Temporal Validation of Case-Mix-Aware Credit-Card Complaint

Response Analytics

Irina V. Pustokhina1,* Denis A. Pustokhin2

1 Department of Entrepreneurship and Logistics, Plekhanov Russian University of Economics, Moscow 117997, Russia

2 Department of Logistics, State University of Management, Moscow 109542, Russia

Received: July 20, 2026 Revised: September 05, 2026 Accepted: October 03, 2026 ⋆ Corresponding author

ABSTRACT

Complaint dashboards often pool unlike cases and report an overall resolution rate. For a financial-services operations

team, that average can conceal which issues account for redress workload and whether a triage signal remains useful

on later cases. We analyze 75,988 publicly reported U.S. credit-card complaints received in 2024. The outcome is the

company’s recorded “closed with monetary relief” category, not a verified payment amount or customer satisfaction.

Intake issue, sub-issue, submission channel, and company identify a transparent logistic screening model trained on

January–September records and evaluated on October–December records. Report-related issues comprise 31.9%

of complaints but have a 0.63% monetary-relief rate, compared with 19.16% for other issues. An exact shift-share

decomposition attributes 1.81 of the 2.64 percentage points of decline from the first to fourth quarter to changing

issue mix. Issue alone achieves an out-of-time area under the receiver operating characteristic curve (AUC) of 0.819;

adding sub-issue and channel raises it to 0.831. A cross-company specification reaches 0.873 (95% bootstrap interval

0.867–0.878) and captures 39.4% of recorded monetary-relief responses in its highest-scoring tenth of holdout cases.

The company term captures institutional and routing differences, so that result does not measure issuer quality or

the gain attainable within a single firm. Case-mix-aware reporting and temporal validation support staffing review

queues without automating redress decisions.

Keywords: Business intelligence Complaint analytics Financial services Case mix Temporal validation

1. INTRODUCTION

A complaints manager can see a rising case count and still

miss the more important change: the share of fee disputes may

have fallen while cases involving credit-report information

increased. Those two issues can lead to different recorded

company responses. Staffing, specialist routing, and an explanation

of quarter-to-quarter performance therefore require

more than a portfolio-wide response rate. The relevant business

intelligence task is to distinguish case composition from

operational outcomes, then ask whether an analytic signal

survives on complaints received later.

We study the public Consumer Financial Protection Bureau

(CFPB) credit-card complaint portfolio for calendar 2024.

The database exposes intake descriptors and company response

categories, which makes it possible to test a specific

operational question: how much of subsequent monetaryrelief

response can be anticipated from information available

before closure? The study contributes a case-mix audit, a sequence

of feature-ablation models, and a strictly later-quarter

holdout. Its unit is a published complaint, not a credit-card

customer or a dollar of compensation. We avoid translating

a response category into welfare, fairness, or an optimal

remedy.