Evolutionary Game Theory for Marketing Strategy: Why Categories Converge on One Playbook, and When Deviating Pays

TL;DR: Every software category converges on the same homepage, every direct-to-consumer brand runs the same playbook, and the standard explanation, that marketers lack imagination, is wrong. Strategy spreads by imitation of what appears to work, which is replicator dynamics: whatever earns above-average payoff grows its share of the population, and reasoning never enters. Maynard Smith and Price (1973, Nature 246) gave the tool, the evolutionarily stable strategy, a state that no rare alternative can invade. Applied to markets it predicts three things positioning frameworks miss. First, convergence: Hotelling (1929) showed why competitors crowd the center, and d'Aspremont, Gabszewicz and Thisse (1979) showed the one condition under which they separate instead. Second, cycles: premium, value, and convenience positionings can chase each other in a rock-paper-scissors loop, the same non-transitive dynamic Sinervo and Lively (1996, Nature 380) documented in lizards over a six-year period. Third, and most usefully, the payoff to a contrarian position is frequency-dependent: it depends on how many others are already deviating, the one variable "be different" advice ignores. Deviation pays only in a narrow window, when the deviant is rare, the niche has real payoff, imitation is slow, and the position rests on a commitment rivals cannot copy fast. Outside that window, the crowd is right and being different is a way to lose money confidently.

---

Open the Ten Best-Funded Sites in Any Category

Open the homepages of the ten best-funded companies in any software category and read only the top fold. The hero sentence has the same shape in all ten: the noun phrase for the product, the word "platform," the audience, a verb about speed or scale. Below it sits the same triptych of three feature cards, the same horizontal strip of customer logos in gray, the same gradient, the same pair of buttons where the left one says start free and the right one says book a demo. Scroll and the sections arrive in the same order. The typeface is one of four. A visitor who lost the tab could not tell you which company she had been reading.

The same convergence repeats in consumer goods, despite entirely different surface details. A direct-to-consumer brand in almost any category now arrives with a lowercase sans-serif wordmark, a founder paragraph about a problem nobody had solved, a claim that the product is clinically studied or dermatologist-tested or third-party-verified, a muted palette of one dusty color, and a subscribe-and-save toggle set to on. Yet the categories differ, while the playbook does not.

The comfortable explanation is that the people writing this copy are unimaginative, or lazy, or copying each other out of cowardice. The explanation feels good, yet predicts nothing. We should be suspicious of any account of a persistent, industry-wide pattern that reduces to a character flaw shared by thousands of independent professionals, because the pattern is too regular and the professionals are too numerous for a personal failing to be the cause. Something structural is producing it.

The structure is evolutionary. Strategies in a market spread the way traits spread in a population: not because anyone reasons their way to them, but because the ones that appear to work get copied, and the ones that appear to fail get abandoned. A homepage layout that seems to convert gets cloned by the next founder who studies the category. Over enough rounds of copying, the category converges on whatever configuration resists being displaced by an alternative. That configuration has a name in biology, the evolutionarily stable strategy, and the mathematics that governs how it takes over, replicator dynamics, was worked out for animals in the 1970s. Marketing categories obey the same equations because they run the same process.

This essay makes three claims that the standard "differentiate or die" literature cannot. Convergence is predictable, not pathological. Positioning cycles, so the winning strategy this year is a moving target rather than a fixed peak. And the payoff to breaking from the herd is frequency-dependent, meaning it depends entirely on how many others have already broken from it, which is why "be different" is good advice only under conditions that most of the people repeating it never state.

The Logic of Animal Conflict, and Why It Is About Marketing

The founding paper is about deer, not demand generation. John Maynard Smith and George Price published "The Logic of Animal Conflict" in Nature (1973, 246, 15-18) to solve a puzzle in ethology: why animals fighting over a resource so often display, posture, and back down rather than escalating to injury. Group-selection stories were popular and loose. Maynard Smith and Price replaced them with a game.

