PE-301a · Module 1

What Propensity Models Predict

3 min read

A propensity model answers one question: given everything we know about this deal right now, what is the probability it will close? Not the probability the rep assigned based on their gut — the probability calculated from patterns in your historical data. If deals with your top 3 features (decision-maker engaged, budget confirmed, and 2+ meetings in 14 days) have historically closed at 72%, then a new deal with those same features gets a 72% propensity score. The model learns from your data what predicts your wins.

Do This

  • Build propensity models on your own closed-won and closed-lost data — your patterns are unique to your business
  • Use propensity scores alongside rep judgment, not as a replacement — the model sees patterns, the rep sees context
  • Retrain the model quarterly as new closed deal data accumulates — the patterns evolve as your market and process change

Avoid This

  • Use industry benchmarks as propensity scores — your conversion patterns are specific to your sales motion
  • Override rep-assigned probabilities with model scores without explanation — adoption requires trust, and trust requires transparency
  • Train a model on fewer than 200 closed deals — the sample is too small to learn reliable patterns

The minimum data requirement for a propensity model is 200 closed deals — roughly 100 won and 100 lost. With fewer, the model cannot learn reliable patterns. With 500+, the patterns become statistically robust. If you do not have enough historical data yet, start collecting the features now and build the model when you reach the threshold.