Visa processed 106 million payment disputes in 2025—a 35% jump since 2019. Their new AI tools promise to let merchants spot and fix disputes before they escalate, using generative AI to draft responses and dig into order data. On the surface, this looks like sensible automation. But for UK regulated firms, it exposes a dangerous gap in how the industry thinks about AI. Faster processing of disputes is not the same as better dispute resolution. And automation without clear audit trails and human accountability violates the spirit of FCA Consumer Duty PS22/9, which demands that firms act in customers' best interests—not just in their own operational convenience.

What Visa's announcement really shows is that the industry has embraced a particular model of AI deployment: throw generative AI at high-volume, repetitive processes and assume scale solves quality. This works for some tasks. It fails catastrophically for others. Disputes touch consumer money, trust, and legal obligation. When a merchant uses generative AI to draft a dispute response without human review, or when an algorithm flags which disputes to escalate, the firm is making a judgement call about a customer's rights. That judgement call is not merely a technical decision. It is a regulatory and ethical one. The fact that Visa's tools exist does not mean every regulated firm should adopt them wholesale. Too many mid-market insurers, lenders, and financial services firms are copying the 'move fast' playbook without asking who is accountable when the AI gets it wrong.

Here is what matters: generative AI is a lousy substitute for transparent, explainable decision-making in regulated workflows. Tools like Harvey and Legora have built reputations in legal services partly because they focus on augmentation—helping humans work faster—rather than replacement. Trovix takes the same view across financial services and accountancy. Our approach, embedded in Trovix Audit, is to build AI governance into the workflow from day one. That means logging every AI decision, flagging where human judgment is required, and proving to auditors (and the ICO under UK GDPR) that you made an informed choice about which tasks are safe to automate and which are not. Visa's tools may be technically impressive. But if your firm adopts them without a governance layer, you are betting that scale and speed matter more than accountability. In a regulated environment, that bet will lose.

What should you do? First, do not assume that because a major payment processor has launched AI tools, your firm should follow. Second, audit your own dispute workflows against FCA Consumer Duty principles and PRA SS1/23 if you are a larger institution. Ask: Which steps in dispute resolution require human judgment? Which decisions affect customer outcomes materially? Which could attract regulatory scrutiny if they fail? Use Trovix Watch to stay across the regulatory guidance emerging around AI in financial services—the EU AI Act is already forcing change, and the FCA's AI roadmap is tightening. Third, when you do automate, build in human review gates and full traceability. A merchant using Visa's generative AI to write a dispute response should know exactly what the AI suggested, who approved it, and why. That is not a bottleneck. It is compliance.

Source: CNBC