Authors:

Shreyaa Karan

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Financial institutions often struggle to personalize offerings to their diverse client base. They should move beyond demographic segmentation and towards using behavioral personas based on decision styles, since traditional methods fall short of capturing evolving customer behaviors. By understanding client decision-making goals and autonomy levels, banks can offer AI-personalized financial journeys, increasing engagement and upsell. This project explores how agentic behavioral traits, characteristics reflecting intentional, and goal-directed financial activity, can provide a more dynamic and insightful basis for client segmentation. Using over one million real-world bank transactions, we engineer agentic features such as transaction consistency, frequency, time-of-day usage, etc.. These features were used to cluster clients through unsupervised learning in JMP Pro, comparing k-means and Self-Organizing Map (SOM) algorithms. The resulting clusters revealed distinct financial personas (consistent planners, impulsive transactors, etc.) that aligned with varying levels of agentic behavior. Model evaluation included cross-validation, silhouette analysis, and interpretability of clusters. Business implications included tailoring digital journeys, recommending budgeting advisors based on behavioral archetypes, and enhancing loyalty through personalized cross-sell strategies. K-means clustering outperformed SOMs in silhouette score (0.67 vs. 0.49), cluster separation, and real-world interpretability. It produced five actionable personas validated through multivariate analysis. This approach is valuable for banking analysts seeking to personalize user experiences based on intent rather than identity. The clarity of the segments supported stakeholder decision-making and was benchmarked to industry results, estimating a 12-18% increase in product cross-sell success. Future enhancements include applying semi-supervised † Presenting Undergrad Author; ‡ Contributing Undergrad Author; * Undergrad Acknowledgment learning or time-evolving cluster analysis to anticipate customer lifecycle transitions. Keywords: Predictive Modeling; Unsupervised Learning; Behavioral Segmentation; Agentic Features; Cluster Analysis

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Shreyaa Karan