How Lock-In Breaks — and Where the Model Lies

Five lessons in, you could be forgiven for walking away with a fatalist’s worldview: the first-mover always wins, the worse standard rules forever, and history is a prison. That’s half the truth, and the dangerous half. Entrenched standards do get overturned — QWERTY’s descendants fall, empires of format collapse, the “unbeatable” incumbent gets beaten — and there are only a handful of ways it ever happens. Knowing those escape routes is the difference between describing a trap and picking its lock.

Then the harder job. This is the last teaching lesson, so it’s also the confession booth: path dependence is one of the most over-used models in social science, and a huge fraction of “it’s just a locked-in accident!” claims are lazy storytelling that would collapse if anyone asked for evidence. If you leave this course able to spot the escape routes and able to catch yourself misusing the model, you’ll have something most people who throw the term around don’t: judgment.

Commit to an answer before you read on.

If you picked the first option, you already sense the theme: escaping lock-in isn’t about being better, it’s about overwhelming or side-stepping the switching cost and the coordination trap. Every escape route below is a different way to do exactly that.

Escape route 1 — the external shock that resets the field

Analogy. A logjam on a river can hold for months — every log wedged against every other, none able to move alone. Then a flood comes through, lifts the whole mass at once, and the jam that looked permanent clears in an afternoon. The logs didn’t get stronger or weaker; the water level changed the rules for all of them simultaneously.

Precise definition. An external shock (or exogenous discontinuity) is a jolt originating outside the locked-in system — a war, an economic crisis, a regulatory mandate, a whole new technological generation — large enough that switching costs are temporarily swamped or reset for everyone at once. Lock-in is durable because moving is individually irrational while everyone else stays put; a big enough shock changes what “everyone else” is doing in one stroke, so the coordination trap momentarily dissolves.

Worked example. Consider a regulatory mandate. Suppose a government rules that, as of a fixed date, every new car sold must support a single charging standard. Before the mandate, no automaker dares back a new plug alone — they’d strand their customers. The rule doesn’t make the new standard better; it makes the switch mandatory and simultaneous, which is precisely the thing individual actors couldn’t coordinate on their own. The same shape appears when a new computing generation arrives: desktop’s lock-in didn’t have to be “beaten” head-on — mobile opened an entirely new context where the old installed base and habits didn’t carry over, and the field re-formed around new leaders.

Pitfall. Don’t confuse a shock with a slow trend. Lock-in erodes to a sharp, coordinated discontinuity, not to gentle pressure. A standard can sit under steady, gradual competitive pressure for decades and never break, then fall in months when a discontinuity hits. If you’re waiting for “gradual improvement” to dislodge an incumbent, you may be waiting forever; the history of these breaks is punctuated, not smooth.

Escape route 2 — a discontinuous leap in quality (10x, not 10%)

Analogy. To get someone to abandon a house they own and have furnished for twenty years, “a slightly nicer house” won’t do it — the hassle, the moving costs, the risk swallow the gain. You need a mansion, offered so far above their current home that even after paying every cost of the move they come out clearly ahead. A better challenger has to pay for the entire switch out of its advantage, with margin to spare.

Precise definition. A challenger overturns a locked-in standard when it is not marginally better but dramatically better — the informal bar is often stated as 10x, not 10%. The advantage has to be large enough to cover, for the typical adopter, the sum of (a) the private switching cost and (b) the coordination risk of moving before the crowd does — and still leave a gain obvious enough to act on. Below that bar, each individual’s rational move is to stay; above it, moving pays even if you’re early, which is what lets a stampede start.

Worked example. Recall from the QWERTY lesson why the Dvorak keyboard never won. Even in its most favourable studies, Dvorak was only somewhat faster — a marginal, contested edge. Retraining every typist, replacing hardware, and coordinating a whole workforce cost far more than a few percent of speed could ever repay, so the rational move was always to stay on QWERTY. Dvorak is the textbook case of a merely-better option failing to clear the bar. Contrast that with genuine displacements — a new medium that is not 20% better but transformationally better (orders of magnitude cheaper, or capable of something the old one simply couldn’t do). That gap is large enough to fund the whole cost of switching, so people switch.

Pitfall. Founders and analysts routinely mistake “clearly better” for “enough to win.” Against an un-entrenched rival, a 20% edge is decisive. Against a locked-in one, it’s often a rounding error, because the incumbent’s moat isn’t its quality — it’s the installed base and the coordination trap. The practical takeaway: to beat lock-in, don’t ask “is my thing better?” Ask “is it better by more than the total cost of switching plus the coordination risk?” That’s a far higher bar, and most challengers never clear it.

