Learning Curve Calculator: How to Use It

This learning curve calculator works out the time (or cost) needed to produce a given unit when a known learning rate applies. You enter the time for the first unit, the learning rate (as a percentage), and the unit number, and you get back the predicted time for that unit plus the cumulative average time per unit. It uses the standard Crawford and Wright learning curve formulas.

Formula

The learning curve is modeled by the equation:

T(n) = T(1) × n^b

where:

• T(n) = time (or cost) to produce the nth unit (hours, minutes, dollars, etc.)
• T(1) = time to produce the first unit
• n = the unit number (1, 2, 3, …)
• b = the learning exponent = log(learning rate) / log(2)

The learning rate is expressed as a decimal (e.g., 80% = 0.80). The exponent b is negative for learning rates below 100%, because each doubling of output reduces the unit time by a fixed percentage. For an 80% learning curve, b = log(0.80)/log(2) ≈ -0.3219.

Worked Example

Using the default values: first unit time = 100 hours, learning rate = 80%, unit number = 10.

1. Convert the learning rate to a decimal: 80% = 0.80.
2. Compute the exponent: b = log(0.80)/log(2) = -0.3219 (using natural logs).
3. Compute 10^b = 10^-0.3219 ≈ 0.4765.
4. Multiply by the first unit time: 100 × 0.4765 = 47.65 hours.

The calculator shows 47.65 hours for the 10th unit. The cumulative average time formula gives a slightly different value because it averages all units from 1 to n. For n=10, the cumulative average is about 62.06 hours per unit (total time for 10 units divided by 10).

When to Use a Learning Curve

Learning curves apply whenever a repetitive task shows predictable improvement with experience. Common uses include:

• Manufacturing: estimating labor hours for assembly, machining, or quality checks.
• Construction: predicting time for repetitive tasks like laying bricks or installing drywall.
• Software development: estimating time for similar coding tasks or bug fixes.
• Healthcare: projecting time for surgical procedures or diagnostic tests as staff gain experience.
• Cost estimation: predicting unit costs for pricing, budgeting, or bids.

The curve is most accurate in the early stages of production (units 1–100) and becomes less reliable as the process matures and improvements plateau.

How to Read the Results

The first output, Time for unit n, is the predicted time for that specific unit. The second output, Cumulative average time, is the average time per unit across all units from 1 to n. The cumulative average is always higher than the individual unit time because early units take longer.

For example, with a 100-hour first unit and 80% learning rate:

• Unit 1: 100 hours
• Unit 2: 80 hours (100 × 0.8)
• Unit 4: 64 hours (80 × 0.8)
• Unit 8: 51.2 hours (64 × 0.8)

Each doubling of cumulative production reduces the unit time by 20% (for an 80% learning rate). The cumulative average after 8 units is about 70.5 hours per unit, not 51.2.

Common Mistakes and Misconceptions

• Using the wrong learning rate. The learning rate is the percentage of time remaining after each doubling, not the percentage improvement. An 80% learning rate means the second unit takes 80% of the first unit’s time, not 20% less than the first unit (though that is equivalent).
• Confusing unit time with cumulative average. The time for unit n is not the same as the average time across all units. The cumulative average is always higher.
• Applying the curve beyond its valid range. Learning curves assume a stable process. If the process changes (new tools, new workers, automation), the curve resets.
• Using the wrong base for logarithms. Any base works as long as you use the same base for both numerator and denominator. Most calculators use natural log (ln) or base-10 log; the ratio is the same.
• Rounding too early. Round the exponent to 2–3 decimal places only at the end, otherwise the result can be off by several percent.

Limitations of the Learning Curve

The model assumes a constant learning rate, which is rarely true in practice. Early improvements are often rapid, then slow down. The curve also ignores external factors like breaks, shift changes, material variability, and worker fatigue. For small batch sizes (n < 5), the curve is highly sensitive to the first-unit estimate, so a small error in T(1) can produce large errors in predictions.

Additionally, the learning rate itself is an estimate. It is usually derived from historical data or industry benchmarks, but it varies by task and team. If you do not have a reliable learning rate, run a few pilot units and fit the curve to actual data.

Alternatives and Related Models

• Wright’s cumulative average model – uses the average time per unit, not the individual unit time. This calculator provides both, but the individual unit formula is Crawford’s.
• Experience curve – applies the same concept to total costs, including materials and overhead, not just labor.
• Log-linear regression – if you have real production data, you can fit a line to log(time) vs. log(unit number) to estimate the learning rate empirically.
• Plateau models – some tasks stop improving after a certain point; a pure learning curve will underestimate time if you extrapolate too far.

Frequently Asked Questions

What is a good learning rate?

Typical learning rates range from 70% to 95%. A lower percentage means faster improvement. For example, 70% is common for complex manual assembly, while 95% might be seen in highly automated processes. There is no universal “good” rate; it depends on the task.

Can I use this for cost instead of time?

Yes. The formula works identically if you substitute cost for time. Just enter the cost of the first unit in the “Time for first unit” field and read the result as cost. The unit label is arbitrary.

What if I only know the time for unit 5, not unit 1?

You can work backwards: divide the known time by n^b to get T(1). For example, if unit 5 takes 60 hours with an 80% learning rate, then T(1) = 60 / (5^-0.3219) ≈ 60 / 0.595 ≈ 100.8 hours. Then use that as your input.

Does the learning curve apply to software development?

It can, but with caution. Software tasks are often less repetitive than manufacturing. The curve works best for similar, well-defined tasks like writing the same type of module or running the same test suite repeatedly. It is less reliable for novel features.

How do I find the learning rate from my own data?

Collect times for several units, then plot log(time) vs. log(unit number). The slope of the best-fit line is the exponent b. The learning rate is 2^b. For example, if the slope is -0.3219, the learning rate is 2^-0.3219 = 0.80, or 80%.

What is the difference between Crawford and Wright models?

Crawford’s model predicts the time for a specific unit, while Wright’s model predicts the cumulative average time per unit. Both use the same learning rate, but Wright’s cumulative average is higher for any n>1. This calculator gives both, so you can use either.

Related

• Curve Calculator Grade: How to Find It
• Curve Calculator App: What It Does & How to Use
• Curve Calculator Radius: Find the Radius

in t1 | Time for first unit | hours | 100 | 0
in rate | Learning rate | % | 80 | 1
in n | Unit number | - | 10 | 1
out tn | Time for unit n | hours | t1 * (n ^ (log(rate/100) / log(2))) | 2
out avg | Cumulative average time | hours | (t1 * (n ^ (log(rate/100) / log(2))) - t1 * ((n-1) ^ (log(rate/100) / log(2)))) / (n - (n-1)) | 2