Renewal forecasting with AI (L4)

The steps

1. 01

   Aggregate 12 months of historical data

   To build a reliable AI model, you must aggregate 12-24 months of historical customer data into a single source of truth (the 'Flat File'). • What to do: Export data from your CRM (Salesforce/HubSpot), CS platform (Gainsight/Totango), and Support ticketing system (Zendesk). • Exact settings: Extract 'Account ID', 'Contract End Date', 'Churn/Renewal Status', 'Monthly Active Users (MAU)', 'Support Ticket Volume', and 'CSM Sentiment Score'. • Data prep: Use a tool like BigQuery, Snowflake, or even a robust Excel sheet. Ensure every row represents a unique contract period. • Example Query: SELECT AccountID, SUM(LoginCount) as Usage, AVG(CSAT) as Sentiment FROM DataPool GROUP BY AccountID. • Owner: RevOps or Data Analyst. • Time Estimate: 10-15 hours. • Pitfalls: Avoid 'Data Leakage',ensure you aren't including data from *after* a churn event occurred to train the model, or it will 'predict' the past perfectly but fail in the future. • Definition of Done: A cleaned dataset with at least 500+ historical renewal outcomes ready for model training.

2. 02

   Configure the AI forecasting model

   You need an environment to train your predictive model. If you don't have a data science team, use 'AutoML' tools like Amazon SageMaker Canvas, Google Vertex AI, or specialized platforms like Akkio. • What to do: Upload your high-quality CSV from Step 1. Select your 'Target Variable' (the column indicating if a customer renewed or churned). • Settings: Choose a 'Binary Classification' model type. Map your input features like 'Days since last login' and 'Open high-priority tickets'. • Configuration: Set hisorical data as the training set (80%) and a small portion as the test set (20%) to validate accuracy. • Owner: RevOps Lead. • Time Estimate: 3-5 hours. • Pitfalls: Including 'Account Name' as a feature; the model might learn names instead of behaviors. • Definition of Done: A trained model showing an 'Accuracy' or 'F1 Score' of at least 80% on the test dataset.

3. 03

   Generate explainable probability scores

   The AI must provide both a probability (0-100%) and 'Feature Importance',essentially the 'why' behind the number. • How to do it: In your AutoML tool, enable 'Explainability' settings (often called SHAP values). • Example output: 'Account X has a 40% renewal probability. Top drivers: 30% decrease in usage, 4 open bugs, 0 executive touchpoints in 90 days.' • Integration: Use an API or Zapier to push these probabilities back into your CRM fields: 'AI_Renewal_Score' and 'AI_Churn_Drivers'. • Owner: RevOps or Systems Admin. • Time Estimate: 4-6 hours. • Pitfalls: Ignoring the human element. If a score is 90% but the driver is 'High usage' and the CSM knows the company just went bankrupt, the model is missing context. • Definition of Done: Every active account in the CRM displays a dynamic AI probability score and the primary reason for that score.

4. 04

   Implement the CSM override workflow

   To improve the model, you must track where the human disagrees with the AI. • What to do: Create two separate fields in Salesforce/HubSpot: 'AI_Forecast_Amount' and 'CSM_Forecast_Override'. • Process: Require CSMs to view the AI score before submitting their monthly forecast. If they change the number by more than 20%, they must select a 'Reason for Override' (e.g., 'Merger/Acquisition', 'Point of Contact Left'). • Tracking: Use a 'Field History Tracking' report to log the original AI number vs. the final human number. • Owner: CS Manager. • Time Estimate: 2 hours (setup) + ongoing. • Pitfalls: CSMs blindly following the AI to avoid being 'wrong,' or completely ignoring it out of spite. • Definition of Done: Reporting dashboard showing 'AI vs. Human' forecast variance across the entire book of business.

5. 05

   Present the dual-forecast to Finance

   The final goal is to give the CFO a forecast that removes 'Happy Ears',the tendency of CSMs to be overly optimistic. • What to do: Create a 'High-Stakes Forecast' dashboard in your BI tool (Tableau, PowerBI, or CRM Dashboards). • Layout: Display three columns: 1. Total Renewable Pipeline, 2. Sum of CSM Forecasts, 3. Sum of AI Weighted Forecasts (Probability * Contract Value). • Strategy: Present the 'AI Number' as the conservative floor and the 'CSM Number' as the ceiling. • Formula: WeightedValue = [Contract_Value] * [AI_Renewal_Score_Decimal]. • Owner: VP of CS and CFO. • Time Estimate: 3 hours. • Pitfalls: Over-relying on the AI during the first 3 months. Use this time to 'shadow' the AI before making budget decisions based on it. • Definition of Done: A monthly CFO-ready report where the AI forecast is the primary benchmark for NRR (Net Retention Rate) projections.

Next playbooks

• CSM call summaries via Gainsight + AI (L2)
• AI-generated QBR decks (L3)
• Intercom Fin AI for support deflection (L3)
• Support deflection with AI search (L3)