Buyer behavior · Hierarchical Bayesian empirical modeling of purchase timing, purchase amount, and defection risk, with model comparison on a separate holdout/validation dataset.
Customer Lifetime Value Measurement
Management Science · 2008
Summary
Jointly modeling purchase timing, amount, and defection predicts customer lifetime value and valuable-customer targeting better than extended NBD–Pareto, RFM, nested, and heuristic benchmarks. Longer interpurchase times predict larger purchase amounts and greater leaving risk; female and male customers show similar timing/risk patterns, while female customers spend less.
For your marketing
Possible founder application
Use this paper as evidence about customer lifetime value measurement in the stated context; compare its population, treatment and outcome with your own decision.
Reported result
What the study found
Jointly modeling purchase timing, amount, and defection predicts customer lifetime value and valuable-customer targeting better than extended NBD–Pareto, RFM, nested, and heuristic benchmarks. Longer interpurchase times predict larger purchase amounts and greater leaving risk; female and male customers show similar timing/risk patterns, while female customers spend less.
Research question
What the authors investigated
What does customer lifetime value measurement show about customer behavior or marketing outcomes?
Design
Hierarchical Bayesian empirical modeling of purchase timing, purchase amount, and defection risk, with model comparison on a separate holdout/validation dataset.
Sample and setting
Customer-level longitudinal records from a membership-based direct-marketing company where joining and termination dates are observed; estimation and holdout samples are used.
Scope and transfer
Limitations to keep in view
The data are from one membership-based direct-marketing company and the model is tailored to a setting with observed membership termination, limiting transfer to noncontractual or multi-product SaaS settings; the authors also note that longer lifetime does not necessarily equal higher value, underscoring dependence on the chosen behavioral model and observables. Full-text methods and correction history were not independently checked.
Research sources
Candidate study metadata and reported evidence were cross-checked against a direct publisher-page retrieval; the full article was not independently reviewed.