gtmpodStart
REVOPSforecasting · advanced

Pipeline forecasting with AI sanity-check on each deal

Last reviewed: 2026-05-23 · saves ~1 day/month/run

Our take

Forecasts are wrong because the data feeding them is wrong: AE optimism, stale stages, no MEDDIC. Adding AI as a sanity-check on each deal — comparing what the AE said vs what the data says — surfaces 15-25% of deals that need pipeline movement. Forecast accuracy improves by 5-10 percentage points.

Tool stack

Steps

  1. Weekly: AI runs across all stage-3+ opportunities.
  2. For each, compares AE-stated commit/upside/best-case against signal from Gong + email + MAP progress.
  3. Outputs a 'commit confidence score' per deal: high / medium / low / flagged.
  4. RevOps reviews flagged deals with sales manager pre-forecast call.
  5. Adjusted forecast is the one submitted. AI inputs are saved for accuracy tracking.

Prompts

Score commit confidence per deal · Claude Sonnet 4.6
You are a sales operations analyst. For opportunity {ID}, given:

- AE commit category: commit / strong-upside / upside / best-case / pipeline
- AE stated close date + amount
- Last 14 days of email activity (count, sentiment)
- Last 30 days of Gong call data
- MAP (mutual action plan) status: on-track / slipping / not-set
- Stage entry date + time-in-stage vs team median

Output a JSON score:

{
  "commit_confidence": "high" | "medium" | "low" | "flagged",
  "reasoning": "1-2 sentences",
  "risk_signals": ["array of specific signals"],
  "recommended_action": "specific next step for AE or manager"
}

Flag if:
- AE says commit but no buyer activity > 7 days
- AE says commit but MAP shows commitments slipping
- Time-in-stage > 1.5x team median
- Recent Gong call sentiment from buyer trends negative
- Champion has been silent > 10 days

Be specific in risk_signals — quote data, not vibes.

Pitfalls

  • AI optimism vs AE optimism: both can be wrong. Calibrate against actual close outcomes monthly.
  • Don't let AI score deals < stage 3 (too noisy).
  • If a deal is flagged 3+ weeks in a row, it's stuck. Escalate, don't keep flagging.
Last reviewed 2026-05-23. Independent.