A subscription business might train a model on past customers to give each current customer a churn score: the estimated chance they cancel in the next 90 days. The retention team then calls the highest-risk customers first. That is predictive analytics. The output is a probability or forecast, not a certainty, and it is only as good as the history it learned from.
IBM defines predictive analytics as a branch of advanced analytics that makes predictions about future outcomes by combining historical data with statistical modeling, data mining and machine learning. Common techniques include regression analysis and supervised learning classifiers.
Why it matters: predictive analytics is how most businesses used machine learning before generative AI, and it still drives forecasting, fraud detection and lead scoring. Shmueli’s paper “To Explain or to Predict?” makes a useful distinction: a model that predicts well is not necessarily a model that explains why something happens, so a high churn score does not by itself tell you what to change. No Chain of Thought episode has discussed predictive analytics as a named practice.
Sources
- IBM: What is predictive analytics? — Vendor explainer from IBM; supports the definition combining historical data, statistical modeling, data mining and machine learning, checked October 24, 2026
- Shmueli, To Explain or to Predict? (Statistical Science, 2010) — Peer-reviewed paper; supports the distinction between explanatory and predictive modeling, checked October 24, 2026