Experience-curve pricing

Why would a manufacturer deliberately sell its first units at a price below what they cost to make?

Selling below cost looks like the one mistake no manufacturer could make twice. Yet it is a deliberate, budgeted, board-approved decision often enough to have a name. The obvious reading — that the firm is buying market share and hoping — explains nothing, because any loss at all buys share. What has to be true for the loss to be a purchase rather than a gift?

Start with an observation from aircraft factories rather than from strategy. In 1936 T. P. Wright noticed that the labour needed to build an airframe fell in a startlingly regular way: each time the cumulative number of units ever built doubled, labour per unit fell by roughly a fifth. Not each year. Not at higher output rates. Each doubling of total units ever made.

That word "cumulative" carries the entire argument, so hold onto it. Take a starting cost of 100 and a curve that removes 20 per cent per doubling. The first doubling gives 80, the second 64, the third 51.2, the fourth 41. Four doublings is sixteen times the output, and the unit cost has fallen by 59 per cent. If instead the rate were 10 per cent per doubling, the same sixteenfold expansion leaves cost at 66, a 34 per cent fall. Same volume, very different worlds.

Now the pricing follows almost mechanically. If cost is a function of accumulated output, cost is not a fact about the firm — it is a variable the firm moves by selling. And if a low price sells more units sooner, the low price is what causes the cost to fall. The firm is not selling below its cost; it is selling above its future cost and paying the difference to get there.

Which is a lovely argument, so let us try to break it, in three places.

First: is the driver really cumulative volume? This is the load-bearing claim, and the shakiest part. Cost also falls with calendar time, with plant scale, and with technologies arriving from outside, and all four move together in the data. If the true driver were time or plant size, rushing volume forward buys nothing and the low price is simply a loss. There is decent evidence for genuine learning-by-doing in assembly-heavy processes — shipyards, airframes, semiconductor fabrication — and much weaker evidence that a firm can conjure it by pricing.

Second: does the firm keep what it learns? Learning leaks. It leaves in the heads of engineers, in the equipment vendors who sell the improved machine to everybody, in reverse-engineered products. Where it leaks fully, the pioneer pays for a cost reduction its rivals also receive, which is worse than doing nothing. Photovoltaic modules are the well-known case: prices have followed a steep learning curve for decades — most estimates put it near a fifth off per doubling, though the figure is disputed and depends heavily on the period chosen — and yet no manufacturer converted that into a lasting cost advantage, because the learning belonged to the industry.

Third, and fatal to the strategy even when it works: doublings get scarcer. The first might take a quarter. After a decade of growth, doubling cumulative output means producing again everything you have ever produced. The cost decline per unit of time therefore collapses even though the curve is unchanged, and a firm that priced against a decline that has effectively stopped keeps the low price and none of the compensation.

So what survives all three? Something narrower and more defensible than the 1970s version. Pricing ahead of cost is rational where the learning is real, largely proprietary, and early — an emerging product, a process nobody has run at volume, a firm still in its first few doublings. It is a bet on a specific causal chain, and every link of it is checkable in advance rather than assumable.

One refinement holds the pieces together: the experience curve is a claim about accumulated units, and scale economies are a claim about units per period. They are frequently confused, they behave completely differently, and only the first justifies pricing below present cost. A big plant is cheap immediately; experience must be bought.

Now the misconception to name as wrong. Selling below cost at launch is regularly described as predatory pricing — driving rivals out to raise prices later. Predation is a real thing, but it is a poor account of launch pricing in a growing market: there is often nobody to drive out, the strategy is only worth it if entry can be prevented afterwards, and the experience-curve story requires no rival at all to make sense. Reaching for predation explains the wrong phenomenon.

Two further honest caveats. The Boston Consulting Group's generalisation of Wright's finding from labour hours to total cost, and from aircraft to everything, went well beyond what the evidence supported, and by the 1980s "we are on the experience curve" had become a respectable way of describing losses that were simply losses. And the direction of causation is contested to this day: cheaper products sell more, so falling cost drives volume as surely as volume drives falling cost, and separating the two convincingly in observational data is very hard.

• What makes some technologies follow Wright's law cleanly while others plateau early.
• How a firm can protect learning it cannot patent, and why process knowledge travels differently from product knowledge.

Every term the collection defines is gathered in the glossary.