Remedy-Aware Financial Innovation: A Hierarchical RegTech

Framework for Consumer Complaint Resolution

Laith Farhan1,* Raad S. Alhumaima2

1 School of Engineering, Manchester Metropolitan University, Manchester, M1, UK

2 Brunel University, Uxbridge UB8 3PH, UK

Received: March 05, 2025 Revised: May 06, 2025 ⋆ A Cc ocrepretespdo: ndAinugg uasutt ho0r4, 2025

ABSTRACT

Financial complaint systems keep track of instances of service failure as well as of the remedy chosen by the

responding institution. The crucial question for operational supervision isn’t just one of categorization of complaints,

however; compensation is scarce, expensive, and fits into a larger framework of explanation vs. relief. In this

paper, an approach for the development of a regulatory technology framework for remedies is proposed, which

conceptualises the resolution of complaints as a hierarchy. The first stage determines if a case will be decided on

relief or explanation, and the second stage distinguishes monetary from non-monetary relief. This is done by using

sparse and regularized models, which are calibrated on the following validation period, and tested on an untouched

chronological holdout, combining structured intake attributes with complaint narratives. In the empirical application,

268,570 complaints were received by California in 2024. Monetary relief makes up 1.04% of independent test period,

and accuracy is not the best criterion. The hierarchical model yields a monetary-relief precision-recall area of 0.335,

whereas the flat text-tabular model and a flat model with the event prevalence yield 0.314 and 0.010, respectively. At

the 2 percent review budget, it is able to identify 58.9 percent of monetary-relief cases and has a lift of 29.4 times

over a random review. The flat fusion model is slightly better for overall three-class classification, demonstrating the

benefit of a hierarchy of remedies when institutional capacity is focused on rare, consequential outcomes, rather

than labelling. Results provide a practical design of human-supervised complaint triage while retaining calibration,

interpretability, chronological validation, and clear usage limitations for automated complaint analysis.

Keywords: Financial innovation Regulatory technology Consumer complaints Hierarchical classification Monetary

relief Text analytics Responsible artificial intelligence