Follow the Money: A Composite Behavioral Risk Scoring Approach to Detect Cognitive Decline in Older Bank Customers

Abstract

Cognitive decline in older adults exposes individuals and their families to serious and lasting financial harm. Despite its increasing prevalence, cognitive decline is systematically underrecognized, and by the time a diagnosis is made, significant financial damage is often already done. Banking institutions hold a unique touchpoint to this at-risk population through the granular financial and behavioral data they continuously collect on their customers. This paper investigates whether transaction-level banking data contains signals consistent with early cognitive decline and whether those signals can be used to construct individual-level risk profiles. Using de-identified data from 856 randomly selected customers aged 50 and older at a top-10 U.S. commercial bank over a three-year window, the study analyzes variables spanning transactions, digital engagement, authentication logs, and power-of-attorney (PoA) events. This study employs age gradient validation via Spearman correlations, Mann-Kendall monotonicity tests, and Kruskal-Wallis tests and then builds a composite cognitive risk score proxy from six variable components (authentication failures, password resets, IVR failures, late payments, financial hardship, login disruption). Next, regressions of risk score on age with different baseline controls are executed and analyzed, and PoA-activated customers are used as an out-of-sample face validity check. The findings of this study are cautiously positive. Six variables show perfect monotonic decline across age groups consistent with cognitive deterioration; age is a robust predictor of elevated risk scores, with effects accelerating after age 65; and 62.5% of PoA-activated customers score above the population median on the composite risk score. This exploratory study contributes to research on identifying cognitive decline through behavioral markers and has implications for future intervention-focused work aimed at shortening diagnosis timelines for people living with dementia and mild cognitive impairment (MCI).