The key idea

No single accuracy percentage explains whether a book forecast is useful. Measure the difference between forecast and actual demand, then examine the inventory consequences under the same operating policy. Separate underprediction from overprediction, and show results for high volume titles, quiet backlist and new editions rather than only one catalog average.

This is an original operational explainer. Its examples and decision rules are illustrations, not measured customer results or universal service commitments.

Start with measures people can interpret

Mean absolute error expresses the typical absolute difference in the original unit, such as copies per week. It is easy to explain, but it does not distinguish expensive shortages from surplus. Percentage errors can become unstable or undefined when actual demand is zero. Report sample size, horizon and excluded periods with every result. A percentage without those definitions can be impossible to compare across titles or markets.

Connect measurement to a decision

A business evaluation should ask how often the stock policy would have run short, how much extra inventory it would have held and how frequently it would have changed production. These results depend on lead time, batch size and service choices, not only forecast quality. Use the same policy to compare candidate forecasts. If results differ by segment, consider a controlled segment specific rollout instead of applying one model everywhere immediately.

Worked example · illustrative assumptions

For actual demand of 10, 0 and 5 copies and forecasts of 8, 2 and 6, absolute errors are 2, 2 and 1. Mean absolute error is 5 divided by 3, about 1.67 copies. A simple percentage error for the zero demand period is not meaningful. The example also shows why the direction matters: the first forecast is short by two while the other two are above demand.

Put it into practice

  • State the unit, horizon and sample behind every accuracy result.
  • Separate underprediction, overprediction and inventory outcomes.
  • Compare models on the same titles and operating assumptions.

Watch out for

Do not advertise an accuracy percentage derived from a synthetic demonstration as a measured customer result. Demonstrations test calculations; real performance claims need an appropriate real evaluation dataset.

Is lower average error always commercially better?

Not necessarily. A modest average improvement can hide more severe shortages on important titles. Evaluate the distribution and business consequences before changing automatic ordering.

AI assisted editorial content with original planning examples. Source links support the stated source notes; operational suggestions are guidance rather than guarantees. Provider rules and product coverage can change. Editorial approach · BooksAhead.ai connection status.