Kundsegmentering med tvåstegsklustring : En fallstudie inom banksektorn

Customer segmentation, the practice of grouping customers based on shared characteristics, is pivotal for data-driven strategies, enabling businesses to tailor their products and services effectively. However, real-world customer behaviour can be influenced by numerous interrelated factors, which can complicate segmentation. This thesis investigates the potential of a two-stage clustering method combining Self-Organising Maps (SOMs) with k-means clustering to identify meaningful customer segments from high-dimensional behavioural data at a leading digital bank in the Nordics. Specifically, SOMs are explored as a dimensionality reduction pre-processing step aimed at enhancing clustering effectiveness. Using a dataset of 5,000 anonymised customers, encompassing demographic attributes, trading history, and platform usage patterns, the study evaluates the two-stage SOM + k-means method against standalone k-means. Results demonstrated that the combined approach consistently outperformed standalone k-means according to internal validity metrics. The resulting segments were interpretable, highlighting clear patterns in trading activity, mobile usage, and customer tenure. Nonetheless, segment boundaries remained somewhat diffuse, reflecting the inherently continuous and overlapping nature of customer behaviours. This research contributes to machine learning-based customer segmentation by applying and critically evaluating a two-stage clustering method in a real-world financial context. The findings underline the value of SOM-based preprocessing for complex segmentation tasks and support treating segmentation as exploratory rather than definitive. Future research could explore additional high-dimensional datasets or assess the generalisability of this method across various customer groups and industries.