Key takeaways
– The learning curve (also called the experience curve, efficiency curve, or cost curve) describes how the time or cost to produce a unit falls as cumulative production rises and people gain experience.
– The standard mathematical form is Y = a X^b, where Y is the cumulative average time per unit after X units, a is the time for the first unit, and b = log(learning rate)/log(2).
– A “90% learning curve” means that every time cumulative production doubles, the cumulative average time per unit falls to 90% of its previous value (a 10% improvement).
– Manufacturers can use learning curves for cost forecasting, pricing/bidding, workforce planning, process improvement, scheduling and procurement.
Background and definition
– Origin: The concept of learning from repetition dates to psychologist Hermann Ebbinghaus (1885) and was later formalized for manufacturing and cost analysis.
– Practical meaning: As a worker or process repeats a task, learning and efficiency gains generally reduce the time and cost per unit. Gains tend to be large early on and then flatten as the process matures.
Learning-curve formula and interpretation
– Formula: Y = a X^b
• Y = cumulative average time per unit (or per batch) after producing X units
• a = time (or cost) for the initial unit (X = 1)
• X = cumulative units produced
• b = learning-curve exponent = log(learning rate)/log(2)
– Learning rate: expressed as a percentage (for example, 80%, 90%). If the learning rate is L, then b = log(L)/log(2). When production doubles (X → 2X), Y_new = a (2X)^b = a X^b × 2^b = Y_old × L.
• Example: L = 0.90 (90%): b = log(0.9)/log(2) ≈ −0.1520. Doubling cumulative production reduces the cumulative average time per unit to 0.9× previous.
• Common interpretation confusion: in everyday language “a high learning curve” often means something is hard to learn. Technically, a higher learning-rate percentage (closer to 100%) means slower improvement; a lower percentage (e.g., 80%) means faster improvement.
Worked numerical examples
– 90% learning curve, initial time a = 1,000 hours:
• X = 1 → Y = 1,000 hours average for 1 unit.
• X = 2 → Y = 1,000 × 2^(log(0.9)/log(2)) = 1,000 × 0.9 = 900 hours average → total cumulative hours = 900 × 2 = 1,800 → incremental time for unit 2 = 800 hours.
• X = 4 → Y = 1,000 × (0.9)^2 = 810 hours average → total cumulative hours = 810 × 4 = 3,240 → incremental time for units 3–4 = 3,240 − 1,800 = 1,440 (average 720 per unit for units 3–4).
– 80% learning curve, a = 1,000 hours:
• X = 2 → Y = 1,000 × 0.8 = 800 hours average → total = 1,600 → second unit incremental = 600 hours.
• X = 4 → Y = 1,000 × (0.8)^2 = 640 hours average → total = 2,560 → incremental for units 3–4 = 960 hours (average 480 each).
Graphing learning curves (practical tips)
– Linear axes: plotting cumulative total hours versus cumulative units will slope upward (more total time for more units) but hides efficiency gains.
– Average-per-unit plots: plotting cumulative average time-per-unit versus cumulative units shows the downward-sloping learning curve.
– Log-log plot: plot ln(cumulative average time) versus ln(cumulative units). The learning-curve model becomes a straight line; slope = b. This is the usual way to estimate b from data (linear regression).
– Excel quick method: compute cumulative average times, take LN of X and LN of Y, use LINEST or SLOPE to estimate b and intercept. Then a = exp(intercept).
How to estimate a learning curve from real data (step-by-step)
1. Collect data
• Record unit-level or batch-level production times (or costs) and cumulative units produced.
• Use consistent task definitions (same product variant, same scope of work).
2. Compute cumulative average time (Y) at each cumulative production level X: Y = (total cumulative time to produce X units) / X.
3. Convert to logs: compute ln(X) and ln(Y).
4. Fit a linear regression: ln(Y) = ln(a) + b · ln(X). The slope is b; intercept = ln(a).
5. Calculate learning rate L = 2^b (because when X doubles, factor = 2^b). Alternatively L = exp(b · ln 2).
6. Validate: check R^2, residuals, and whether the model fits early period vs later period data. If process changes occurred (automation, new tooling), estimate separate learning rates for each regime.
7. Periodically update the model as more data arrives.
Practical steps for manufacturers — how to use learning-curve information
1. Cost forecasting and quoting
• Use predicted Y values combined with labor and overhead rates to forecast unit costs over planned volumes.
• For bids, include realistic learning assumptions and sensitivity ranges (best-case/worst-case).
2. Pricing and break-even analysis
• Map expected cost declines to pricing strategy over production runs, especially for long production ramps or new product introductions.
3. Capacity and workforce planning
• Convert expected hours per unit into staffing needs over time. Plan hiring, training schedules, and overtime rather than using constant-per-unit times.
4. Scheduling and lead times
• Use predicted reductions in unit time to build more accurate production schedules and expected completion dates.
5. Inventory and procurement
• Align material purchasing and supplier lead times to the projected production curve; some suppliers require longer lead times early, before learning effects reduce internal bottlenecks.
6. Process improvement and automation decisions
• Compare projected manual-learning savings versus capital investment in automation. If learning curve flattens early and volumes are high, automation may pay sooner.
7. Training programs
• Identify how much faster workers become productive and design training resources to accelerate that slope.
8. Risk management and sensitivity analysis
• Run scenarios with several learning rates (e.g., 80%, 90%, 95%) to see financial and schedule impacts. Use conservative and optimistic cases in decision-making.
Limitations and cautions
– Model assumes consistent process and product: changes in product design, materials, or process invalidate a single learning rate.
– Worker turnover, shift changes, or skill mix changes will alter observed learning.
– Batch-size effects: different lot sizes can distort cumulative averages.
– Productivity improvements from deliberate process change (automation, tooling) are not the same as “learning” — treat them as separate regime changes.
– The simple model captures a smooth decline; real-world data may have noise and discontinuities.
Practical Excel example (quick formulas)
– Given a table of cumulative units (X) and cumulative total time (T_total):
• Cumulative average Y = T_total / X
• In Excel: LN_X = LN(X), LN_Y = LN(Y)
• Slope b = SLOPE(LN_Y_range, LN_X_range)
• Intercept ln(a) = INTERCEPT(LN_Y_range, LN_X_range)
• a = EXP(intercept)
• Learning rate L = 2^b
• Predict Y for any X: = a * X^b
What does a “high” learning curve mean?
– Everyday usage: “high learning curve” often means something hard to master.
– Technical usage: A high learning-rate percentage (e.g., 98%) means small improvements when production doubles (slow learning). A low percentage (e.g., 70–80%) means large improvements (fast learning). Clarify which usage you mean when discussing it with stakeholders.
What a 90% learning curve means (concise)
– Every time cumulative production doubles, cumulative average time per unit falls to 90% of the previous average — a 10% improvement in average unit time when doubles occur.
Fast facts
– Learning curves are widely used in manufacturing, project estimating, and strategic cost analysis.
– The Boston Consulting Group popularized an experience curve concept in the 1960s that extended learning-curve ideas to wider cost declines.
The bottom line
– Learning curves provide a simple, data-driven way to model how per-unit time or cost declines as cumulative production increases. When properly estimated and applied (with attention to regime changes and variability), they improve costing, scheduling, staffing, quoting, and investment decisions. Always validate the model with actual production data and run sensitivity tests for planning.
Related reading (suggested topics)
– Experience curve and experience effects
– Process improvement and lean manufacturing
– Cost estimation techniques and regression analysis
– Workforce learning and training program design
Editor’s note: The following topics are reserved for upcoming updates and will be expanded with detailed examples and datasets.