Unit Costs Fall Predictably With Everything You Have Ever Made
Across many industries, each doubling of cumulative production has reduced unit cost by a roughly constant percentage. That regularity turned into a strategy: buy share now at a loss, because volume itself lowers your cost later.
A Regularity Nobody Expected to Be So Consistent
The observation originated in aircraft manufacturing, where analysts noticed that the labour hours required to build an airframe fell by a predictable percentage each time the total number built doubled. Later work extended it across many industries and found the pattern repeatedly, in semiconductors, chemicals, automobiles, and eventually solar panels and batteries.
The relationship is stated as a percentage: an eighty percent experience curve means that each doubling of cumulative output reduces unit cost to eighty percent of its previous level. Observed rates commonly fall between seventy and ninety percent depending on the industry.
The critical word is cumulative. This is not economies of scale, which describe cost falling with the volume produced in a given period. Experience effects depend on everything ever produced, which means they are path dependent and cannot be replicated simply by building a larger plant.
Where the Savings Actually Come From
The curve is an empirical regularity rather than a law, and it is worth being specific about the mechanisms behind it, because they are what determine whether it applies.
Labour learning is the original source: workers become faster and make fewer errors at a repeated task. Process improvement accumulates as engineers find better sequences, tooling, and layouts. Product redesign removes components and simplifies assembly once the design is understood in production. Yield improvement matters enormously in processes where a share of output is scrapped, since raising yield lowers cost per good unit without changing anything else. Supplier learning extends the effect up the chain as vendors run their own curves.
| Concept | Cost Falls With |
|---|---|
| Economies of scale | Output per period |
| Experience curve | Cumulative output ever produced |
| Technological change | New process, independent of volume |
The Strategy It Produced
Consulting work in the 1960s and 1970s drew an aggressive conclusion from the regularity. If cost falls with cumulative volume, then the firm with the largest cumulative volume has the lowest cost, and the way to acquire cumulative volume is to price aggressively early and take share.
The resulting prescription was to price below current cost in a growing market, accept early losses, accumulate volume faster than competitors, and arrive at a cost position rivals cannot match. Combined with growth share portfolio thinking, this became one of the most influential strategic frameworks of the era.
It worked in several documented cases, particularly in semiconductors and consumer electronics, where the mechanisms behind the curve were strong and the products were standardised.
The experience curve is a description that was converted into an instruction. The description was mostly right. The instruction assumed cost leadership would translate into profit, which requires that customers care about price above everything else.
Where the Strategy Failed
The failures are as instructive as the successes and they cluster into recognisable categories.
Cost leadership is not profit leadership. Buying share through price can produce the lowest cost position in an industry with no margin left in it. Several industries ran this experiment collectively, with every participant pricing for volume and none earning a return.
The curve applies to a process, not to a market. A firm that accumulates enormous experience in one manufacturing process gains nothing when the process is replaced. Experience in vacuum tubes was worthless in semiconductors. This is the classic disruption pattern, and the accumulated experience can even be a liability if it anchors the firm to the obsolete method.
Experience spills over. The assumption that learning is proprietary frequently fails. Engineers move between firms, equipment vendors sell the same improved tools to everyone, and suppliers pass on their own learning to all customers. Where spillover is high, the leader funds a curve its competitors descend for free.
Differentiation is an alternative. A competitor that avoids the volume race and sells a differentiated product at a premium is not required to be on the curve at all.
Where It Is Still Doing Real Work
The concept remains genuinely useful in industries with standardised products, high initial costs, and strong process learning. Solar module costs and lithium ion battery cell costs have both followed remarkably consistent experience curves over decades, and those curves have been used to forecast the point at which each technology becomes competitive without subsidy.
That forecasting use is arguably the most valuable modern application. If a technology is on a reliable curve, and you can estimate cumulative deployment over time, you can estimate when it crosses a cost threshold. Policy support for early deployment then becomes an investment in moving down the curve faster, which is an explicit rationale behind renewable energy subsidy programmes.
How to Use It Without Being Misled
Three questions determine whether the framework applies. Are the underlying mechanisms present, meaning is there repeated process learning and yield improvement rather than a one time design cost? Is learning proprietary or does it spill to competitors through shared suppliers and equipment? And is the process itself durable, or is there a plausible technology that resets the curve to zero?
Where the answers are favourable, aggressive volume strategy is defensible. Where they are not, it is a well documented route to owning the largest share of an unprofitable industry.
The Bottom Line
The experience curve is a real and widely observed empirical pattern that was oversold as a strategic rule. Cost genuinely does fall with cumulative production where process learning and yield improvement are the drivers, and it genuinely does not confer advantage where learning spills over or where the process gets replaced. The most durable modern use is not competitive strategy at all, it is forecasting when an emerging technology becomes cheap enough to win on its own.