Override

Definition

Manually overruling a system recommendation — setting a different rate than the engine proposed, or correcting a forecast by hand. Not a failure of the tool: the override is where information the model cannot see enters the process.

What it tells you

Override discipline separates mature RM teams from nervous ones. The healthy rule: override only on information the model demonstrably lacks (the cause, not a gut discomfort), state the reason in writing, and re-measure. Your override rate is itself a signal — near zero can mean rubber-stamping; consistently high and winning means the model has drifted from reality; consistently high and losing means the overrides are the problem.

How to track it

An override log: date, recommendation, your decision, the stated reason, and the outcome once known. Review it monthly next to the forecast-accuracy report — it is the fastest way to learn whether you or the model reads your market better, and where.

Where it fits

The human half of the human–machine division of labour in ML-based pricing. The free Academy covers it in depth: Pricing Engine — ML-based rate recommendation.