The move that made the theory portable was to drop the assumption that anyone reasons. Classical game theory, in the tradition of Nash, assumes players compute best responses to each other's strategies. Animals do not compute. What Maynard Smith and Price showed is that you get equilibrium-like behavior anyway, provided strategies are heritable and the ones that pay reproduce faster. The equilibrium concept they needed was stronger than a plain Nash equilibrium: it had to be stable against invasion by rare mutants, because in a population the test of a strategy is not whether it is a best response but whether it can be displaced.

Formally, a strategy σ* is an evolutionarily stable strategy when, against a population playing σ*, no alternative σ does strictly better, and in the knife-edge case where an alternative ties, σ* does strictly better against that alternative than the alternative does against itself. Writing u(σ, σ') for the payoff to playing σ against an opponent playing σ', the condition is:

$$u(\sigma^{*}, \sigma^{*}) \ge u(\sigma, \sigma^{*}), \quad \text{and if } u(\sigma^{*}, \sigma^{*}) = u(\sigma, \sigma^{*}) \text{ then } u(\sigma^{*}, \sigma) > u(\sigma, \sigma).$$

The first clause is Nash: the incumbent strategy is a best response to itself, so a mutant cannot earn more by facing the crowd. The second clause is the evolutionary addition, and it is the one that matters for markets. When a mutant does exactly as well against the crowd as the crowd does against itself, the tie is broken by how the two do against the mutant. If the incumbent handles the rare mutant better than the mutant handles itself, the mutant cannot gain a foothold, and the incumbent is stable. A position can be a perfectly good best response and still not be evolutionarily stable, which is the gap that most competitive-strategy advice falls into.

The hawk-dove game

The workhorse example, and the one worth carrying into every category review, is hawk-dove. Two animals contest a resource worth V. A Hawk escalates; a Dove displays and yields. When two Hawks meet they fight, one wins V and the other pays an injury cost C, so each earns (V - C)/2 on average. A Hawk against a Dove takes the whole resource, V, while the Dove gets nothing but walks away unhurt. Two Doves split the resource, V/2 each.

Read the matrix as a category. Hawk is the aggressive posture: undercut on price, escalate ad spend, match every feature, contest every keyword. Dove is restraint: hold price, stay in your lane, let the contested customer go. The interesting question is which posture is stable when everyone in the category is choosing, and the answer depends on one ratio.

If the prize is worth more than the fight costs, V > C, Hawk is a dominant strategy and the stable state is all-Hawk: everyone escalates, because escalation pays even against another escalator. That is a price war, and it is stable in the grim sense that no single firm can stop fighting unilaterally without losing the contested customer. But when the cost of fighting exceeds the value of the prize, C > V, neither pure posture is stable. A population of all Hawks is invadable by a Dove, because Hawks facing Hawks bleed out (V-C)/2 < 0 while a Dove at least earns zero. A population of all Doves is invadable by a Hawk, because a lone Hawk among Doves takes V every time. The stable state is a mixture.

When the stable state is a mixture

The mixed evolutionarily stable strategy is the point where a Hawk and a Dove earn the same expected payoff, so neither posture can grow at the other's expense. Solving that indifference condition for the hawk-dove payoffs gives a clean result:

$$p^{*} = \frac{V}{C}, \qquad \text{valid when } C > V,$$

where p* is the equilibrium fraction of the population playing Hawk. The intuition is worth more than the algebra. The more damaging the fight relative to the prize, the smaller the stable share of aggressors. Aggression is self-limiting: it is profitable only as long as it is rare, and it stops being rare exactly when it stops being profitable.

