SD-301e · Module 1
Multi-Input Forecast Models
3 min read
Stage probability is the baseline. Layering additional inputs improves accuracy, but only if the inputs are reliable. The second input is velocity — how fast the deal is moving through stages compared to the historical norm. A deal that reaches Stage 3 in fifteen days when the average is thirty has higher probability. A deal that took sixty days has lower. The third input is engagement — meeting frequency, stakeholder count, response latency. The fourth is deal characteristics — size, industry, new versus expansion. Each layer narrows the confidence interval. Four inputs typically reduce forecast error from 12% to 8%. Diminishing returns kick in after four.
Do This
- Start with stage probability as the baseline, then layer velocity and engagement
- Add inputs one at a time and measure whether each improves accuracy on historical data
- Rebuild the model quarterly to capture changes in your sales process and market
Avoid This
- Throw twenty inputs into the model and assume more complexity means more accuracy
- Use inputs that reps self-report unreliably — the model amplifies data quality issues
- Build the model once and never recalibrate — your sales process changes, the model must follow