Escape route 3 — coordinated switching (solving the coordination trap)

Analogy. A crowded room, everyone standing, everyone’s neck aching — but no one sits, because sitting alone means missing what the standers can see. If a single voice could say “on three, everybody sit,” the whole room would gladly drop at once. The problem was never that sitting is bad; it’s that no one could move first. Lock-in is that room.

Precise definition. Coordinated switching is any mechanism that lets the crowd move together rather than one-by-one, so the fear of “moving before everyone else” — the thing that keeps the bad equilibrium stable — disappears. The bad standard is a Nash equilibrium: given that everyone else uses it, your best move is to use it too, even if a different standard would be better for all of you if you all switched. Anything that coordinates the jump breaks the trap: a standards body that declares the new spec, a legal mandate, a flag day (a pre-announced instant when everyone switches at once), a dominant buyer big enough to force its suppliers over, or a focal point everyone can see and converge on.

Worked example. In 1967, Sweden switched from driving on the left to driving on the right — literally the mother of all coordination problems, where being out of step is fatal. It worked because it was Dagen H (“H-day”): a single pre-announced date, 3 September, when everyone changed at 5 a.m. simultaneously. No driver had to gamble on switching before the others; the flag day made the move universal and instantaneous, which is the only way that particular equilibrium could ever be jumped. Tie this back to your critical-mass prerequisite: you don’t even need to force everyone at once — coordinate enough switchers to cross the critical-mass threshold and the reinforcing loop that used to defend the old standard flips and pulls the rest across for you.

Pitfall. Coordinated switching needs credible common knowledge, not just a good idea. If people doubt that others will actually move — if the flag day isn’t believed, the mandate isn’t enforced, the standards body isn’t trusted — everyone hedges by staying put and the coordination fails. The hard part is rarely choosing the better standard; it’s manufacturing the shared belief that everyone else is really going to jump. A half-believed flag day is worse than none.

Escape route 4 — sponsorship and subsidised switching

Analogy. A new nightclub across from the packed one can’t win on merit alone — an empty room stays empty. So it does what new clubs do: free entry, free first drink, pays a few well-known faces to show up early. It’s buying the crowd it can’t yet attract, funding the walk across the critical-mass threshold out of its own pocket, betting that once the room is full it’ll stay full on its own.

Precise definition. Sponsored (or subsidised) switching is when a deep-pocketed backer deliberately absorbs the early adopters’ switching costs to push a new standard past its critical mass — after which increasing returns take over and sustain it unaided. The tools are familiar: loss-leader pricing, giving hardware away, paying developers to build for the new platform, buying exclusive content. The logic mirrors how many incumbents originally won — an early lead that got amplified — except the sponsor manufactures that early lead on purpose with money instead of receiving it from luck.

Worked example. Think back to the intro’s lock-in explorer, and its external-shock button that could jolt a frozen market loose. A sponsor is that button with a budget. A firm that wants its format to become the standard can sell the players below cost, hand development kits to studios for free, and secure a must-have exclusive — each move eating the early adopters’ switching cost so that the installed base climbs to the tipping point on the sponsor’s dime. Once past critical mass, network effects do the rest and the subsidies can stop. The whole play is: pay to reach the threshold, then let the loop pay you back.

Pitfall. Sponsorship is expensive and fails if it stops short of the threshold. Subsidise adoption to below critical mass and the moment you turn off the money the market rolls right back to the incumbent — you’ve bought a temporary bump, not a tipped equilibrium, and burned the cash for nothing. Sponsorship only “works” if it actually crosses the tipping point; a half-funded standards war is a very expensive way to lose.

Escape route 5 — riding a new layer (envelopment)

Analogy. You can’t win the argument about which is the best horse if the world quietly switches to cars. You didn’t out-argue the horse people — you moved to a different layer where their entire debate stopped mattering. The old lock-in is still there; it just no longer binds anything anyone cares about.

Precise definition. Envelopment (or layer-jumping) is escaping lock-in not by beating the entrenched standard head-on but by moving to a new layer, platform, or use-case where the old standard’s switching costs don’t apply — making the incumbent’s moat irrelevant rather than surmounting it. Because you’re not asking anyone to abandon their installed base, you dodge the coordination trap entirely: you’re building a fresh context, and the fresh context has no legacy to defend.

