Demandbase predictive scores use machine learning to analyze data from your CRM, Marketing Automation System (MAS), CSV imports, and other available sources.
Follow these best practices when setting up and maintaining Pipeline Predict and Qualification Scores to improve scoring accuracy and identify accounts to prioritize for marketing and sales.
Best Practices for Pipeline Predict Scores
Retrain scores after configuration changes: Retrain your scores whenever you update your engagement points or intent keyword configurations. Auto Retrain is enabled by default. See Retrain Pipeline Predict Scores.
Use recent opportunities: Include opportunities opened in the last 12 months.
-
Include enough accounts for training: Use at least 50 distinct accounts with qualified opportunities.
Important:- The model only considers accounts with engagement activity before opportunity creation.
- For growth (renewal, upsell, or cross-sell) scores, include existing customer accounts when training the model.
- Include as much engagement data as possible: Instead of selecting a limited subset of engagement data, exclude only the data you don't want to include. This allows the model to train on a larger dataset.
- Avoid Pipeline Date filters: Don't use selectors that reference Pipeline Date.
-
Avoid Activity Date filters: Don't use Activity Date filters in the Engagement activities selector. Activity Date is evaluated relative to the model's training date, not the opportunity creation date or score calculation date. Pipeline Predict already accounts for activity recency. To narrow the activities included in the model, filter by activity type or use an activity segment instead.
Important: Remove Activity Date filters from existing models and retrain them. These filters reduce training data and can cause scores to compress over time, preventing accounts from reaching score thresholds.
Best Practices for Qualification Scores
Retrain scores after configuration changes: Retrain your scores whenever you update your technographics or intent keyword configurations. Auto Retrain is enabled by default. See Retrain Qualification Scores.
- Use recent customer accounts: Include accounts that became customers within the last 12 months. Expand the date range if you have longer sales cycles.
- Include enough accounts for training: Use at least 50 distinct accounts. We recommend 100 or more accounts for better model training.
- Include as much engagement data as possible: Instead of selecting a limited subset of engagement data, exclude only the data you don't want to include. This allows the model to train on a larger dataset.