When to do margin

Every business can raise its margin tomorrow morning. Price is a decision, not a discovery, and the cost of raising it is always paid later and somewhere else — in volume, in share, in a competitor who now has a business case. So the interesting question is never can we, and rarely how much. It is when. This is a review of what the strategy and industrial-organisation literature actually establishes about that timing, organised around a single decision rule and applied at three levels: the account, the market, the geography. It is the companion to Where is the margin in the world?, which maps where margin has already collected.

The question, stated precisely

Extraction trades a certain gain now against an uncertain loss later. The gain is arithmetic: raise price by a fraction δ and, on the volume you keep, you earn δ more per unit. The loss is a stream: the customers the increase costs you would have bought from you again next year, and the year after, in a market that may be growing. Timing the move means knowing when the first number beats the second. Everything below is a way of making those two numbers comparable.

The static rule: break-even elasticity

Start with one period. Write m for the contribution margin ratio and ε for the elasticity of volume to price. Raising price by δ leaves you with volume Q(1 − εδ) earning a margin of (m + δ) per unit of the old price, so profit changes by

ΔΠ ≈ P·Q·δ·(1 − ε·m)

which is positive exactly when ε < 1/m. This is the break-even elasticity every pricing desk knows, and it is the same object as Abba Lerner's 1934 index, (P − MC)/P = 1/ε, read from the other end: the margin you can hold is the reciprocal of the elasticity you face. Ramsey had already given the portfolio version in 1927 — across several things you can price, mark up most where demand responds least.

The rule is correct and nearly useless on its own, because it treats a lost customer as a lost sale. In any business with repeat purchase, a lost customer is a lost annuity.

The dynamic rule: the franchise multiple

Suppose the customers you lose would have been retained at a rate ρ each period, in a market growing at g, and that you discount at r. One unit of volume lost today costs you not one period of contribution but

M = 1 / (1 − ρ(1 + g)/(1 + r))

periods of it — a geometric sum, and the same algebra as a growing perpetuity. Call M the franchise multiple. Putting it back into the static comparison, extraction is worth it when

ε · m · M < 1

Three things follow immediately, and they are the substance of the whole article.

• Growth is the dominant term, and it is non-linear. M rises hyperbolically in g and diverges when ρ(1 + g) ≥ (1 + r). At that point the customer base is a perpetuity growing at least as fast as it is discounted, its value is unbounded, and no price increase whatsoever is worth the customer it costs. This is not a metaphor for “growth is important”; it is a singularity, and it is the formal content of the land-grab strategy.
• Retention enters twice, in opposite directions. High ρ raises M, which argues against extracting — and simultaneously lowers ε, because a customer facing a switching cost does not leave over a price rise, which argues for it. Which effect wins is the entire switching-cost literature in one line. Klemperer's 1987 answer is that it depends on where you are in the life of the base: build it while it is being allocated, price it once it is locked. Bargain, then rip off.
• A high contribution margin makes you less free, not more. The m in the product means that the businesses with the most attractive margins — software, pharma, luxury — have the least room to raise them, because every unit they lose was carrying a large contribution. A 75%-margin business tolerates a third of the volume loss a 25%-margin business can. This is counterintuitive and it is right.

The extraction frontier

Rearranged, the rule gives the largest volume response you can absorb and still be right to extract: ε* = 1/(m·M). The chart plots that frontier against market growth, one curve per retention rate. Sit below a curve and taking price is correct; sit above it and you are selling an annuity to book a quarter.

One pleasant feature of the algebra: although M is hyperbolic in g, the frontier itself is exactly linear, since ε* = (1 − ρ(1+g)/(1+r))/m. It has slope −ρ/(m(1+r)) and crosses zero at

g* = (1 + r)/ρ − 1

which is the growth rate above which extraction is never justified, whatever your margin. At 95% retention and a 12% discount rate that threshold is about 18% — a number well inside the range of ordinary commercial markets, not an exotic corner case.

Read the middle curve at 5% growth, a 50% contribution margin and a 12% discount rate: the tolerable elasticity is about 0.4, meaning a 10% price rise must cost under 4% of volume. At 20% market growth the same business can afford almost nothing. The frontier is steep in exactly the range where most real decisions sit, which is why this question feels harder in practice than it looks in a spreadsheet.

