Why Two Identical Deals in Your CRM Can Have 85% and 15% Real Odds

Two $120,000 deals can sit at the identical “70%” in a CRM and have wildly different real odds of closing — one at 85 percent, the other at 15 — because stage-based forecasting only tracks where a deal sits in a pipeline, not what the buyer is actually doing. That’s the diagnosis from Fluint’s Nate Nasralla, who lays out a framework for deal-level, behavior-driven forecasting built to replace it. His argument: every deal at a given CRM stage gets treated as equally likely to close, which masks exactly the variance that determines whether a forecast is worth trusting. His fix is a “Plays → Predictors → Outcomes” model, where seller actions (plays) drive observable buyer behaviors (predictors like multi-threading or a confirmed timeline) that in turn predict win rate, cycle time, and deal size — using signals like stakeholder count, CFO attendance, and response speed instead of a rep’s stage guess.

Building it requires six specific pieces: a time-series event store instead of stage snapshots, derived feature engineering, three outcome labels instead of a binary win/loss, interpretable models (regression or gradient-boosted trees, not black boxes), measurement of each play’s actual lift, and nightly retraining. For teams without 18 months of clean CRM history, Nasralla suggests a cold-start workaround using existing calendar and email records instead of waiting to accumulate data. Operationally, he recommends forecast calls focus specifically on gaps of 20-plus points between a rep’s number and the model’s — the disagreements, not the averages, are where the real forecasting work happens.

The stakes for getting this wrong are already quantified elsewhere: Outreach estimates that eighteen people spending twenty combined hours a week on manual, spreadsheet-based forecasting costs one organization roughly $83,200 a year in labor alone — separate from the cost of the bad decisions that stale, error-prone spreadsheets produce. Nasralla’s summary line cuts to the point: “the stage was never the thing predicting the outcome. The buyer’s behavior was, and we just didn’t have a way to see it.”