Visa announced in April that it processed 106 million charge disputes globally in 2025, up 35% since 2019, and has now launched six AI tools to automate dispute detection, validation and resolution across merchants, issuers and acquirers. For UK regulated financial services firms, this should matter because dispute volumes are genuinely growing and manually handling them costs real money and creates compliance friction under the FCA Consumer Duty PS22/9. What Visa is doing makes sense at their scale. But here is the problem: Visa sells tools to firms that already have API integration capacity and dedicated payments infrastructure. Most mid-market UK law firms, insurers and accountancy practices using payment processing or managing payment disputes do not have that infrastructure. They have spreadsheets, three different software systems that do not talk to each other, and a compliance team that is already stretched.

This story is symptomatic of how enterprise AI vendors are building solutions for enterprise problems and assuming the integration layer exists. It does not, especially in UK regulated firms with £20m–£500m turnover. The broader pattern is clear: every vendor from Harvey to Luminance to Copilot is shipping tools that work brilliantly in isolation and assume you have a data engineering team to plug them in. The dispute process is a perfect example. Visa's tools are genuinely clever at flagging false disputes or prioritizing high-value claims. But if your disputes live in three different systems—your payment processor's portal, your CRM, your accounting software—then AI cannot see the full picture. You end up with an expensive tool that solves 40% of your problem and adds another system to your integration backlog.

Trovix's view is blunt: AI that does not solve your integration problem is AI that will fail compliance review and waste budget. Under PRA SS1/23 and the FCA's expectations on third-party operational resilience, you cannot implement a new system without mapping data flows, audit trails and responsibility. When you buy Visa's tools, or any point solution, you are buying an island. We see this repeatedly. Firms buy AI for document review or intake and then realise they cannot feed it the data without rebuilding their document infrastructure first. Visa's approach—building purpose-built, cloud-hosted dispute tools—works for Visa. It does not work for a 150-person insurance broker or a mid-market practice where data lives across five systems and your IT team is one person. The honest difference: our approach, starting with Trovix Sift for data extraction and Trovix Brief for intake, assumes your systems are fragmented and builds from there. We solve the integration layer first, not after.

If you process charge disputes, do this: stop evaluating AI dispute tools in isolation. Map where your dispute data actually lives today—payment processor, case management, finance system, email. Then ask any vendor: how does your AI access all three without me building custom connectors? If they say 'our API integrates with the main platforms', ask which platforms and how long integration takes. Visa's tools will work if you are an acquirer or processor. If you are a regulated financial services firm with messy data and stretched IT, you need a different approach. Start with the integration problem, not the AI problem. That is how you avoid spending £50k on tools that sit unused in six months.

Source: CNBC