Build an account scoring model from closed-won data

Prerequisites

• 30+ closed-won deals in CRM
• Warehouse access (BigQuery or Snowflake)
• CRM API access for write-back

Steps

1. Export closed-won. Pull all won opportunities with account attributes, technographics, and signal history.
2. Feature engineering. Extract firmographic, technographic, and behavioral features in dbt.
3. Train model. Logistic regression or gradient boosting with Laplace smoothing if sample < 50.
4. Score TAM. Run model against full account universe.
5. Write to CRM. Sync scores as custom property with last-updated timestamp.
6. Validate. Compare conversion rate by score decile over 30 days.

Success criteria

Top decile converts at 3x+ the bottom decile. Reps report using scores for prioritization.