Cm Renewal Forecaster

Renewal Forecaster

Produces probability-weighted renewal forecasts by combining health signals, usage data, engagement patterns, and historical outcomes. Part of the Commercial Motion pillar.

This is an analytical and reporting skill. It produces forecasts and projections. It does not make retention decisions or set strategy -- those are human calls informed by the forecast.

When to Run

• Scheduled: Weekly forecast refresh for all accounts renewing in next 180 days
• On-demand: When leadership requests a forecast update
• Triggered: When a significant risk or health signal changes on an account within the renewal window

Core Execution Logic

Step 1: Identify Renewal Cohort

Pull all accounts with renewal dates in the next 180 days. For each, assemble:

• Current health score and trend (from bi-health-score)
• Risk signal count and severity (from bi-risk-detector)
• Usage trend classification (from bi-usage-monitor)
• Engagement pattern (from ra-engagement-tracker)
• Renewal risk classification (from lo-renewal-manager: On Track / Watch / At Risk / Critical)
• Expansion signals (from bi-expansion-detector -- renewals with expansion opportunity are a different conversation)
• Contract details (ARR, term, auto-renewal clause, notice period)

Step 2: Assign Renewal Probability

Default probability model (use until historical calibration data is available):

Probability bounds: Floor at 5% (even "certain" churns occasionally reverse). Cap at 98% (even "certain" renewals occasionally surprise).

Calibration: After 12 months of data, replace the default model with historically-calibrated probabilities. See references/probability-calibration.md.

Step 3: Compute Portfolio Forecast

Aggregate individual account probabilities into portfolio-level projections:

Step 4: Identify Actionable Accounts

Flag accounts where intervention could meaningfully change the forecast:

Step 5: Generate Forecast Report

{
  "forecast_date": "2026-03-10",
  "horizon": "next_180_days",
  "portfolio": {
    "renewing_accounts": 42,
    "renewing_arr": 3200000
  },
  "forecast": {
    "most_likely": {
      "retained_arr": 2880000,
      "churn_arr": 320000,
      "expansion_arr": 180000,
      "nrr": 0.956
    },
    "optimistic": { "nrr": 0.982 },
    "pessimistic": { "nrr": 0.921 }
  },
  "by_classification": {
    "on_track": { "count": 28, "arr": 2100000, "avg_probability": 0.94 },
    "watch": { "count": 8, "arr": 680000, "avg_probability": 0.72 },
    "at_risk": { "count": 4, "arr": 340000, "avg_probability": 0.41 },
    "critical": { "count": 2, "arr": 80000, "avg_probability": 0.13 }
  },
  "actionable_accounts": [
    {
      "account": "Acme Corp",
      "arr": 85000,
      "renewal_date": "2026-05-15",
      "days_remaining": 66,
      "classification": "at_risk",
      "probability": 0.45,
      "category": "saveable_at_risk",
      "top_risk": "Champion departed February, usage declining since",
      "intervention_opportunity": "New VP Engineering hired last week -- window to establish relationship before renewal window",
      "csm": "Jane Doe"
    }
  ],
  "period_over_period": {
    "nrr_change_vs_last_week": -0.003,
    "accounts_worsened": 2,
    "accounts_improved": 1,
    "new_at_risk": 1
  }
}

Handoff to Human

Confidence and Limitations

• High confidence for timeline tracking and milestone computation -- deterministic calendar logic
• Medium confidence for probability estimation using the default model -- health and signal data are strong predictive inputs, but the model has not been calibrated against your specific portfolio's churn patterns until you have 12 months of data
• Medium confidence for portfolio aggregation -- individual account probabilities may be off, but errors tend to cancel out at the portfolio level (some pessimistic, some optimistic). The aggregate forecast is more reliable than any individual account prediction
• Low confidence for accounts with limited data history (<6 months) or accounts where the risk is purely relationship-based with no data signal
• Probability estimates should be treated as directional, not precise. "75% renewal probability" means "more likely than not with moderate risk," not a guarantee
• The forecast does not account for portfolio-level events (pricing changes, product launches, market shifts) that may move multiple renewals simultaneously. The human must factor these in
• The optimistic/pessimistic scenarios are bounded estimates, not worst-case/best-case. True worst case (major product failure, market crash) is outside the model

Dependencies

Required:

• lo-renewal-manager (renewal classifications, timeline data)
• bi-health-score (health data for probability computation)
• CRM API (contract data, ARR, renewal dates)

Strongly recommended:

• bi-risk-detector (risk signal count and severity)
• bi-usage-monitor (usage trend for trajectory assessment)
• bi-expansion-detector (expansion signals for combined renewal-expansion opportunities)
• ra-engagement-tracker (engagement pattern for probability adjustment)

Downstream consumers:

• CS leadership (weekly forecast review)
• CRO/CFO (monthly/quarterly revenue planning)
• Board reporting (quarterly NRR projection)
• ic-cross-func-prep (pipeline review materials)

References

• references/probability-calibration.md -- Historical calibration methodology, minimum data requirements, and model validation process