Predicting Customer Lifetime Values -- ecommerce use case

TL;DR

This paper compares a statistical 'buy-till-you-die' model and a neural network for predicting customer lifetime value in ecommerce, analyzing their effectiveness and providing recommendations for future research.

Contribution

It offers a comparative analysis of traditional statistical and neural network approaches for customer lifetime value prediction in ecommerce.

Findings

Neural networks outperform statistical models in prediction accuracy.

Qualitative analysis highlights strengths and weaknesses of each approach.

Recommendations for choosing models based on specific ecommerce scenarios.

Abstract

Predicting customer future purchases and lifetime value is a key metrics for managing marketing campaigns and optimizing marketing spend. This task is specifically challenging when the relationships between the customer and the firm are of a noncontractual nature and therefore the future purchases need to be predicted based mostly on historical purchases. This work compares two approaches to predict customer future purchases, first using a 'buy-till-you-die' statistical model to predict customer behavior and later using a neural network on the same dataset and comparing the results. This comparison will lead to both quantitative and qualitative analysis of those two methods as well as recommendation on how to proceed in different cases and opportunities for future research.