Worked example. An operating system might be utterly locked in on desktop computers — decades of installed base, switching costs, developer habits. A challenger that tried to win that fight head-on would lose. But when computing shifted to phones, that desktop lock-in simply didn’t carry over: a new device class, new form factor, new set of habits, a blank field with no entrenched winner. The desktop incumbent’s moat was perfectly intact and completely irrelevant to the new layer. The winner there didn’t beat the old standard; it routed around it. Same pattern whenever a dominant standard gets enveloped by a new platform that makes the old battlefield beside the point.

Pitfall. Envelopment is often misread after the fact as “the old giant was toppled.” It usually wasn’t — it’s still dominant on its own layer; it was simply made less important by growth elsewhere. Reading every layer-jump as a head-on defeat teaches the wrong lesson (that entrenched standards are fragile to direct assault) when the real lesson is the opposite: they’re so hard to beat directly that the smart move is usually to not fight them at all and change the layer instead.

Synthesis — lock-in is durable, not eternal

Step back and notice that all five routes are the same move wearing different clothes. Lock-in is defended by exactly two things: the switching cost each adopter pays to move, and the coordination trap that makes moving irrational unless enough others move too. Every escape either overwhelms that defence (a shock resets it; a 10x leap out-pays it; a sponsor buys past it) or side-steps it (coordination lets the crowd move as one; envelopment moves to a layer where it doesn’t apply). There is no sixth kind of magic — if you can name which of these is (or isn’t) in play, you can predict whether a given incumbent is actually vulnerable or just looks beatable.

That’s the constructive half of the lesson. Now the harder, more valuable half: learning where this whole model lies if you use it carelessly — because a tool this powerful is also this easy to abuse.

Where the model lies 1 — the just-so-story trap

Analogy. “The leopard has spots because long ago it needed camouflage” is a just-so story: a tidy narrative that explains the outcome, sounds plausible, flatters our love of reasons — and offers not one shred of evidence that it’s actually why. Path dependence generates these by the thousand: any market winner can be re-narrated as “a historical accident locked in over something better.” It always sounds good. That’s the danger.

Precise definition. The just-so-story trap (named in this literature by the economists Liebowitz and Margolis) is declaring an outcome a locked-in accident without establishing the two things the claim actually requires: (a) that a genuinely better alternative existed, and (b) that switching costs, rather than the winner’s ongoing merit, are what keeps it in place. Absent both, “it’s just locked-in path dependence” is a story, not a finding — and markets often do correct themselves when the prize for switching is big enough, which is precisely what routes 1–5 above describe.

Worked example. Someone claims a dominant software product “only won by lock-in and is worse than its rivals.” To turn that from a story into a real claim, demand the counterfactual: name the specific better alternative; show it was actually available; show it’s genuinely better on the merits and not just to the claimant’s taste; and show that what keeps users on the incumbent is switching costs and not the fact that it’s still, boringly, good enough or even best. If a rival really were dramatically better, route 2 says it could clear the switching-cost bar — so the persistence of the incumbent is itself weak evidence that no such 10x-better option exists. Very often the honest answer is “the winner is actually still competitive,” and the lock-in story evaporates.

Pitfall. The just-so story is seductive precisely because it’s unfalsifiable as usually told — it explains any winner equally well, which means it explains none. The discipline is a single demand: produce the counterfactual, or withdraw the claim. If you can’t name the better option and show that switching costs (not merit) are what’s blocking it, you don’t have path dependence — you have a narrative that flatters your priors.

Where the model lies 2 — path dependence vs. mere causation

Analogy. “I’m standing here because I walked here” is true, and it is not a theory of anything — it’s just what “here” and “walked” mean. Now compare “I’m standing here because a coin flip three hours ago sent me left instead of right, and every step since had to build on that, so I ended up somewhere no planner would have chosen.” Only the second is a claim. The first is grammar.

Precise definition. Mere causation — “the past influenced the present” — is trivially true of literally everything and is not path dependence. As lesson 1 hammered, the real claim requires a small, contingent early difference that got amplified into a large, persistent, hard-to-reverse gap. If the outcome was instead determined by big, obvious factors — better resources, a real and durable advantage, a decisive early superiority — then it isn’t path dependence, it’s ordinary cause and effect that would repeat in any re-run. Path dependence lives specifically in the gap between a tiny trigger and a huge, sticky, contingent result.

Worked example. “This city became a major port because of its history.” Weak reading: yes, ports have histories — empty. Strong (path-dependence) reading: a small, chance early event (one trading post, a single early rail junction) gave it a tiny lead that got amplified until it froze, and a dozen other locations could have played the role equally well. But suppose the city sits on the only deep natural harbour for 500 miles. Then it isn’t path dependent at all — a big, obvious, durable geographic advantage picked it, and it would win every re-run. Same sentence, opposite verdict, and the whole difference is whether the trigger was a small contingent accident or a large determining factor.

