Skip to content
Sales and Operations Planning | Part 4 of 10

Forecast Accuracy Shows Size, Bias Shows Direction

Measure forecast accuracy and bias per family at a fixed lag, and treat a steady bias as a habit to correct, not as noise.

Every planner knows that forecasts are wrong. The useful questions are how wrong, in which direction, and whether it is getting better. Two measures answer them: accuracy and bias.

Measure at a fixed lag

Compare the actual sales of a month with the forecast made a fixed time before. Lag 2 means the forecast made two months before the month. Choose the lag that matches the reaction time of operations, because that is the forecast operations used to plan capacity and materials.

A forecast made last week always looks good. It is also useless for planning.

Accuracy

Forecast accuracy is 1 minus the sum of absolute errors divided by the sum of actual sales, per family over a period. In the pump example, six months from July to December 2026:

  • Compact pumps: 93.7 percent
  • Process pumps: 90.4 percent
  • Engineered systems: 77.3 percent

Weighted by revenue over all families: 88.0 percent.

Do not average monthly percentages. A small month with a large percentage error would count as much as a large month, and the measure would mislead.

Bias

Bias is the sum of forecast minus actual, divided by the sum of actual sales. Positive means the forecast was too high.

  • Compact pumps: plus 2.7 percent
  • Process pumps: plus 9.7 percent
  • Engineered systems: plus 8.2 percent

Within plus or minus 3 percent, errors cancel out over time. That is noise. Process pumps are different: six months in a row the forecast was higher than actual sales. Operations built capacity and bought modules for almost one tenth more than was sold.

That is not noise. It is a habit, and it is the first topic of the next demand review.

Why bias matters more than accuracy

An accurate but biased forecast does steady damage. Stock grows, capacity is reserved for orders that do not come, and nobody notices because each single month looks acceptable.

In the example, the executive review decided to lower the process pump plan by 5 percent and to look for the reason with sales. Typical reasons are optimistic targets entered as forecasts, projects counted at full value, or a fear of running out that is answered with a higher number.

Low accuracy is not always a problem

Engineered systems show 77 percent accuracy because a handful of projects decide every month. More statistics will not change that. For such families, plan capacity instead of stock: reserve engineering and assembly hours per month and fill them with confirmed projects.

Make it routine

Calculate accuracy and bias in week 1 of every cycle, per family, and show both at the start of the demand review. Show the trend, not only the last month. Mark every family with a bias beyond 3 percent and give it an owner.

What is measured at a fixed lag improves. What is not measured is defended in meetings.

Newsletter

One Lean idea.
Every week.