Adaptive RegTech for Financial Complaint Operations:

Temporal Institutional Risk Signals Outperform Text–Tabular

Fusion in Predicting Untimely Responses

Samandarboy Sulaymanov1,* Olimjonov Abbosjon Olimjonovich2

1 Tashkent State University of Economics, Tashkent, Uzbekistan

2 Tashkent State University of Economics, Uzbekistan

Received: December 11, 2025 Revised: February 04, 2026 Accepted: May 01, 2026 ⋆ Corresponding author

ABSTRACT

While the innovation in the financial sector can be measured by the products it offers customers, many valuable

innovations are generated by operational technologies that facilitate the regulatory system to respond more quickly.

This research designs and pilots an adaptive regulatory-technology approach to prioritize consumer complaints

at high risk of an untimely institutional response. A leakage-safe chronological design is used to compare four

approaches: static historical institutional-risk benchmark, short-horizon rolling-risk score, interpretable structured

classifier, and text–tabular fusion classifier. Selection and calibration of models come before a non-biased one-month

holdout period for evaluation. Untimely responses are very uncommon, and performance is evaluated based on

precision–recall discrimination, calibration, and lift and recall under fixed review-capacity constraints, not just on

accuracy. The seven-day rolling institutional-risk score proved to be the best performing operational group at a 2%

review budget, identifying 75.7% of cases that arrive late with 9.9% accuracy, a 37.8-fold lift over random review.

When the review budget was increased to 5%, 92.8% of cases that were untimely were captured. Unlike a common

belief in financial NLP, incorporating any complaint-language indicator failed to improve financial ranking results:

The fused model performed worse than any of the institutional-risk dynamic or static benchmarks. When service

failures are concentrated in institutions, the findings indicate that the operational state near to the time of the failure

could convey more information than richer complaint content. The suggested framework provides a clear low cost

human-in-the-loop RegTech solution that allows to channel limited compliance focus and maintain auditability and

chronological validity.

Keywords: Financial innovation RegTech Consumer complaints Explainable machine learning Rare-event

prediction Operational risk Human-in-the-loop compliance

1. INTRODUCTION

Digital lending, alternative credit scoring, mobile payments,

and platform-based intermediation are often considered the

hallmarks of financial innovation. These developments have

impacted the economics of screening, distribution and service

delivery and have introduced new operational and conduct

risk challenges [1, 2, 3]. Another, more hidden but increasingly

significant field, is regulatory technology (RegTech):

the use of data, analytics and automated controls to enhance

the execution, monitoring and supervision responsiveness of

compliance with regulations. In this area, the problem is not