Concept · Economics

Experience curves

Experience curves describe how unit costs may change with cumulative output as firms learn, redesign, and improve; the slope must be estimated for the relevant process.

Cumulative experience may reduce hours or cost per unit.

An experience curve summarizes how a unit outcome changes with cumulative activity. In a power-law model, unit cost at cumulative output Q is c(Q)=aQ^b, with a and b estimated from comparable observations. The progress ratio for a doubling is c(2Q)/c(Q)=2^b; the implied fractional reduction is 1−2^b when b is negative. A progress ratio is an estimated relationship over a range, not a promised improvement schedule.

In one sentence

An experience curve represents the observed relationship between cumulative production or activity and a unit cost or performance measure.

The curve is empirical, not guaranteed. Theodore Wright’s 1936 paper on aircraft costs says the curve was worked up from limited prior production points and corrected as new experience arrived. Modern estimates should separate labor learning, redesign, material prices, supplier changes, product mix, automation, and scale utilization. If unit definitions or technology change, historical slope may not carry forward.

Use an experience curve for scenario planning and learning investment, not as a promise of future margin. Estimate on consistent cohorts of cumulative output; plot residuals; test structural breaks; and report a range. Be cautious when applying industry-level slopes to a specific company or when early data span a different product generation. A strategy that buys market share at losses needs stronger economics than a projected curve alone.

Learning mechanisms

A curve can combine several mechanisms that should be separated when making a forecast.

Labor learning

Workers and teams become faster or make fewer errors as they repeat and refine tasks.

Process and product redesign

Engineering, tooling, and workflow changes reduce material or time per unit.

Scale and supplier learning

Higher cumulative volume may enable better utilization, purchasing terms, and supplier process improvement.

A continuum, not a switch

An experience curve captures an empirical cost or performance pattern across cumulative output. Its slope is a local estimate that should be rechecked when products, processes, or inputs change.

LowLittle cumulative learningHighRepeatable learning captured in unit economics

“Experience helps only when the organization learns from it.”

Why it matters

Experience curves can explain why early production is expensive and why training, process engineering, and feedback loops matter. They help firms decide when to standardize, automate, or protect time for learning. But the cost reduction may be captured by customers through lower prices or lost through input inflation, defects, or changing specifications.

Wright’s paper analyzed aircraft production costs and explicitly discusses adjusting a labor curve as new data became available. Toyota’s official production-system account also emphasizes iterative improvement and work routines. The examples support disciplined learning; neither justifies assuming a fixed learning rate for another industry or era.

Real-world examples

The same concept shows up in different ways across industries.

ToyotaLearning embedded in ongoing process improvement

→

Toyota describes continuing improvement within its production system. This shows an organizational method for acting on experience; the article does not assign Toyota a numerical experience-curve rate or infer cost savings from the description.

Curtiss-WrightEarly aircraft cost-curve analysis

→

Theodore Wright’s 1936 paper, written at Curtiss-Wright, developed aircraft cost curves from prior production experience and revised them with later data. Its historical method is relevant; the original observations are not a universal benchmark for present-day production.

When it breaks

Forecasts break when volume is mistaken for learning. Repeating a defective process can reinforce waste. Track unit-level causes and operational changes, and check quality, yield, labor hours, and total cost rather than assuming each doubling lowers expense.

A company can lower reported unit cost by shifting work to suppliers, changing product mix, or deferring maintenance. Keep the system boundary and quality consistent. Identify constraints and learning investments, and test whether competitors can learn at a similar pace.

Cumulative output and time commonly rise together, so a fitted relationship can also capture changing input prices, technology or product mix. Separate a forecast fit from evidence that cumulative experience caused the decline. Check whether the model uses unit cost, cumulative average cost or labor hours; these are distinct dependent variables.

Key takeaways

01
Define cumulative output and the exact unit outcome.
02
Estimate a learning rate from comparable production data.
03
Separate process learning from scale, mix, and input-price effects.

Sources

Factors Affecting the Cost of Airplanes · T. P. Wright, Journal of the Aeronautical Sciences. PDF index 0, printed p. 122, empirical curve construction and correction with new observations; indexes 2–4, distinct labor/material/overhead mechanisms and changing complete-airplane slope.
Toyota Production System · Toyota Motor Corporation. Jidoka, stopping and improvement; Just-in-Time, continuous flow, minimum parts stock and replenishment.