The curve above is the equilibrium the model implies, plotting p* = V/C as the cost-to-value ratio C/V climbs from 1 to 5. When fighting costs twice the prize, only half the category should be aggressive; when it costs five times the prize, only a fifth. A category manager who feels that the whole market is irrationally aggressive is usually looking at a market where C/V is low, which means the aggression is not irrational at all. Changing the mix requires changing the ratio, by raising the cost of fighting or lowering the value of the contested prize, not by appealing to anyone's restraint.

An evolutionarily stable strategy is one that, if adopted by most of a population, no rare alternative can do better against.

, John Maynard Smith, Evolution and the Theory of Games (1982), paraphrased

Maynard Smith's 1982 book, Evolution and the Theory of Games (Cambridge University Press), generalized the hawk-dove result into a full apparatus and is the source most worth reading first. The lesson we carry forward is that stability is a property of the whole population, not of any one player's cleverness, and that the stable state is frequently a mixture rather than a single winning move.

Strategy Spreads by Copying, Not by Reasoning

The ESS tells us where a population comes to rest. Replicator dynamics tells us how it gets there, and that is the part that maps onto marketing, because the mechanism is imitation.

Peter Taylor and Leo Jonker gave the dynamics its canonical form in "Evolutionary Stable Strategies and Game Dynamics" (1978, Mathematical Biosciences 40, 145-156). The equation says that the growth rate of a strategy's share is proportional to how far its payoff sits above the population average:

$$\dot{x}_i = x_i \left[\, f_i(x) - \bar{f}(x) \,\right], \qquad \bar{f}(x) = \sum_j x_j\, f_j(x),$$

where xᵢ is the fraction of the population using strategy i, fᵢ(x) is its payoff given the current mix, and f̄(x) is the average payoff across everyone. Strategies that beat the average grow their share; strategies that trail it shrink; and a strategy exactly at the average holds steady. No player maximizes, forecasts, or even understands the game. The share of a strategy rises purely because it is doing better than what surrounds it.

Substitute "founders" for "organisms" and "copying a competitor's playbook" for "reproduction" and the equation describes a market. A positioning that appears to be working attracts imitators, and the rate of imitation scales with how far ahead the position looks. Jan Weibull's Evolutionary Game Theory (1995, MIT Press) is the rigorous treatment of why this is more than a metaphor: the replicator equation arises from several plausible micro-level copying rules, including "imitate the most successful firm you can observe" and "switch to a strategy in proportion to how much better it is doing," which is exactly how founders and marketers actually behave. We do not need marketers to be biological to get biological dynamics. We only need them to copy what works, which they demonstrably do.

Invasion, and the shape of a bandwagon

The dynamics also tell us what a successful deviation looks like on the way up, and the shape is instructive. Consider a category dominated by a consensus playbook, into which one firm introduces a genuinely different position, a rare mutant at 2 percent of the category. Suppose the deviant earns a payoff of 5 whenever it faces a consensus firm, against the 4 that consensus firms earn against each other, but that two deviants competing head to head earn only 1, because they are now fighting over the same small pocket of demand. Those payoffs make an anti-coordination game with an interior stable mix, and the replicator equation traces the path.

The trajectory is the path the replicator equation produces for those payoffs, integrated numerically from a 2 percent start. The deviant explodes early, when it is rare and every encounter is against the consensus it beats, then decelerates as copycats arrive and more of its encounters are against other deviants, and settles at a stable 40 percent. The window of extraordinary return is the first few generations. By the time the position is common it is no longer special, and the last firm to copy it earns exactly what the consensus earns. Being early is not a nicety here; it is the entire source of the excess payoff, and the math says so.

Martin Nowak and Robert May added a wrinkle that matters for how fast this happens. In "Evolutionary Games and Spatial Chaos" (1992, Nature 359, 826-829) they put the players on a grid so that each interacts only with neighbors, and found that strategies which go globally extinct under well-mixed dynamics can persist forever in local clusters, generating perpetually shifting spatial patterns. The market analogue is that a position wiped out at the level of the whole category can survive indefinitely in a region, a vertical, or a community, because imitation is local before it is global. A strategy does not have to win everywhere to be worth holding somewhere.