Six regimes

The rule has four parameters, and in practice businesses cluster into six recognisable configurations of them. Each has a canonical move and a canonical way of being played wrongly.

The failure modes are worth as much as the moves, because each regime has a characteristic way of being misdiagnosed:

• Land grab. Believing you are in a land grab when the market is neither growing fast nor sticky. Subsidising customers who churn is the most expensive mistake in business, and it is the default one.
• Skim. Walking the price down too slowly and handing the volume tier to whoever entered second, who now has the scale and is coming for your tier next.
• Capacity squeeze. Mistaking a demand spike for a capacity constraint. If rivals can add capacity in twelve months, you have pre-announced the price that makes it worth their while.
• Harvest. Harvesting a business that was merely having a bad two years. The move is close to irreversible: the capability you cut is what you would need to come back.
• Deter. Paying the premium when nobody was going to enter, or when the entrant's cost of capital is political rather than financial and no price deters them.
• Defend. Defending a position that structurally cannot be held, which converts an orderly harvest into a disorderly one and destroys the cash that was still recoverable.

What the evidence says about momentum

The most common intuition about timing is that a hot market is a high-margin market: demand is strong, so take the price. The literature does not support that as a general claim, and the disagreement is instructive enough to be worth laying out rather than resolving.

• The collusion view says mark-ups fall in booms. Rotemberg and Saloner (1986) showed that where prices are held up by tacit cooperation rather than by structure, a demand boom raises the one-off payoff to undercutting and therefore makes discipline harder to sustain. Price wars break out in good times, not bad ones. Anyone who has watched an industry discount into its best year knows the pattern.
• The capacity view says the opposite, conditionally. Kreps and Scheinkman (1983) established that when capacity is committed before price, the outcome is Cournot rather than Bertrand: prices clear well above cost. The reconciliation is that a boom raises margin only when capacity cannot follow it. Strong demand with loose capacity is a price war waiting to happen; strong demand against a hard capacity ceiling is the one clean margin window there is.
• The recent macro evidence is unsettled. Nekarda and Ramey (2020) re-examined mark-up cyclicality with better measurement and found them weakly procyclical or acyclical, overturning the countercyclical consensus. The honest reading is that the average across all industries carries almost no information, and you are obliged to read the capacity position of your own.
• A common shock is a coordinating device. Weber and Wasner (2023) document the mechanism behind the 2021-22 episode: a visible, publicly understood cost shock legitimises an increase that would be punished as a unilateral move. The window is real, it is opened by the shock being legible rather than by it being large, and it closes when the shock stops being news.

So “the market has momentum” is not by itself a reason to extract. Momentum raises g, which pushes the frontier against extraction, and it only helps if it arrives at an industry that cannot add supply. Growth plus tight capacity is the buy signal. Growth plus loose capacity is the classic trap.

The three levels

The same rule applies at the account, the market and the geography, but the observable signals differ, and a firm is routinely in different regimes at different levels at once — harvesting a mature account inside a market it is still land-grabbing, or defending share at home while extracting in a country where it is entrenched.

At the account level the literature is unusually concrete. Reichheld and Sasser (1990) found that a five-point improvement in retention raised customer net present value by between 25% and 85% depending on the industry — which is the ρ term measured directly. Gupta, Lehmann and Stuart (2004) then showed that aggregated customer lifetime value tracks firm value closely enough to be used for valuation. Put together, they make the account-level trade explicit and slightly alarming: a price rise that raises reported profit while lowering retention can reduce the value of the firm at the same time as it improves the quarter, and both effects are measurable.

Where the literature agrees, and where it does not

A review should be clear about which of its claims are settled.

