Predicting Customer Goals in Financial Institution Services: A Data-Driven LSTM Approach

TL;DR

This work tackles predicting customer goals and actions in financial services using two LSTM-based models: a standard LSTM and a GNN-enhanced LSTM that leverages a state-space graph of customer states. Built on a semi-synthetic dataset generated by a domain-independent simulator, the approach constructs handcrafted features and a graph representation to capture both temporal and structural patterns in multi-interface banking interactions. Results show that graph embeddings improve goal, type, and trajectory predictions, though goal accuracy remains below 80%. The study provides a scalable pipeline for behavior prediction in banking contexts and discusses future work on using graph-informed insights to influence customer behavior through targeted incentives. The practical impact lies in enabling personalized experiences and more effective interface-migration strategies in financial institutions, supported by a data-driven, graph-aware modeling framework.

Abstract

In today's competitive financial landscape, understanding and anticipating customer goals is crucial for institutions to deliver a personalized and optimized user experience. This has given rise to the problem of accurately predicting customer goals and actions. Focusing on that problem, we use historical customer traces generated by a realistic simulator and present two simple models for predicting customer goals and future actions -- an LSTM model and an LSTM model enhanced with state-space graph embeddings. Our results demonstrate the effectiveness of these models when it comes to predicting customer goals and actions.

Figures (2)

• Figure 1: An example of bag-of-words and one-hot encoding representation, and a state-space graph representation.
• Figure 2: Loss function of LSTM and GNN+LSTM approaches when predicting the next 1, 5, and 15 actions of the customer when viewing the customer's last 20 actions.