Forecast bias: measure direction, not just error

Absolute error is only half the measure

Absolute error answers a necessary question: how far was the submission from the outcome? It makes over-forecasting and under-forecasting comparable and stops opposite misses cancelling each other out.

Using it alone discards the sign. That removes the difference between systematic optimism, systematic caution and error that changes direction. Those patterns need different responses.

Predictable bias is a gift

A consistent direction tells the team where to look. Repeated over-forecasting may point to qualification, close-date discipline or the point at which Best Case becomes Commit. Repeated under-forecasting may mean upside is recognized too late or that the review punishes visible ambition.

The correction isn't to subtract eight per cent forever. It's to inspect the decisions underneath, find the recurring judgment and test a change in the review process.

Compare like with like

Keep role, submission point, segment, region and deal size visible. A Seller position at the start of the quarter should not be compared blindly with a leadership position submitted days before close.

The same person can also show different bias across enterprise and transactional deals. A single annual number may describe neither. Start with cohorts large enough to be meaningful and narrow enough to share an operating context.

Use bias to improve the review, not rank people

A public league table changes the behavior being measured. People learn which number protects them and adjust the submission before the organization learns anything.

Keep the analysis in the coaching and forecast process. Show the person their pattern, the deals that produced it, the evidence available at the time and whether later Manager or leadership judgment corrected it. The purpose is a better decision, not a label attached to a person.

Also keep the corrections between layers. If Managers repeatedly reduce one Seller's position, or leadership repeatedly restores value that Managers removed, the pattern belongs in the operating review. The question isn't who won the argument. It's which layer saw the outcome most clearly, what evidence it used and whether that information can reach the earlier decision next time.

Where bias analysis goes wrong

Small samples create confident stories from noise. Changing territories, targets or forecast definitions can also create an apparent shift that has nothing to do with judgment. Keep the effective definitions and cohort changes with the result.

Don't treat random error as evidence of poor judgment before checking the record underneath. If the necessary evidence never arrived, the problem may be capture, qualification or decision timing rather than the person submitting the position.

A note on evidence

PROOF doesn't yet publish a benchmark for acceptable forecast bias. A useful threshold depends on horizon, sales cycle, role and the decision the number supports.

The reusable distinction is simpler: distance shows how far the forecast missed. Direction shows how the process tends to miss. Keep both.