Deal-Close Forecast — payment-verified ML forecast of which deals actually close
Know which deals will actually pay — measured against money received, not CRM optimism.
What was getting in the way.
CRM close dates are aspirational: deals marked "closing this week" for a month, and won dates edited after the fact. Leadership couldn't trust the pipeline view for cash planning, and the handoff from sales to onboarding depended on someone noticing a deal was about to land.
Context
The marketing team’s own sales pipeline (prospective local-business clients).
How the work runs.
- 01
Trigger
scheduled runs (forecast reads on demand; watcher on a 2-hour cycle; monthly retrain).
- 02
Ground truth is the client's
Ground truth is the client's first successful payment — not the CRM close date. Payment history syncs nightly and joins to deals.
- 03
trained model scores every open
A trained model scores every open deal: probability of first payment within 14 and 30 days, plus a predicted close week.
- 04
Output includes a "lie factor"
Output includes a "lie factor" — how CRM close dates historically compare to actual payment dates — and zombie flags for deals long past their stated close.
- 05
watcher escalates high-probability deals to
The watcher escalates high-probability deals to leadership for onboarding prep, with cooldowns and dedupe so it never nags.
- 06
Outcome
a calibrated, payment-verified pipeline view. Honest status note: the escalation watcher is currently PAUSED at leadership's request (2026-09-18); the forecast itself remains available on demand.
Evidence from the workflow.
Each system has a role.
Intake
CRM (deals, stage history)
Record of truth (ground truth)
Payments platform
Action (scoring)
ML model (local)
Action (escalations — paused)
Team chat
Why this is Operator.
Owns a workflow end to end, running the process inside the authority you set.
Watcher escalates internally only, with per-deal dedupe and cooldowns; currently paused per leadership instruction (paused ≠ deleted; resumes on request). Internal test deals are filtered from any client-facing readout. Retrain is calibrated against holdout data before replacing the model.
Effects / Outcomes
Model trained on 1,468 pipeline deals (121 won / 1,325 lost), with 84% of won deals matched to a verified first payment.
Calibration holds where it matters: deals scored ≥50% closed at a 75% actual rate on holdout (14-day head, AUC 0.696).
Quantified the CRM honesty gap: 17.4% of won deals actually paid later than their recorded close date — leadership now sees both numbers side by side.
Find where a workflow like this fits.
Start with the systems, work, constraints, and authority already present in your operation.