Why Everyone Ends Up in the Same Place

Replicator dynamics explains how a strategy spreads once it exists. It does not by itself explain why the strategy the market converges on is so often the bland one in the middle. For that we need the oldest result in the theory of location, and it predates evolutionary game theory by four decades.

Harold Hotelling published "Stability in Competition" in the Economic Journal (1929, 39(153), 41-57) with an example that has outlived every other part of the paper. Two vendors sell an identical product to customers spread evenly along a beach. Each customer buys from the nearer vendor. Where do the vendors stand? The intuitive answer, one at the quarter mark and one at the three-quarter mark, minimizing everyone's walk, is not an equilibrium. Whichever vendor is nearer the center can capture more of the beach by edging toward the middle, and the other must follow. The only stable configuration is both vendors back to back at the exact center, splitting the beach, having erased every difference between them. Hotelling called it a tendency toward minimum differentiation, and it is the formal reason two gas stations sit on the same corner, two news networks converge on the same centrist script, and two software homepages become indistinguishable.

The generalization to marketing is direct. The "beach" is any dimension of positioning along which customers have a spread of preferences: price, from budget to premium; tone, from playful to serious; scope, from focused to all-in-one. Wherever customers distribute themselves along such a line and buy from the nearest option, competitors are pulled toward the center of the distribution, because the center is where the most contestable customers are. The result is a category of near-clones clustered on the median preference, each terrified to move because moving cedes the center to the rival.

The condition was pinned down fifty years later, and it is the correction every strategist should know alongside the original. Claude d'Aspremont, Jean Gabszewicz and Jacques-François Thisse, in "On Hotelling's Stability in Competition" (1979, Econometrica 47(5), 1145-1150), showed that Hotelling's own math contained an error: with linear transport costs, no price equilibrium exists when the firms are close together, so the clean minimum-differentiation story does not go through. Fix the model by making the cost customers bear for a poor match rise with the square of the distance, quadratic transport costs, and the result flips. Firms now maximally differentiate, fleeing to opposite ends of the line, because standing close to a rival triggers ruinous price competition that both would rather avoid.

The two results are not a contradiction; they are a switch, and the position of the switch is the cost customers pay for a bad match. When a mismatch is cheap to the customer, so that a budget buyer will grudgingly accept a premium product placed next to a budget one rather than search further, firms crowd the center and the category converges. When a mismatch is expensive, so that a customer forced onto a poorly fitting option suffers badly and price competition between neighbors turns vicious, firms separate and the category spreads out. Convergence and differentiation are both equilibria of the same model under different values of one parameter, which tells us precisely what to measure before advising anyone to be different: how much a customer actually suffers from buying the wrong-fit product.

The Sociology of the Herd

Economists model imitation as a payoff-maximizing response. Sociologists studied why firms copy each other even when the payoff is unclear, and their answer fills in the micro-mechanism that the replicator equation abstracts away.

Paul DiMaggio and Walter Powell, in "The Iron Cage Revisited" (1983, American Sociological Review 48(2), 147-160), argued that organizations in a field grow more similar over time through three pressures, and named the one that concerns us: mimetic isomorphism, the copying of other organizations as a response to uncertainty. When a firm cannot tell what will work, the cheapest and most defensible move is to do what the visibly successful firms are doing. Nobody is fired for building the homepage that every funded competitor already has. Unlike a lapse of effort, the copying is a rational response to genuine uncertainty about cause and effect, and in practice it produces convergence as a byproduct.

