system. Three properties push a system toward complexity: multiplicity (how many elements interact), interdependence (how connected they are) and diversity (how different they are). A large organization, a market, a regional supply chain: all three dials at maximum.

Why does the distinction matter? Because in complex systems three managerial reflexes fail systematically —the same three that serve you well in the complicated:

The point forecast. In the complicated, more analysis converges on a better answer. In the complex, the information required for an exact forecast doesn't exist yet —it is created as the actors react. Demanding "the number" doesn't produce knowledge; it produces a theater of precision (and we already saw in Part 1 what the confidence intervals of the world's best CFOs are worth: a 36.3% hit rate).

The average. In the complex, events far from the mean are more frequent than statistical intuition expects —the tails are fat, as in the Flyvbjerg database from Part 1: the IT project doesn't overrun "a little", it overruns by 447% once it enters the tail. Managing a fat-tailed system with averages is driving by the rear-view mirror.

The single consequence. In an interdependent system, every intervention has second- and third-order effects nobody designed. Sargut and McGrath illustrate it with cases where a reasonable local decision —an incentive, a cut, a price change— cascades into global outcomes nobody wanted. The management question stops being "what is going to happen?" and becomes "what could happen, and how do I detect it early?".

For that terrain, the tools they propose are different ones: simulate scenarios including the low-probability, high-impact ones, instead of projecting the base case; decouple and add redundancy so that a local failure doesn't propagate (in the complex, efficiency and fragility are the same purchase); watch three baskets of information —lagging indicators, the present, and early indicators— instead of the accounts alone, which always arrive late; use counterfactuals and narrative ("what would have to happen for this to break us?") where statistics fall short; invest in real options —small, staged, reversible bets— instead of all-in commitments; and recruit cognitive diversity, because a team that thinks alike sees only one future.

In the clear and the complicated, boring pays

The map has a vanity trap: read quickly, it suggests that the clear and the complicated are minor terrain —a matter for operators and technicians— and that "real" leadership lives in the complex. The evidence says exactly the opposite: the most underexploited competitive advantage in the world is managing the clear and the complicated well.

Nicholas Bloom, Raffaella Sadun and John Van Reenen have spent two decades measuring, through the World Management Survey, the adoption of basic management practices —performance monitoring, targets with a coherent horizon, talent management— across more than 12,000 companies in 34 countries. Nothing exotic: are processes measured? are deviations corrected? do targets exist, and do they bite? are high performers promoted and retained? The results are brutal in both directions. The differences are enormous —and persistent—: moving from the worst-managed decile to the best is associated with 75% higher productivity, 25% faster annual growth and some US$15 million more in profit. In their academic work, the same authors estimate that these practices explain around 30% of total productivity gaps between countries —on the order of what capital or technology explain. And yet most companies don't adopt them. They found the most humbling reason simply by asking: managers rate themselves, on average, 7 out of 10 —and the correlation between the self-assigned score and measured quality is zero. (WYSIATI, Kahneman and Part 1 would say: the internal story feels complete.)

The title of their HBR article is a rhetorical question: why do we undervalue competent management? Because it isn't glamorous. Because it looks "operational" while strategy looks "executive". But in the clear and complicated terrains —where the rules are known and good practices exist— executing the boring things with discipline is the strategy, and it pays better than any episodic stroke of genius. Walter Kiechel called the twentieth century "the management century": a hundred years of accumulated knowledge about how to manage. The finding of the World Management Survey is that the century ended and most companies still haven't adopted it —which turns disciplined adoption, today, into competitive advantage.

In the complex, you experiment (and most ideas lose)

If in the complicated the unit of work is the analysis, in the complex it is the controlled experiment. And on what happens when an organization takes experimenting seriously, the most sobering data comes from those who practice it most. Ron Kohavi (who ran experimentation at Microsoft) and Stefan Thomke (Harvard) documented it: at Google and at Bing, only 10% to 20% of experiments produce positive results. At Microsoft overall, a third improve the metrics, a third move nothing and a third make them worse. Read that again: at some of the most capable organizations on the planet, most ideas —proposed by talented people, approved by competent managers— fail the reality test. That is why the big platforms each run more than 10,000 experiments a year: because they know they don't know which idea is the good one. And the prize for experimenting is real: one Bing experiment with ad formatting —an idea that had sat in the backlog for months because nobody thought it was a priority— produced 12% more revenue: over US$100 million a year in the United States alone.

The translation for companies that aren't digital platforms isn't "A/B test everything": it is adopting the logic of complex terrain. Pilot before rollout: one plant, one branch, one segment —with a comparison group and a success criterion written down beforehand (otherwise the coherent-story machine from Part 1 will read any result as a victory). Staged bets with exit options, instead of monolithic commitments. Reversibility as a design criterion: the reversible decision gets made fast and cheap; the irreversible one is the only kind that deserves the full heavy process. And a budgeted tolerance for small failure: if two out of every three Microsoft ideas improve nothing, what failure rate is implicit in your plan —and what does your culture do to the manager whose honest pilot came back negative?

Misreading the terrain (the three classic errors)

Treating the complex as complicated. The costliest and the most common. You recognize it by its symptoms: five-year plans with decimal places, "the number" demanded where there is no basis for any number, variance punished as though it were incompetence, and the conviction that the failure of the last plan is fixed by planning the next one more finely. There is a statistical version of this error that Kahneman illustrated with a famous anecdote: Israeli flight instructors were convinced that praise made cadets worse and criticism made them better —because after an excellent flight the next one tended to be worse, and after a disastrous one, better. It was pure regression to the mean: extremes tend to return to the average on their own, with neither the criticism nor the praise doing anything. In a system with randomness, raw feedback lies: it credits management with what is noise. The manager who "learns" from every monthly data point of a complex system is learning, in good measure, mythology.

Treating the complicated as clear. Skipping the analysis where analysis does have an answer: setting price by instinct when there are measurable elasticities, sizing the plant by analogy when demand can be projected, hiring "on feel" when there is evidence on how that selection is decided better (Part 3 brings those numbers, and they are uncomfortable). It is the inverse of the previous error, and it is usually committed by the successful founder: his intuition worked in the terrain where it was trained, and he exports it to terrains where there is no reason for it to work.

Staying in analysis when the terrain calls for movement. In the complex and the chaotic, the cost of additional information is time —and time is exactly what the terrain doesn't give. Analysis paralysis is, frequently, loss aversion dressed up as rigor (Part 1): asking for one more study is the socially acceptable way of not deciding. The opposite discipline has a name in some companies: telling reversible decisions from irreversible ones, and forbidding yourself the heavy process for the former.

The terrain, its question, its method and its risk. Snowden & Boone, HBR, 2007; Sargut & McGrath, HBR, 2011.

Questions for Monday

For your three most important open decisions:

The map leaves one question open, and it is the most personal of them all. In the complicated, experts rule; in the complex, experiments do. But what about gut feel? When is the intuition of someone with twenty years in the industry worth something —that immediate certainty, without arguments, that this candidate is the right one or this deal doesn't add up? And when is it better to submit to the humiliation of a three-variable formula forecasting better than the entire committee? That is what Part 3 is about: the exact conditions under which expert intuition is reliable, half a century of evidence on judgment versus algorithm, and the flaw in professional judgment almost nobody talks about —noise.