Operating leverage is the most reliable idea in business strategy, and for roughly a decade it has been quietly failing. The premise is simple enough to fit on a slide: fixed costs stay fixed, so every additional pound of revenue drops through to profit at a higher rate than the last. Grow, and margins expand on their own. It is the reason so many businesses were told to prioritise scale over profitability, and the reason so many investors were willing to fund the wait.
Key points
- Incremental operating margin has fallen by roughly two-thirds across a broad cohort of scaled businesses since the early 2010s.
- The cause is not weak management. Costs the model treats as fixed have become variable, and the model has not been updated.
- Three specific culprits: cloud and compute, distribution costs that scale with revenue, and compliance that scales with footprint.
- What replaces operating leverage is narrower and harder: pricing power, genuine network effects, and disciplined scope.
The trouble is that the businesses which grew did not, on the whole, see the margin expansion the model predicted. Some did. Many held margins flat while doubling in size, which the model says should be impossible. And an uncomfortable number expanded revenue substantially while margins deteriorated, an outcome the framework treats as a management failure, and which was blamed on management with some regularity.
We think the framework is wrong more often than the managers are.
The mechanism, as taught
The classical version divides costs into two buckets. Fixed costs, the factory, the head office, the core engineering team, do not move with volume. Variable costs move proportionally. Because the fixed bucket is spread across a larger revenue base as the business grows, unit economics improve automatically, and the incremental margin on new revenue exceeds the average margin on existing revenue.
This was an excellent description of an industrial economy. A steel plant genuinely has enormous fixed costs and genuinely does become more profitable per tonne as utilisation rises. The model was built on a real observation about real businesses.
Its extension to the modern economy rested on an analogy: software has near-zero marginal cost, therefore software businesses have extreme operating leverage, therefore growth is the dominant priority and margin will arrive later. The first clause is true. The chain that follows from it has turned out to be considerably weaker than assumed.
What actually broke
The model did not fail because managers became worse at managing. It failed because the costs it classifies as fixed have, one by one, become variable, and nobody reclassified them.
Compute is the clearest case. A company that owned its servers had a genuine fixed cost: buy the hardware, spread it over as many users as possible. A company renting compute has a cost that scales, with reasonable precision, with usage. The industry replaced a fixed cost with a variable one, correctly celebrated the flexibility this bought, and then continued to model margins as though the fixed cost were still there. Add inference workloads to that base and the variable component grows faster than revenue rather than slower.
Distribution is the second. When customer acquisition runs through auction-based channels, the cost of the next customer is set by what the next-highest bidder will pay, not by an internal efficiency curve. This is close to the opposite of a fixed cost: it rises as the category becomes more valuable, which is precisely when a growing business is trying to acquire most aggressively. Businesses that built genuinely owned distribution avoided this. Most did not build it, because renting was cheaper at the time.
Compliance is the third and least discussed. Regulatory cost scales with jurisdictional footprint, product surface area and headcount, all of which grow with the business. Data protection, financial crime, employment law, sector-specific regimes and the reporting apparatus around each are not a fixed head-office overhead. They are a tax on complexity, and complexity is what scale produces.
The model did not stop working because managers got worse. It stopped working because the costs it calls fixed became variable, and nobody moved them across the page.
Zimba Capital Research
| Cost line | 2015–18 | 2024–26 | Change |
|---|---|---|---|
| Cost of revenue (incl. compute) | 27p | 34p | +7p |
| Sales & marketing | 24p | 29p | +5p |
| Research & development | 13p | 15p | +2p |
| General, admin & compliance | 7p | 11p | +4p |
| Retained as operating profit | 29p | 11p | −18p |
The table is the argument in miniature. No single line moved catastrophically. Four lines each moved a few pence in the same direction, and the residual, which is the entire point of operating leverage, absorbed the whole of it.
What replaces it
If scale no longer produces margin automatically, then margin has to be produced deliberately. Three mechanisms still work, and each is narrower and more demanding than the one it replaces.
Pricing power. The only cost structure that reliably improves with scale is the one where the customer accepts a higher price. This requires genuine differentiation, meaningful switching costs, or a position in a customer’s workflow that makes replacement expensive. It is difficult to build and straightforward to test: raise prices and observe what happens. Most businesses have never run that test honestly, and a surprising number discover on running it that they have less power than their strategy documents assume.
Genuine network effects. Not the diluted version where more users make a product marginally more useful, but the strict version where each additional participant raises the value of the network to every existing participant. Real network effects do produce something like operating leverage, because the value created scales faster than the cost of serving it. They are rare, they are usually identifiable early, and they are frequently claimed by businesses that do not have them.
Scope discipline. The least glamorous and most available. Much of the margin erosion in the table comes from businesses doing more things, more products, more segments, more geographies, each of which carries its own compliance, support and complexity cost. A business that grows within a tightly held scope retains far more of its incremental pound than one that grows by addition. This is a decision, not a market condition, and it is the one lever available to almost every management team reading this.
What would change our mind
If incremental margins recover across the cohort as compute pricing normalises and the current infrastructure build-out is amortised, then the last few years read as a capex cycle rather than a structural break, and the classical model survives with an asterisk. We would want to see two consecutive years of improvement across sectors, not one year in one sector.
Equally, if a meaningful group of businesses demonstrates that automation genuinely converts variable cost back into fixed cost, that the compute bill behaves like a factory rather than a meter, then the mechanism is restored in a new form. There are early claims to this effect. There is not yet evidence at the level of reported operating margin, which is where the claim would have to show up.
Until then, the practical guidance is unglamorous. Stop assuming margin arrives with size. Identify which of your costs the model still treats as fixed and check whether they are. And be honest about which of the three replacement mechanisms your business actually possesses, because the answer for most businesses is the third one, and the third one requires saying no.
