You picked the payments provider in week three because it was the one that would approve you. Two years later it sets your unit economics, your settlement times, and which two countries you can expand into.
Nobody made a strategic decision here. A convenience from week three is now the shape of the company.
Why these three models
The decision is which of your existing commitments you can still change, and what changing one would actually cost. The features that fire are an outcome that depends on the order things happened rather than only on their merits, a set of variables where some move freely and others no longer do, and an improvement that is not reachable by small adjustments.
Three lenses. Path dependence produces a cycle answer about why you are here. Comparative statics produces an equilibrium answer about what would move if you pushed. The landscape produces a complex answer about why the obvious improvements are all small and the large one requires letting go of something first. The middle one is the diagnostic and it is the one nobody runs, because it looks like homework rather than strategy.
1. History is doing work in your present
Most business analysis is written as though the current state is a response to current conditions. Often it is not. It is a response to conditions that existed when a decision was made, plus everything that has since been built on top of that decision.
The mechanism is increasing returns. An early choice attracts complements: integrations, staff who know it, processes designed around its quirks, contracts that assume it. Each complement raises the cost of leaving, so later choices get made in a world where leaving is already expensive. This is why an option that was one of three roughly equal candidates in year one can be effectively the only option by year three, without ever having been better.1
Two things follow. First, the current arrangement is not evidence that it is the right arrangement, so benchmarking against it proves nothing. Second, and less obvious, lock-in is not always a mistake. Committing early is often correct because it buys speed, and the complements you accumulate are usually productive. The failure is not commitment; it is commitment that nobody labelled, so that the moment it starts to bind is the moment you discover it exists.
There is a further reason this stays invisible, and it is structural rather than a failure of attention. A structure that has been in place long enough stops being experienced as a structure at all. Everyone who joined after the decision treats it as a fact about the world, describes it in onboarding as how things work here, and defends it in planning without ever having weighed it.4 Nobody argues with a constraint they were never told was a choice.
The practical version is an inventory. List the five decisions your company still runs on that were made in the first year. For each, write what it would cost to reverse today, in weeks and in money. Most founders can complete that list in twenty minutes and have never once done it, which means the constraints they are operating under are unmapped.
Two rules make that inventory useful rather than decorative. Write the reversal cost as a range with the bad case included, because the number people volunteer is the happy path and the migration that matters is the one that goes wrong. And write who would have to do the reversal, by name. A change costing six weeks of the only person who understands the integration is not a six week change. It is six weeks during which nothing else in that area moves.
The inventory also has a category people leave off, because it does not look like a decision at all. Who your first large customer was belongs on the list. So does where you registered the company, which bank would open an account for you, and which country your first three staff live in. Those set your tax position, your currency exposure, and which markets you can hire into without a work permit, and every one of them was chosen on availability rather than on merit. They belong on the same page as the database.
2. Which variables can still move
The inventory tells you what is expensive. It does not tell you what is possible, and those are different questions.
Comparative statics is the discipline of asking what happens to the outcome when one input changes and everything else holds.3 It sounds academic and it is the most useful piece of arithmetic in this article, because it separates the variables you can actually push from the ones you only talk about.
Run it on a live constraint. If your payment provider’s fee dropped by a fifth tomorrow, what would change: pricing, margin, which segments become viable. If your settlement time halved, what changes: working capital, how much you can hold in inventory, whether you need the credit line at all. Do it one variable at a time and write the consequence down.
The result is usually surprising in a specific way. One or two variables turn out to move almost everything, and the rest move nothing, and the ones that move everything are frequently not the ones the team argues about. A founder team can spend a quarter debating a variable whose full swing changes the outcome by two percent, while a variable that would change it by forty percent sits unexamined because it was decided in year one and nobody thinks of it as a variable at all.
You do not need a model to run this. A sheet with the four or five numbers that drive your monthly result is enough: price, volume, the two largest cost lines, and the delay between delivering and being paid. Move one by a fifth, hold the rest, read the bottom line. Repeat for each. Twenty minutes gives you the ranking.
The failure mode is holding the wrong things fixed. If cutting your price by a fifth also changes volume, then reading the result as though volume stood still gives you a number that is not merely wrong but wrong in the flattering direction. Where two inputs are obviously linked, move the pair together and say so rather than pretending they are independent. And treat the output as a ranking, not a forecast. What you want is which variable dominates, and that ranking survives errors in the individual numbers that a forecast would not.
A number you have stopped questioning has stopped being a decision and started being a wall.
3. Why the improvement is not next door
The first two models say where you are and what moves. The third says why the obvious moves are disappointing.
Picture the space of possible configurations for your business as a terrain, where height is how well the configuration performs. If the terrain has a single smooth peak, incremental improvement works: every small step upward gets you closer to the best available arrangement. If the terrain is rugged, with many local peaks separated by valleys, then small steps take you to the top of whatever hill you happen to be standing on and no further. Getting to a higher peak requires descending first, which means a period of performing worse.2
This is the structural reason that sensible incremental optimisation stops working. You are not running out of ideas. You have arrived at the top of a hill that history put you on, and every remaining local move is downhill.
