Predictive Lead Scoring

Predictive lead scoring is the umbrella term for any model — logistic regression, gradient-boosted trees, modern ML — that estimates conversion probability from historical patterns. It is closely related to AI lead scoring; the distinction is mostly tradition. "Predictive" came out of the marketing analytics world in the 2010s and usually meant simpler statistical models. "AI lead scoring" is the same idea with deeper models and richer features. In practice, vendors use the terms interchangeably.

The mechanic is the same regardless of model class: train on past conversions, validate on a held-out set, and score each new lead with a probability. The score then drives handoff (top scores to senior reps), cadence (high-intent leads get faster outreach), and forecasting (sum of probabilities approximates expected revenue).

The discipline that separates good implementations from bad is honest validation. A model that scores 0.95 AUC on training data and 0.55 on real reps is overfitted and will burn rep trust within a quarter. Check calibration, monitor drift, and retrain on a regular cadence — quarterly is a reasonable default for B2B.