What Is Propensity Modelling?
Propensity modelling predicts the likelihood that a customer will take a specific action — purchase, churn, or convert. Learn how it drives targeted business decisions.
Key Takeaways
- Propensity modelling assigns each customer a probability score for a specific action — purchase, churn, upgrade, or response to an offer.
- It enables resource allocation based on likelihood rather than intuition, targeting customers most likely to respond.
- Models are built using historical data where the outcome is known, then applied to current customers to predict future behaviour.
What propensity modelling does
Propensity modelling calculates the probability that a specific customer will take a specific action within a defined time period. A propensity-to-buy model might score each customer from 0 to 1, where 0.8 means an 80% chance of purchasing in the next 30 days. A propensity-to-churn model estimates how likely each customer is to stop buying. These scores let businesses focus resources on customers where intervention will have the greatest impact rather than treating everyone equally.
How models are built
Propensity models learn from historical data. To build a churn propensity model, you analyse customers who churned in the past and identify the behavioural patterns that preceded their departure — declining purchase frequency, fewer site visits, reduced email engagement. Machine learning algorithms like logistic regression, random forests, or gradient boosting learn these patterns and apply them to current customers. The model outputs a probability score for each customer based on their current behaviour matching historical churn signals.
Business applications
Propensity-to-purchase models identify the hottest leads for sales teams. Propensity-to-churn models flag at-risk customers for retention campaigns. Propensity-to-respond models predict which customers will react to a specific offer, improving campaign ROI. For African fintech companies like Paystack merchants, propensity models can predict which customers are likely to try new payment methods or upgrade their service tier, enabling targeted outreach that maximises conversion.
Implementation considerations
Start with a clear definition of the action you want to predict and the time window. Ensure you have sufficient historical data — at least several hundred examples of both positive and negative outcomes. Choose simple models first (logistic regression) before moving to complex ones. Validate model accuracy on held-out data that the model has never seen. Monitor performance over time because customer behaviour patterns shift. Retrain models quarterly at minimum to maintain prediction accuracy.