Ask why the forecast is ignored and you will usually be told it is inaccurate. Ask how inaccurate, and the room goes quiet. Nobody is scoring it.
Score it in public
Publish forecast error every cycle, by owner, without commentary. Accuracy improves within two quarters in almost every case we have measured, and it improves before anyone touches the model.
Accuracy improves before anyone touches the model.
Separate the number from the target
A forecast that doubles as a commitment will be negotiated rather than estimated. Keep the honest number and the agreed number in different columns, and let the gap be the conversation.
Let overrides cost something
Overrides are legitimate — but record who made them and how they performed. An unrecorded override is a decision that escapes review.