Marvin Lieberman and Shigeru Asaba sharpened the question in "Why Do Firms Imitate Each Other?" (2006, Academy of Management Review 31(2), 366-385) by separating two distinct engines of imitation. Information-based imitation happens when a firm copies another because it believes the other has better information, which is DiMaggio and Powell's uncertainty response and tends to make firms herd onto whatever the perceived leader does. Rivalry-based imitation happens when a firm copies another to hold its position and limit the rival's advantage, matching a competitor's move so as not to be left exposed. The two have different signatures. Information-based imitation clusters the whole field on the leader; rivalry-based imitation produces the tit-for-tat matching of feature releases and price moves among close competitors. Knowing which one is driving your category tells you whether the convergence flows from a single admired exemplar or from mutual fear among peers, and the fix differs accordingly.

Underneath both sits a harder claim from organizational ecology. Michael Hannan and John Freeman, in "The Population Ecology of Organizations" (1977, American Journal of Sociology 82(5), 929-964), argued that firms change far less than managers imagine, because structural inertia and selection, not adaptation, are what shape a population of organizations over time. Contrary to the adaptation story managers prefer, markets do not mostly get more efficient because incumbents cleverly adjust; they get more efficient because the misfit firms are selected out and replaced. The claim is starkly evolutionary, and it changes how we should read convergence. When we see a category full of look-alike firms, we are not necessarily watching a room full of imitators. We may be watching the survivors of a selection process that already removed the ones who positioned differently and were wrong. Which of those two stories is true matters enormously for whether we should join the herd or break from it, and it is the question the next two sections exist to answer.

Distinctiveness Is Not Differentiation

Marketing science has its own quarrel with "be different," and it is precise enough to resolve part of the convergence debate directly. Byron Sharp's How Brands Grow (2010, Oxford University Press) argued, against a generation of positioning orthodoxy, that brands in a category are bought by largely the same customers for largely the same reasons, that perceived differentiation is far weaker and rarer than marketers believe, and that brands grow chiefly by being easy to notice and easy to buy rather than by occupying a unique position in the customer's mind. If Sharp is right, much of the differentiation the convergence critics demand is a solution to a problem customers do not have.

The distinction Sharp and Jenni Romaniuk draw in How Brands Grow Part 2 (2016, Oxford University Press) is the one that dissolves the confusion. Differentiation is being meaningfully different in what you offer, a reason to prefer you. Distinctiveness is being recognizably yourself in how you show up, the colors, logos, characters, and assets that let a buyer find you fast. Their evidence is that differentiation is hard to achieve and rarely decisive, while distinctiveness is achievable and matters a great deal, because most purchases are low-attention and the brand that is easiest to recognize and retrieve wins the moment. Convergence in what firms offer is, in practice, often fine; convergence in how they signal identity is the actual mistake. The ten identical homepages are not failing because their offers are similar. The homepages fail because they are visually interchangeable, so that no amount of exposure builds a recognizable asset a buyer can retrieve later.

The double jeopardy law, formalized by Andrew Ehrenberg, Gerald Goodhardt and Patrick Barwise in "Double Jeopardy Revisited" (1990, Journal of Marketing 54(3), 82-91), is the empirical backbone of Sharp's argument and a direct constraint on deviation strategy. Small brands do not just have fewer buyers; those buyers also buy them slightly less often, so market share and loyalty rise and fall together rather than trading off. The uncomfortable implication for the would-be deviant is that a distinctive niche position does not buy the compensating loyalty that positioning folklore promises. There is no hidden reservoir of devotion that a small, differentiated brand can draw on to offset its small reach. A deviation has to pay through the reach it wins, not through a loyalty premium that the data says is not there. We treat this in more depth in the portfolio framing of brand versus performance investment, where the same tension appears as a risk-return trade-off.

The Frequency-Dependent Payoff to a Contrarian Position

Now we can state the thesis precisely, because we have the pieces. The payoff to a contrarian position is not a fixed property of the position. Instead, it is a function of how many other firms have already taken it. The frequency of the deviation is the variable positioning frameworks systematically omit, and omitting it is why so much differentiation advice is simultaneously true and useless, contrary to how it is usually taught.

