Reg E timelines push banks to pay disputes before they can investigate them

The useful part of this is the mechanism, which explains why an expensive detection stack can be working as designed and still see nothing. In a first-party dispute the credentials are real and the customer is who they claim to be, so there is no anomaly to fire at authorization; the tooling was built for stolen credentials, account takeover and synthetic identity ^[13]. The loss shows up later as a case, and the survey's respondents name separating legitimate claims from fraudulent ones as their top operational problem, made worse by manual processes and unintegrated systems at volume ^[14]. That relocates the loss into the part of the bank measured on cycle time rather than on loss avoided.

Compounding it is a timing problem the institution does not fully control. Shanmugam's argument is that provisional credit goes out before a meaningful investigation is finished because Reg E deadlines require it, not because anyone is careless ^[6], and once the credit is out it rarely comes back ^[15]. So the decision that determines the loss is made before the evidence that would inform it exists. Any detection improvement has to fit inside that window or it is only good for post-mortems.

The rate of change deserves a discount. Going from 15% to 36% of reported fraud in a single year ^[3] is 2.4 times the prior figure, a 21 point move ^[16]. Jumps that size in a category defined by classification can reflect banks relabelling losses they previously booked as third-party fraud, and the article does not separate new filings from new labelling. We do not know yet which share is which, though either reading leaves the same caseload sitting in the same understaffed queue.

The diagnosis is published by the co-founder and CEO of Casap, which sells agentic dispute management to banks and credit unions ^[1], and Casap commissioned the survey with Cornerstone Advisors ^[4]. Even weighing that against the findings, the one that survives is the least flattering: nearly half of respondents do not know or do not track what share of false claims they catch before provisional credit goes out ^[8]. That is a statement about a bank's own management reporting, and any institution can check its own number this quarter without buying anything.

The board-deck version is that losses rose and the tools are dated, so the answer is a purchase order. That version leaves out two things the same survey shows. When 91% of bankers say low perceived consequences embolden false filers ^[9], they are describing a policy variable about what happens to a repeat filer, not a model accuracy problem. And when nearly 90% attribute suspicious disputes to financial hardship ^[10], they are describing customers the bank intends to keep, most of whose disputes are legitimate ^[17] and whose treatment carries regulatory and franchise cost.

So the sequencing runs the other way from the procurement instinct. Instrumenting the pre-credit catch rate is what makes or breaks the tooling case next quarter, because a bank that cannot state its current number cannot price the improvement or defend the consequence policy that follows from it.