A simple forecast model that beat our reps' gut feel

Our sales team's manual forecast, submitted by each rep every Friday, missed the actual quarterly closed number by an average of 22% over the prior four quarters. A basic weighted-pipeline calculation, built in an afternoon, cut that miss to 8% over the two quarters since.

The calculation

forecast = sum(deal_value * stage_weight)
// stage_weight learned from your own historical
// close rate at each stage, not a generic guess

Instead of asking reps to predict which deals will close, which invites optimism bias, we weight every open deal's value by the actual historical close rate of deals at that stage, calculated from our own closed-won and closed-lost history, not an industry benchmark.

Why this outperformed manual forecasts specifically

Rep forecasts were not badly calibrated at the individual deal level; reps were usually right about which specific deals were shaky. The error was systemic and directional: reps as a group were reliably optimistic about timing, consistently pulling deals into the current quarter that historically slipped.

A weighted-stage model does not know or care about any individual rep's optimism about a specific deal closing this week. It only knows what deals at that stage have historically done, which turned out to be a better predictor at the aggregate level than any individual's judgment, even though individual reps often knew their own deals better than the model did.

• Recalculate stage weights quarterly — they drift as your sales process and average deal size change.

• This model is weaker for small pipelines; we would not trust it below roughly 30 open deals in a given period.

• We still ask reps for manual forecasts too — the gap between the two numbers is itself a useful signal worth investigating.