Implementing and Training Predictive Customer Lifetime Value Models in Python
summary
Implementation walks through a Pareto/NBD hierarchical Bayesian model implemented in Jupyter notebooks, using PyMC (and noting Stan alternatives) and the lifetimes utilities on the CD Now dataset. It trains on recency/frequency/duration RFM inputs, produces posterior Lambda and mu per customer, and computes probability a customer is alive and expected future purchases. Aggregate validation predicts 1724 versus observed 1788 repeat purchases (RMS C ≈ 1.4), with code and a GitHub notebook provided.
AI-written summary · about
chapters (11)
- 0:00CLV modeling in Python
- 1:01Why CLV matters
- 4:55Contractual vs non-contractual
- 8:26Pareto/NBD model intro
- 14:13Data needs for CLV
- 17:07Training CLV models
- 18:38Pareto/NBD in PyMC
- 22:54Actionable CLV insights
- 25:23Model evaluation
- 30:27Summary
- 30:47Q&A
related talks
34mModelling and predicting customer lifetime value
36mCustomer Lifetime Value: Models, Metrics and a Multitude of Uses
44mEating Humble Py: From Toy Problem to Real-World Solution in Predicting CLV
41mPyMC-Marketing: Customer and Marketing Analytics the Easy Way
35mUsing Survival Analysis to understand customer retention
35mWhat we learned by running a large custom Bayesian...
01h 26mProbabilistic Programming and Bayesian Computing with PyMC
45mKnow Your Customer