Visa processed 106 million disputes in 2025 — a 35% surge since 2019 — and is now deploying generative AI to handle them at scale. For UK financial services firms and their legal teams, this matters because they sit between Visa's automation and the end customer. The FCA's Consumer Duty (PS22/9) makes clear that firms remain liable for algorithmic decisions, regardless of whether they are made by a third party. Visa's new tools will flag disputes, prioritise them, and recommend resolutions using AI. But they will not explain *why* those recommendations were made in a way that satisfies regulatory scrutiny. That explainability gap is now a compliance risk for every UK bank and fintech using them.

This story is part of a larger pattern: payment infrastructure and data providers are racing to automate dispute management without building in the governance that regulators now expect. Visa is not alone — American Express, Mastercard, and UK providers are all moving in this direction. The assumption is that faster = better. But the FCA, ICO, and PRA increasingly treat automated decision-making as a governance event, not just a technology upgrade. The EU AI Act has created a model (albeit outside the UK) where high-risk financial processes require documented risk assessments, bias testing, and audit trails. Firms are beginning to expect the same from their suppliers. Visa's move reveals the gap: the payment rails are automating faster than the compliance frameworks around them.

Here is Trovix's honest view: generative AI excels at pattern matching and speed, which is what Visa is using it for. But pattern matching in dispute resolution creates two problems. First, AI systems like those Visa is deploying do not easily explain their reasoning in regulatory language — they optimise for throughput, not accountability. Second, firms using these tools are outsourcing the decision to a third party without contractual or technical mechanisms to audit it. Compare this to how specialist legal AI systems (like Harvey or Luminance) work in M&A or contract review: they are designed from the outset to show their working, flag uncertainty, and allow human override. Dispute management AI is not built that way yet. The result is that UK firms adopting Visa's tools are taking on compliance risk without control. If a dispute is mishandled and a customer complains to the FCA, the firm's defence cannot be 'the AI decided it' — the firm must be able to demonstrate fair process. Trovix Watch gives firms real-time visibility into these kinds of regulatory shifts, which matter far more than the speed of the underlying process.

What should a mid-market financial services firm, law firm, or insurance broker do right now? First, do not assume that using a major payment provider's AI tools absolves you of FCA Consumer Duty liability — it does not. Second, before deploying Visa's dispute tools, audit the explainability. Can you retrieve and explain the reasoning behind a disputed decision? Can you retrace which data inputs triggered it? If the answer is no, you need contractual language with Visa that allocates liability for algorithmic harm. Third, document your use of these tools as part of your AI governance framework under ICO UK GDPR and FRC ISA UK principles. Fourth, treat dispute AI as a governance question, not just an operational one — involve compliance and legal teams in the deployment decision, not just operations.

Source: CNBC