Return to the two-strategy game behind the invasion chart. Let q be the share of the category already running the deviant position, with payoffs a when deviant meets deviant, b when deviant meets consensus, c when consensus meets deviant, and d when consensus meets consensus. The payoff advantage of deviating over conforming, as a function of how crowded the deviation already is, is linear in q, and it defines the invasion window past which deviating stops paying:

$$\begin{gathered} \pi_D(q) - \pi_C(q) = (b - d) - \big[(b - d) - (a - c)\big]\, q, \\[4pt] q^{*} = \frac{b - d}{(b - d) - (a - c)}, \qquad \text{deviating pays} \iff q < q^{*}. \end{gathered}$$

The first term, b - d, is the edge a lone deviant enjoys when the position is empty: how much better it does against the consensus than the consensus does against itself. The bracketed coefficient is the rate at which that edge erodes as imitators arrive, and q* is the frequency at which it reaches zero. With the numbers behind the earlier chart, b - d = 5 - 3 = 2 and a - c = 1 - 4 = -3, so the edge when rare is +2 and it vanishes at q* = 2/5 = 0.4. A firm that adopts the deviation while fewer than 40 percent of the category has done so earns more than a conformist; a firm that adopts it once the deviation is more crowded than that earns less. The same position is a good idea and a bad idea at different frequencies, and nothing about the position itself changed. Only its prevalence did.

The essence of strategy is choosing what not to do.

, Michael Porter, What Is Strategy? (1996)

Michael Porter's "What Is Strategy?" (Harvard Business Review, November-December 1996) is usually read as a defense of differentiation, but its actual content is sharper and fits the frequency-dependent picture exactly. Porter's claim is that strategy is a set of trade-offs, activities that are valuable precisely because they are incompatible with the activities rivals have chosen, so that a rival cannot copy your position without abandoning its own. A trade-off is what makes a deviation slow to imitate, which is what keeps q low, which is what keeps the payoff positive. Porter and the replicator equation are saying the same thing in different languages: a differentiated position is worth having only to the degree that copying it is costly, because copying is what closes the window.

When deviating pays

The model gives four conditions, and they are testable rather than inspirational. Deviation pays when the deviant is rare, so that q sits well below q* and most encounters are against the consensus the deviation beats. Deviation pays when the niche has real payoff, meaning b - d is genuinely large because a distinct pocket of demand actually prefers the deviant offer, not merely a repackaging of the same offer. In practice it also pays when imitation is slow, because a low copying rate keeps q below the threshold for longer and stretches the interval of excess return; Porter's trade-offs, proprietary assets, and hard-won capabilities all slow the copying. And it pays when the position rests on a commitment rivals cannot match quickly, which is the costly-signaling logic: a deviation backed by an expensive, hard-to-reverse investment both convinces customers it is real and deters imitators who would have to make the same investment to follow. A deviation that any rival can copy by Friday is not a strategy; it is a preview.

When it does not

The symmetric conditions are the ones the differentiation literature almost never voices, and they are where most contrarian bets actually die. Deviation does not pay when the crowd is right, when the consensus position is a genuine ESS because it is a best response that also resists invasion, so that b - d is zero or negative and the lone deviant simply earns less. Deviation does not pay when the niche is empty for a reason, when the pocket of demand you imagine is not there, or is there but too small or too poor to support a firm, which the double jeopardy law warns is the common case. And deviation does not pay when imitation is instant, because a position anyone can copy overnight sends q past q* before you have recouped the cost of pioneering it, leaving the fast followers to split a payoff you paid to discover.

