The FCA’s latest publication on AI in financial services doesn’t introduce new rules. It does make a practical point: firms remain responsible for the outcomes their AI delivers, and need to show that governance works in day-to-day operations. That’s as much a question of people and business processes as it is of technology.
On 19 October, the FCA published AI in financial services: good practice and areas for improvement. It draws on supervisory work to illustrate where AI is being managed well and where firms need to improve. The examples sit alongside existing obligations, including the Consumer Duty and the Senior Managers and Certification Regime (SM&CR).
The following week, the Bank of England and FCA published their 2026 survey of AI in UK financial services. Among respondents, 86% said they were using AI. Yet while 94% of full-survey respondents reported firm-wide AI policies or standards, only 39% reported routinely monitoring customer-impact measures. Having a governance framework and knowing whether it works are different things.
What the FCA Is Saying
A few themes stand out:
- Responsibility stays with the firm. Outsourcing an AI service doesn’t outsource accountability for its effects on customers.
- Governance starts before launch. Risk assessment, challenge, testing and ownership can’t sensibly be bolted on once a tool is in production.
- Human oversight has to mean something. Reviewers need the information, authority and confidence to challenge an AI output, not simply approve it.
- Monitor outcomes, not only technical performance. The FCA points to live monitoring, audit records, complaints and evidence of whether customers are being treated fairly.
- Autonomy requires clear limits. Firms need to decide where AI may assist, act, abstain or escalate, with controls appropriate to the decision and its consequences.
Why This Is a Business and Workflow Problem
There’s an irony in AI adoption that I keep coming back to. One of the most significant technology advances of our time still depends heavily on some very familiar disciplines: organisational change, training, adoption, process design and effective governance. As we delegate more work to technology, the need for clear human responsibility doesn’t simply disappear.
Financial services has expensive experience of what happens when flawed sales practices become widespread. PPI and motor finance arose from different failures, but both remind us of the consequences when problems become embedded in repeatable processes. AI could reproduce the same poor explanation or recommendation across thousands of interactions. Properly designed, it could just as easily make decisions more consistent and help firms spot problems sooner.
Consider an insurance agent using AI to compare policies. What information did the assistant rely on? Which product eligibility rules applied? What was actually explained to the customer? Who made or approved the decision, and how would the firm recognise a pattern of poor outcomes? Capturing an audit trail helps, but it isn’t enough on its own. The controls must influence what happens, and monitoring must test the result. We’ve set out what that looks like in day-to-day operations in What good looks like for AI in financial services operations.
Where ServiceNow Fits
ServiceNow can provide a useful combination of AI governance capabilities and operational workflows, and the two do different jobs. ServiceNow’s AI Control Tower supports the inventory, ownership and governance of AI systems and agents. That’s different from governing each individual customer transaction. For that, configured workflows can apply business rules, route cases for qualified review, manage exceptions and record approvals, overrides and escalation. With the necessary integrations and monitoring, firms can connect AI activity to the decisions and customer outcomes that matter, rather than treating the model as a standalone tool.
None of this happens automatically: the data, integrations, safeguards, role design and testing all have to be right. A workflow platform can help make controls operational; it can’t replace a firm’s own judgement about risk, suitability or compliance.
How Right Prompt Helps
At Right Prompt, our approach is to weave AI into the business processes we’re building, rather than start with AI and look for an application. In financial services, that means using it selectively for tasks such as document understanding and product information, alongside defined rules, integrations and accountable decisions.
Our Smart Assurance insurance distribution solution illustrates the approach. It brings together product eligibility criteria, quotations from insurer systems, application workflows and customer and seller consent steps. We’ve also demonstrated AI assistance to help an insurance agent understand and compare policies.
That’s the opportunity I take from the FCA’s publication. Firms should not have to choose between making AI useful and keeping it under control. But they do need to design both into the way work gets done, and be able to demonstrate the outcomes.
This article is general commentary and does not constitute regulatory or legal advice.