Visa announced in April 2026 that it had deployed six AI tools to automate parts of the charge dispute process, handling over 106 million disputes globally in 2025 alone—a 35% surge since 2019. For UK acquirers, issuers and merchant banks, this looks like good news: less manual work, faster resolution, fewer administrative headaches. The reality is messier. These tools reduce process friction, not commercial or regulatory risk. A merchant disputes a transaction faster. But if that AI decision is wrong, or if it cannot be explained to an FCA investigator or a court, your firm carries the liability, not Visa. Under FCA Consumer Duty PS22/9, you must treat customers fairly and demonstrate it. Opaque automation fails that test.

This is part of a wider pattern in financial services AI: vendors build tools that optimise for speed and cost reduction, then leave their clients to solve the governance problem. We see it in underwriting (where tools like Luminance and Harvey offer classification and prediction), in KYC screening, and now in dispute triage. The assumption is that faster = better. But regulated firms operate under different constraints. The FCA, PRA, and ICO care less about throughput than about evidence. Can you show your work? Can you audit what happened? Can you explain why customer X got decision Y? Most vendor tools give you the first two, barely. The third—explanation and defensibility—is where they tend to get thin.

Here is where Trovix's approach differs, and why it matters here. When we built tools like Trovix Sift, we started from the regulatory requirement backward, not the efficiency benefit forward. Sift extracts and classifies data from documents, yes. But it also creates a chain of custody: you can see which fields were extracted, which rule applied, which confidence score was assigned, and why the system made that call. That auditability is not a nice-to-have in dispute resolution. It is the whole game. A merchant questions a chargeback decision. Your compliance officer needs to hand an FCA investigator or a claimant's solicitor a clear record: what data was evaluated, by which logic, with what confidence, and by human oversight at which stage. Visa's tools—and most third-party AI platforms—do not force that discipline. They optimise for merchant throughput, not for your regulatory defence.

What should a mid-market UK acquirer, issuer or merchant bank do right now? First: do not assume Visa's AI tools are sufficient governance. They are convenient, and they will reduce manual effort. But they do not replace a documented dispute resolution policy or human review of high-value or flagged disputes. Second: if you are building or buying dispute automation internally, demand explainability and audit trail as core requirements, not post-hoc add-ons. Third: map your dispute rules and thresholds now—before you plug in any AI—and version-control them like code. Your FCA file should contain a clear, dated record of your dispute policy, the AI tools you use, the human oversight points you maintain, and the testing and monitoring you do. If you have not done that yet, do it before the next chargeback season. Visa's tools will move disputes through your system faster. Your governance framework needs to move faster too.

Source: CNBC