Building a forecast you can defend in the room
Confidence intervals do not survive contact with a board meeting. What survives is being able to show your work.
Every forecasting tool sells accuracy. Almost none of them help with the actual moment of difficulty, which is a room full of people asking why the number moved.
Accuracy is table stakes and not the point
A forecast that is accurate but unexplainable gets overridden by whoever is most senior in the room. The override is usually worse than the model. But it happens anyway, because a number nobody can interrogate is a number nobody will stand behind.
- Show which deals moved and why the model changed its mind about them.
- Separate structural change from noise, so a bad week does not read as a trend.
- Keep the history. Being able to show what the model believed six weeks ago, and whether it was right, is what converts scepticism into trust.
The forecast that wins the room is not the most accurate one. It is the one whose author can answer the third follow-up question.
The tier that matters
Split verified from inferred. A deal where the champion confirmed budget and timing in writing is a different object from one where the model inferred momentum from activity patterns. Both belong in the forecast. They should never be summed into a single undifferentiated number.