W. Chan Kim and Renée Mauborgne's Blue Ocean Strategy (2005, Harvard Business School Press) is the popular case for deviation, the argument to create uncontested market space rather than fight in crowded waters, and the frequency-dependent frame explains both why it is right and where it misleads. Creating an empty niche is exactly moving to a region of the strategy space where q = 0 and the edge b - d is large, which is where the payoff is highest. The blue ocean is real. What the book underweights is the survivorship problem we flagged earlier: the celebrated blue-ocean cases are the deviations that found a niche with real payoff and defended it long enough to matter, and we do not observe the far larger number of firms that sailed into an empty ocean because it was empty for a reason, and drowned there quietly. A strategy of seeking uncontested space is a strategy of maximizing b - d while assuming the niche exists, and the assumption is the whole risk. The mapping to a real, underserved job to be done, rather than an imagined one, is where that risk is actually managed, which is the discipline behind jobs-to-be-done segmentation.

Cycles: The Category That Rotates

Convergence is not the only stable pattern. Some categories never come to rest, because their strategies stand in a non-transitive relationship, and a non-transitive game has no single ESS. The biology here is vivid enough to make the mechanism unforgettable.

Barry Sinervo and Curtis Lively documented, in "The Rock-Paper-Scissors Game and the Evolution of Alternative Male Strategies" (1996, Nature 380, 240-243), a population of side-blotched lizards with three male types distinguished by throat color. Orange-throated males are aggressive and hold large territories. Blue-throated males hold small territories and guard a single mate closely. Yellow-throated males hold no territory and sneak matings by mimicking females. The three beat each other in a loop: aggressive orange overruns the small territories of blue, blue mate-guarders foil the sneaking yellows, and sneaking yellows slip past the overextended oranges. None can win outright, because each is beaten by the type it cannot defend against, and the field frequencies of the three morphs were observed to oscillate over a roughly six-year period.

Marketing categories cycle for the same structural reason whenever three positionings stand in a rock-paper-scissors relation. Take premium, value, and convenience. A premium position wins customers away from a plain value position by offering a reason to trade up. A value position undercuts a convenience position when customers grow price-sensitive and decide the surcharge for convenience is not worth it. And a convenience position peels customers off the premium players when it becomes good enough and radically easier, so that effort, not quality, becomes the deciding factor. Each beats one and loses to another. The replicator equation on that cyclic payoff structure produces oscillation rather than settling, and the same three positionings trade the lead indefinitely.

The path above is the replicator equation traced on a cyclic three-strategy payoff, started from a category that is 50 percent premium. No strategy ever wins; each peaks, gets undercut by the one it cannot defend against, collapses, and recovers only after the strategy that beat it is itself overtaken. A firm that reads its category as converging when it is actually cycling will make a predictable and expensive error: it will pile into whatever is peaking, arriving at the top just as the counter-strategy begins to eat it. The focal point a category seems to be settling on can be a phase in a rotation rather than a resting place, and telling the two apart is a measurement problem, not a matter of intuition.

What an Operator Should Actually Do

The framework earns its place only if it changes decisions, so here are the decisions, in the order a category review should take them. The through-line is that every one of them is a measurement, because the frequency-dependent payoff cannot be judged by taste.

First, measure the strategy distribution in the category. Instead of describing your competitors in prose, classify them. Pick the two or three positioning axes that actually vary, code every competitor's position on each, and count. The output is the distribution of strategies, which is the q in the payoff equation and the single most important number nobody computes. A category that feels crowded on a dimension may be crowded only in perception, with the actual mass of firms clustered elsewhere, and you cannot know without the count.

Second, estimate the imitation lag. How long, historically, has it taken for a visibly successful move in this category to be copied by half the field? Read the last several years of launches, repositionings, and pricing changes, and time them. A short lag means any deviation you make will be copied before it pays, pushing you toward positions defended by a real trade-off; a long lag means a temporary deviation can be worth it on its own. The lag is the copying rate in the replicator equation, and it sets how fast q will chase you.

Third, choose the timing of a deviation against the distribution and the lag, not against your appetite for it. Deviation is a good bet when the position is rare, the niche has evidence of real demand, and the lag is long enough to recoup the cost of pioneering. When those do not hold, the disciplined move is to run the consensus offer with a distinctive identity, per Sharp, rather than a differentiated offer that will bleed the contested center.

