Overview
The ML/AI Configuration page is where you train and manage the machine-learning models that power personalised predictions across Revenue Intelligence. It shows whether your organisation has enough data, how each model is doing, and lets you retrain when new data lands. This area is intended for Admin or RevOps users.
Getting there
Open the ML/AI Configuration card from the Settings hub, or go to /revenue-intelligence/settings/ml-ai.
ML Readiness Status
The readiness panel tells you whether your data can support good predictions. It reports a status (for example Optimal Data) and three signals:
- Closed Deals — won or lost deals the models learn outcomes from.
- Deals with Stage History — deals with stage progression, used for timing models.
- Months of Data — how much history is available for forecasting.
Each signal is rated (for example Optimal or Ready), so you can see at a glance what is strong and what needs more data.
Machine Learning Models
Three models are listed, each with its status, the data it uses, and its required versus optimal thresholds:
- Win Probability — predicts the likelihood of closing opportunities from historical patterns. Requires closed deals (won or lost) to learn from actual outcomes.
- Revenue Forecast — forecasts revenue trends using time-series analysis. Needs historical revenue data; the page shows the current model complexity (such as Advanced ML) and how many more months would reach optimal accuracy.
- Deal Velocity — estimates time-to-close and identifies bottlenecks. Requires closed deals with stage-progression history.
Each model shows a ratio of available data against its required and optimal levels (for example 75/30 with Required 15 and Optimal 30), and is marked Active once trained.
Training and retraining
- Confirm the readiness panel shows enough data for the model you want.
- Select Retrain on that model to rebuild it on the latest data.
- Use Refresh to update the readiness figures after new deals close.
Troubleshooting
- A model can’t train: check it has reached its Required data threshold — models need a minimum number of closed deals or months of history.
- Predictions feel weak: aim for the Optimal threshold; accuracy improves as you move from Required toward Optimal.
- Figures look stale: select Refresh to recalculate readiness.