The diagnostic is honest and quick. If your last four improvement efforts each produced a small gain and the aggregate is roughly flat, you are on a local peak. That is a different situation from a team that is failing to execute, and it calls for a different response: a deliberate long jump, made at a time you can afford the valley, rather than more effort on the slope.
Long jumps are expensive and most of them fail, which is why they should be sized like bets rather than like plans. The relevant question is not whether the new peak is higher. It is how long you would be in the valley and whether you survive it, and that returns you to the reversal costs you wrote down in section one.
So size the valley before the peak, in the two units you can check. How many months of reduced performance, and how much cash that consumes at your current burn. If the answer runs longer than your runway, the jump is not available this year whatever its merits, and the honest conclusion is to extend runway first rather than attempt it underfunded and abandon it halfway down.
There is a partial version most founders walk straight past. You can often run the new configuration alongside the old one for a single segment: one country, one product line, one customer tier. That buys a small valley instead of the whole one. It is slower, it duplicates effort, and it converts an irreversible jump into a test whose result you can read before committing. The condition is that the segment has to be representative enough that success there means something. A pilot run on your most patient customers only tells you the new arrangement works for people who would have tolerated anything.
What the three say together
- Inventory the year-one decisions still in force, with a reversal cost in weeks and money for each.
- Run one-variable-at-a-time on your live constraints. Which single input, if it moved, would move the outcome most.
- Check whether your recent improvements aggregate to anything. Flat aggregate plus consistent small wins means a local peak.
- If you are on a local peak, size the jump against the valley, not against the destination.
Where they disagree
Path dependence and the landscape view give opposite advice about incumbency.
Path dependence says that the complements you have accumulated are real value: the integrations work, the staff are trained, the processes fit. Discarding them to chase a theoretically better configuration destroys something you actually built. The landscape view says those same complements are exactly what holds you on a local peak, and that they will keep every remaining move small.
Both descriptions are of the same asset, which is why the disagreement cannot be settled in the abstract. It is settled by whether the aggregate of your recent improvements is still rising. If small moves are still producing compounding gains, the complements are an asset and you should keep building on them. If the aggregate has gone flat while the individual wins continue, the same complements have become the constraint. Founders tend to notice the flattening two or three quarters after it happens, because each individual win still feels like progress.
What none of them contain
None of the three prices the human cost of a jump. A reconfiguration that is correct on the terrain still asks people who built the current arrangement to dismantle it, and some of them will leave rather than do that. The models treat configurations as free to change and they are not, and the people who leave are often the ones who understood the old system well enough to migrate it.
None of them accounts for the terrain moving. The landscape picture assumes the peaks stay where they are while you search. In a market with new entrants and shifting regulation the heights change underneath you, and a jump aimed at a peak that has since eroded is worse than staying put.
And none of them can tell you the difference between a local peak and a plateau you have not finished climbing. The evidence looks the same for at least two quarters. That is a genuine limit rather than an oversight, and the only mitigation is to write down in advance what result would tell you which one you are in, before you have an emotional stake in the answer.
The one action that survives the ignorance: write the five year-one decisions your company still runs on, each with a reversal cost in weeks. You will find at least one you assumed was permanent that is not, and at least one you assumed was flexible that is now load-bearing. Both discoveries change what you should be arguing about this quarter.
Who has to move
The founder has to run this, because everyone else in the company joined after the constraining decisions were made and experiences them as facts about the world rather than as choices. The instinct when growth flattens is to push harder on the current configuration, which is the correct move on a smooth terrain and the wrong one on a rugged terrain, and nobody can tell you which you are on without the aggregate check. The cheapest first test is the improvement audit: total the effect of your last four initiatives. If they sum to less than the smallest of them promised, you have your answer and it cost an hour.
Sources and notes
- Stefan Thurner, Rudolf Hanel and Peter Klimek, Introduction to the Theory of Complex Systems, Oxford University Press. Path dependence, increasing returns and lock-in are treated in the chapters on evolutionary and co-evolutionary processes, where the order of events determines which of several possible stable configurations is reached. Used in section 1 for the mechanism, and specifically for the point that the reached state carries no implication of optimality.
- Stefan Thurner, Rudolf Hanel and Peter Klimek, Introduction to the Theory of Complex Systems, Oxford University Press. Rugged fitness landscapes, local optima, and the requirement to cross a valley in order to reach a higher peak are developed alongside the treatment of evolutionary search. Used in section 3. The landscape is a model of search, not a measurement of any real business, and no claim is made here that a specific company’s configuration space has a particular shape.
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press. Comparative statics as a method, in which one parameter is varied while others are held fixed in order to sign the effect on the outcome, is used throughout the volume as the standard analytical tool. Cited in section 2 for the method rather than for any specific result.
- Donella H. Meadows, Thinking in Systems: A Primer, Chelsea Green Publishing. The observation that a system’s structure, once built, generates its own behaviour regardless of the intentions of the people inside it supports the reading in section 1 that current arrangements are not evidence of current fit. Used for the framing rather than for a specific claim.
A note on a number this article does not give. There is no threshold at which a lock-in becomes worth breaking. It depends on the reversal cost, the height of the alternative, and how long you can operate at reduced performance, and only the first of those three is knowable in advance. What transfers is the inventory and the aggregate check, both of which you can complete this week.
Joshua Agonya Pi’Rwot, Founder.