Statistical modeling and probabilistic composition in the prediction of the customer lifetime value

Abstract

Purpose

Data mining registers of transactions allows for benchmarking customer's evaluation strategies. The purpose of this paper is to provide information on the application of different approaches to explore this kind of data.

Design/methodology/approach

Traditionally, heuristics based on variables such as recency, frequency, and monetary (RFM) value of transactions are used to determine the best customers. In this paper, a new form of directly combining the values of these variables is compared to an approach based on fitting a stochastic model. This last model is a mixture of a model for the number of transactions and another for the value spent. The new direct form of evaluation is based on computing the joint probability of maximizing quality indicators.

Findings

Good fit of the different models tested to the series of individual data as well as coherent predictions are registered. Patterns found provide empirical confirmation of results that theoretically should be expected.

Research limitations/implications

These results are valid for a particular supermarkets network in a Brazilian city. The inner consistency of the results, nevertheless, and the coherence of the results obtained with what was expected, encourage application to other places and sectors of activity.

Practical implications

The results obtained show clearly the effectiveness of the approach based on RFM value measurement.

Originality/value

The models studied are applied for the first time for the kind of data treated, where determination of which customers remain active is a problem of special interest.

Citation

Parracho Sant'Anna, A. and Otavio de Araujo Ribeiro, R. (2009), "Statistical modeling and probabilistic composition in the prediction of the customer lifetime value", Benchmarking: An International Journal, Vol. 16 No. 3, pp. 335-350. https://doi.org/10.1108/14635770910961362