Pitfall. The tell is the size and contingency of the trigger relative to the outcome. If someone points to a big cause for a big effect and calls it path dependence, they’ve mislabelled plain causation. Path dependence is the surprising claim — small in, huge and sticky out, and easily-could-have-gone-otherwise. If it could only have gone one way, it was never path dependent; it was determined.

Where the model lies 3 — assuming lock-in is permanent

Analogy. The mirror image of overusing the model: the general who fortifies against the last war. Convinced the current standard is unbreakable, an incumbent stops watching for the shock, the 10x challenger, or the layer-jump — and gets blindsided by exactly the escape route Part A just catalogued. “It can never be dislodged” is how the dislodged always felt the year before.

Precise definition. Treating every entrenched standard as permanently locked in is the direct opposite error to the just-so story, and just as costly. It ignores that lock-in is durable but not eternal — that shocks, 10x leaps, coordinated switching, sponsorship, and envelopment are live, recurring escape routes. For an incumbent, this breeds fatal complacency; for an analyst, it produces confident predictions of permanence that history routinely embarrasses.

Worked example. A dominant firm reasons: “Our network effects and switching costs are insurmountable; no one can catch us.” That may be true against a head-on, merely-better rival — routes 2 and 3 are hard. But it says nothing about a technological-generation shock (route 1) that resets the field, or an envelopment (route 5) that makes the whole battlefield beside the point. The graveyard of “permanent” monopolies is full of firms that correctly judged themselves unbeatable on the old layer and never looked up. Being right about the head-on fight lulled them into missing the flank.

Pitfall. Don’t let the durability of lock-in curdle into a belief in its permanence. The two errors are symmetric: over-applying the model sees lock-in everywhere; this error sees escape nowhere. The corrective for both is the same — always ask which specific escape routes are, or aren’t, available for the case in front of you.

Where the model lies 4 — not everything locks in

Analogy. To someone holding the lock-in hammer, every market looks like a nail. But most markets aren’t tippy at all: buy a better sandwich, a better haircut, a better umbrella and nothing snowballs — your choice doesn’t make the option more attractive to the next person. Calling those “locked in” is like diagnosing a limp in someone who’s simply standing still.

Precise definition. Over-application is forcing the lock-in model onto outcomes that lack its precondition — increasing returns. Path dependence and lock-in need a reinforcing loop where each adopter makes the option more attractive to the next. Markets with constant or diminishing returns — most ordinary goods and services — simply don’t tip: quality stays self-correcting, no early accident runs away, and there’s no lock-in to explain. Relatedly, most first-movers are overtaken; “first” is not “locked in,” and treating every early leader as permanently entrenched is the same misuse from another angle.

Worked example. Compare two claims. “Our office messaging tool is locked in” — plausible: it has network effects (more valuable the more colleagues use it), so an early lead can freeze. “Our brand of stapler is locked in” — nonsense: staplers have no network effects, your buying one doesn’t make it better for the next buyer, and a better, cheaper stapler will simply take the market. Same word, but only the first market has the increasing-returns engine that lock-in requires. Before you reach for the model, check for the engine; no increasing returns, no lock-in.

Pitfall. The meta-mistake of the whole course: path dependence is a specific claim about a specific mechanism, not a universal theory of why-things-are-the-way-they-are. Applying it to everything with a history dilutes it into the empty “the past influenced the present” — the exact trivial version lesson 1 spent its length disarming. A model that explains everything explains nothing. Reserve it for cases with genuine increasing returns, a small contingent trigger, and a real counterfactual — and let ordinary competition explain the rest.

Now the discriminating skill that separates people who understand this model from people who just enjoy it: sorting real path-dependence claims from stories.

Putting both halves together

One last mixed quiz spanning the escape routes and the honest limits — because the whole point is to hold both in your head at once.

Where this goes next

That’s the whole model — the claim, the mechanisms, the canonical cases and their debate, the four engines, standards wars, institutions and careers, and now the escape routes plus the honest limits. You can state path dependence precisely, spot it in the wild, tell it from mere causation, name how a lock-in breaks, and catch yourself misusing it. That’s genuine expertise, not a slogan.

One thing remains: proving it. The Final Exam is a graded, one-way run through the whole course — one question at a time, and submitting an answer locks it in (no going back, no retries, no restart), which is fitting for a course about irreversibility. You’ll need 70% to pass, and your score appears only at the very end. Bring the scalpel one last time.