• Settled: margin is bounded by structure, not by ambition. From Lerner through Bain and Porter to Sutton, the finding is consistent. Where entry is easy or capacity is elastic, extraction calls forth the supply that removes it. Sutton's contribution is the sharpest: in industries where R&D or advertising spend is endogenous, a fat margin triggers an escalation in spending rather than an entrant, which removes it just as surely and more expensively.
• Settled: the turn from building to extracting is real and has an order. Klemperer's bargain-then-ripoff path, and Farrell and Klemperer's 2007 survey, establish that land grab and harvest are phases of one strategy rather than rival ones. The controversy is about timing the turn, not about whether there is one.
• Contested: whether share itself is worth buying. The PIMS work of Buzzell, Gale and Sultan (1975) found roughly three and a half points of return on investment per ten points of share, and that number drove a generation of share-buying. Jacobson and Aaker (1985) showed the correlation largely dissolves once unobserved firm quality is controlled for: share and profit are joint products of being good, and purchasing the share does not purchase the profit. The modern position is that share is worth buying only where a specific mechanism converts it into advantage — Lieberman and Montgomery's pre-emption, learning or switching costs — and Golder and Tellis (1993) showed that correcting for survivor bias, pioneers failed about half the time.
• Contested: the cyclicality of mark-ups, as above. Rotemberg and Saloner against Nekarda and Ramey, unresolved.
• Under-studied: the reversibility of extraction. The literature models the price increase; it rarely models coming back. Dixit and Pindyck's option-value framework is the right tool and it is not much applied here. In practice a reputational reset is closer to irreversible than to costly, which argues for treating extraction as a commitment and requiring a margin of safety on the rule rather than a bare inequality.

A practical decision sequence

Reduced to something usable, in order, because the sequence matters more than any single answer:

• 1. Can supply follow? If a competitor can add capacity within your planning horizon, you are setting their business case, not your price. Stop here.
• 2. Is g above or below r? Compare the market's growth to your cost of capital honestly, at the level of the segment rather than the industry. Above it, the franchise multiple does the deciding for you.
• 3. Is the customer base still being allocated? If buyers are choosing a supplier for the first time, you are in the phase where share is bought once and held; if they chose years ago, you are in the phase where it is priced.
• 4. What is ε here, not in general? Estimate it on this account or this segment, from renewal behaviour, not from a category average. It is the parameter most often assumed and least often measured.
• 5. Then apply ε·m·M < 1, and require room. If the inequality only just holds, it does not hold: the parameters are estimates and the move is close to irreversible.

The literature

24 works, tagged by the regime or the term of the rule each one bears on. The rule tag marks the results that constrain margin in general rather than in one regime.

Densest around the land grab (7 works) and the capacity squeeze (4), thinnest on deterrence (2) — which is itself a finding. The situations where a firm deliberately holds margin down are the least studied, because they are the hardest to observe: a price that was never raised leaves no trace in the data.

Appendix — the model and its limits

The derivation

With price P, volume Q and contribution margin ratio m, profit is Q·m·P. A relative price rise of δ gives volume Q(1 − εδ) and a per-unit contribution of P(m + δ), so the first-period change is P·Q·[δ − εδm − εδ²], which for small δ is P·Q·δ(1 − εm). The lost volume εδQ is then treated as a stream retained at ρ per period in a market growing at g and discounted at r, giving the geometric sum M. Setting the gain against the discounted loss yields ε·m·M < 1. The approximation drops the second-order term εδ², which is small for the price moves anyone actually makes and which biases the rule slightly toward extraction, so the margin of safety asked for above is not decorative.

What the model leaves out

• Competitive response. ε is written as a property of demand, but in a concentrated market it is mostly a property of what rivals do next. A price rise that is matched has an ε near zero; the same rise unmatched can have an ε several times the historical estimate. Rotemberg and Saloner is precisely about this, and it does not reduce to a single elasticity.
• Segment heterogeneity. One ε for a customer base averages over segments that differ by an order of magnitude, which is why price discrimination usually dominates a uniform increase. The right first question is often not how much to raise but on whom.
• The regulator and the counterparty. Extraction in healthcare, utilities, defence or public procurement invites a response from something other than a competitor, and that response is not priced by any of these terms.
• Reputation across markets. The model treats each market separately. A firm known for raising price on locked-in customers pays for that reputation in every negotiation it has not yet had.

On the sources

The works cited are the primary literature, given with venue and year so they can be checked. Findings are stated as the authors reported them; the mapping onto the six regimes, the franchise multiple and the decision sequence are this page's synthesis, not theirs, and the disagreements noted above are live in the literature rather than settled in favour of the reading offered here.