Design to Cost · Toolkit

Learning Curve

Each doubling of cumulative volume cuts unit cost by a predictable percentage — use it to plan cost-down, not just observe it.

Time30 min
FormatSolo
StageToolkit

Learning Curve: what it is and why it works

The learning curve describes how the labor and process content of a product falls as cumulative production grows: each time cumulative volume doubles, the cost or hours per unit fall to a fixed percentage of their previous value. With an 85% curve, unit 2 takes 85% of unit 1, unit 4 takes 85% of unit 2, and so on. Mathematically, Y = A x N^b, where A is the first-unit value, N the cumulative unit number and b = log2(learning rate), about -0.234 for 85%. Wright's original 1936 formulation applied this to the cumulative average; the unit formulation, where Y is the cost of the Nth unit, is also widely used. State which one you use, since they give different numbers.

The curve is useful because it turns an observed phenomenon into a planning tool. Target costs, quotes for multi-unit contracts and supplier cost-down paths can all be set from a credible rate rather than from wishful percentages. It applies mainly to labor-intensive and process-driven cost, such as assembly hours, welding, testing and setup, and much less to purchased material. The curve holds only while the product and process are stable; a redesign, a new plant or a long production break resets or degrades it. Learning rates feed Cost-Down Agreements, add quantity effects to Parametric Costing, and help interpret the high costs measured in Prototype Costing.

What you need

Unit hours or unit costs against cumulative unit number for stable products
Records of design, process or staffing changes that could break the series
Separation of labor or process cost from material and purchased parts
Planned volumes and build sequence for the new product

What you get

A fitted learning rate with the formulation used (unit or cumulative average)
Validation across several products, with the range of observed rates
Planned unit and average costs for the new product by lot or year
A quarterly comparison of actuals against the curve

When to use it

When cost targets ignore that the first thousand are the most expensive.

How to do it, step by step

Collect unit costs against cumulative volume for a stable product.
Fit the learning curve: cost drops by a fixed percentage per doubling.
Validate the rate on several products before trusting it.
Apply the curve to plan cost-down for new products.
Check reality against the curve quarterly and re-fit.

Worked example: Pricing a 50-unit order of modular pump stations

Illustrative scenario — figures are realistic but not from a real company.

A manufacturer of prefabricated pump stations for municipal and industrial water customers received an inquiry for 50 identical stations delivered over two years. The first-unit assembly and testing estimate was 620 labor hours at a loaded rate of $85 per hour. Sales wanted to quote 620 hours per unit to be safe.

The cost engineer took unit hours from three previous multi-unit orders and fitted a unit learning curve to each on a log-log plot. The rates ranged from 86% to 89%, with 87% as the central value.
Applying an 87% unit curve (b about -0.20) gave about 540 hours for unit 2, 410 for unit 8, 310 for unit 32 and 283 for unit 50.
The average across all 50 units came to roughly 347 hours, about $29,500 of labor per station instead of the $52,700 implied by the first-unit estimate.
The team quoted labor on an 89% curve, the conservative end of the history, and planned internally on 87%.

Result. The quote carried about $20,000 less labor per station than the first-unit approach, and the company won the order. Actual hours tracked slightly better than 87% until a customer-requested design change at unit 21 bumped hours back up by about 12%, after which learning resumed. The team added a change-reset allowance to future quotes.

Open it in the FieldKit Design to Cost app → Get the Design to Cost template pack — join the waitlist

Common pitfalls and how to avoid them

Applying the learning rate to total cost including purchased material.Apply it only to the cost that learns, mainly labor and process time, and handle material separately.
Mixing unit and cumulative average formulations.Choose one formulation, state it and use it consistently in fitting and forecasting.
Assuming learning continues through redesigns, relocations or long breaks.Reset or degrade the curve after major changes and flag interruptions in the data.
Trusting a rate fitted on one product or a few units.Validate the rate on several comparable products and use a conservative value for commitments.

Frequently asked questions

What does an 80% learning curve mean?

It means that every time cumulative production doubles, the unit cost or hours fall to 80% of their previous value, a 20% reduction per doubling. If the first unit takes 1,000 hours, the second takes 800, the fourth 640 and the eighth 512 under a unit formulation. The exponent is log2(0.80), about -0.322.

What is the difference between the Wright and Crawford learning curves?

Wright's model, from 1936 aircraft production, applies the learning rate to the cumulative average cost of all units produced so far. The unit model often attributed to Crawford applies it to the cost of each individual unit. For the same rate, they give different values, so fitting and forecasting must use the same formulation.

Does the learning curve apply to all types of cost?

No. It applies best to labor-intensive and process-driven activities such as assembly, fabrication, testing and setup, where repetition improves methods, tooling and skills. Purchased materials and components follow supplier pricing and commodity markets, and fixed overhead does not scale with units, so they should be treated separately.

Origin

Learning curve — T. P. Wright, 1936; unit curve theory in aircraft production.

Related methods

More in “Toolkit”

Cross-cutting methods: curves, complexity, quality and estimating tools.