Fourth, defend a deviation with a commitment that cannot be copied fast. A position that rests on an expensive, hard-to-reverse investment, an exclusive supply relationship, a proprietary data asset, a brand built over years, a business model a rival would have to cannibalize itself to match, keeps q low by making imitation costly, which is the only thing that keeps the payoff positive. Porter's trade-off and the costly-signaling logic are doing the same work from two directions.

Loading diagram...

The uncomfortable conclusion, and the one worth ending on, is that most of the time the crowd is not wrong. Convergence is usually an equilibrium, not a failure, and the conventional wisdom that a brand must differentiate is, in the median category, advice to abandon the center where the demand actually is. The frequency-dependent payoff is what separates the cases: a deviation is worth making only when it is rare, backed by real demand, slow to copy, and defended by a commitment, and those conditions are measurable rather than matters of courage. When they hold, deviate and defend. When they do not, run the ordinary offer and win on distinctiveness and execution, which is where categories full of look-alike firms are actually decided. The market concentration this produces, and who ends up owning a converged category, is the subject of the winner-take-most analysis that this framework sits underneath.

Key Takeaways

1. Categories converge because strategy spreads by imitation of what appears to work, which is replicator dynamics, not because marketers lack imagination. Taylor and Jonker (1978) gave the equation: a strategy's share grows in proportion to how far its payoff sits above the population average, and no reasoning is required for equilibrium-like behavior to appear.
2. The stable state is often a mixture, not a single winning move. In the hawk-dove game of Maynard Smith and Price (1973), when the cost of fighting exceeds the value of the prize, the equilibrium share of aggressive play is exactly V/C, so aggression is self-limiting and a price war is a solved ratio rather than a lapse of discipline.
3. Convergence and differentiation are both equilibria of Hotelling's (1929) location model under different values of one parameter. d'Aspremont, Gabszewicz and Thisse (1979) showed firms crowd the center when a poor-fit purchase costs the customer little, and separate to the extremes when a poor fit is expensive. Measure the mismatch cost before advising anyone to be different.
4. The payoff to a contrarian position is frequency-dependent. Deviating beats conforming only while the deviation's share q stays below a threshold q* = (b-d)/[(b-d)-(a-c)], and the same position is a good idea when rare and a bad idea when crowded, with nothing about the position itself having changed.
5. Distinctiveness is not differentiation. Sharp (2010) and Romaniuk and Sharp (2016) show that convergence in what firms offer is often harmless, while convergence in how they signal identity is the real error, and the double jeopardy law of Ehrenberg, Goodhardt and Barwise (1990) warns that a niche position buys no compensating loyalty premium.
6. Some categories cycle rather than converge, in a rock-paper-scissors loop among premium, value, and convenience, the same non-transitive dynamic Sinervo and Lively (1996) found in lizards over a six-year period. Piling into whatever is peaking is the worst-timed move in a cycling category, because the counter-position is about to invade.
7. Deviation pays only when four conditions hold together: the position is rare, the niche has real demand, imitation is slow, and the position rests on a commitment rivals cannot copy fast. Each is measurable. When they do not hold, run the consensus offer with a distinctive identity and win on execution.

Further Reading

• Category Entry Points: A Quantitative Approach to Byron Sharp's Mental Availability Theory, the distinctiveness half of this argument, and how being easy to retrieve beats being different.
• Market Sensing Systems: Automated Competitive Intelligence with LLMs, how to build the pipeline that measures the strategy distribution this essay says every category review needs.
• Coordination Games, Standards Wars, and Product Launch, the sibling dynamic where expectations, not payoffs alone, decide which position becomes focal.
• Signaling Theory and Advertising as a Costly Signal, how a deviation is made credible and hard to copy through commitments that only a serious firm can afford.