The Real Barrier to AI Adoption in Financial Services Isn’t Regulation

On July 14th, HM Treasury published its Financial Services AI Adoption Plan: ten recommendations, five themes, and a lot of careful language about coordination and prioritisation. Most of it will get skimmed and filed. But buried in the first theme is a sentence that should make every risk-averse board in the sector slightly uncomfortable: “the core challenge now is not the absence of regulatory support, but its accessibility, consistency and practical application.”

Read that again. The government’s own commissioned review just said, in writing, that the rules are not the reason your AI programme is stuck.

What the Financial Services AI Adoption Plan Actually Says

For eighteen months, “we’re waiting for regulatory clarity” has been the single most useful sentence in UK financial services. It stalled procurement decisions, justified pilot purgatory, and gave risk committees a reason to say not yet without saying no. It was a legitimate position when the FCA’s stance on generative AI genuinely was undefined.

It is no longer legitimate, and the plan says so directly. The UK has a principles-based, outcomes-focused regulatory framework. Regulators have already built sandboxes, AI Live Testing and innovation pathways through the FCA’s AI Lab. What the plan flags as the actual gap is that most firms still cannot navigate any of it, because guidance is fragmented and few know how to apply high-level principles to a specific AI use case.

That distinction matters more than it looks. If the barrier is regulatory ambiguity, the fix is patience: wait for the next consultation, the next guidance note. If the barrier is something else entirely, patience stops being a strategy.

The Real Barrier to AI Adoption in Financial Services Isn’t Regulatory Clarity

The plan calls the gap “accessibility, consistency and practical application,” but there’s a more precise word for it: translation. Consumer Duty, model risk management, the Senior Managers and Certification Regime, explainability, accountability – none of these are new concepts invented for AI. They are existing obligations that most firms have never had to apply at the level of a specific workflow, a specific agent, a specific handoff between a model and a human being.

The plan’s own recommendation isn’t “wait for more rules.” It’s regulators making expectations clearer, running alongside firms doing the harder internal work of applying principles they already have. That second half is the part most organisations skip.

This is why the excuse mattered so much, and why losing it changes the calculus. If the barrier was regulatory ambiguity, slow-moving firms could tell themselves the constraint was external. HMT’s plan removes that alibi. The infrastructure to move exists. What’s missing sits inside the organisation and not with the regulator.

Why AI Governance in Financial Services Is a Design Problem, Not a Compliance Problem

The instinct in regulated industries is to treat “are we allowed to use AI here” as the hard question and hand it to legal and compliance. That’s the wrong question, and it’s the wrong owner.

The real question is: in this specific workflow, where does the model assist, where does it recommend, where does a person retain final accountability, and can the organisation evidence that decision when someone asks. That’s not a compliance sign-off. It’s a design exercise — one that has to happen workflow by workflow, not once at a policy level.

Firms that map this per workflow, rather than waiting for a single blanket ruling from above, are the ones the plan’s own language points to as ready to scale. That mapping work  (not another vendor evaluation or another AI strategy deck) is where the next six months of effort should go.

The pattern shows up the same way in almost every risk committee we sit in on. Everyone in the room can recite Consumer Duty and model risk requirements from memory. What actually stops the discussion is a much smaller question: if the model surfaces a recommendation and an analyst clicks approve without reading it properly, who owns that outcome under the Senior Managers and Certification Regime? Nobody has mapped it. The rule was never the gap. The workflow was.

Firms keep treating “what does the regulator allow” as the hard question and “what does our own accountability chain actually look like at the point of use” as an afterthought. The Treasury plan just confirmed, in a government-commissioned document, that the second question is the one worth solving first.

Closing the Gap Between Principle and Practice

This is precisely where Experience Haus works. We don’t come in with a generic AI strategy or a technology recommendation. We map specific workflows against existing regulatory obligations, identify exactly where a model should assist versus recommend versus stay out entirely, and build the accountability trail that lets a risk committee sign off with confidence rather than caution. It’s the difference between an AI policy that sits in a folder and an AI deployment a risk lead will actually defend.

The Question Worth Asking This Week

Pull up recommendation one of HM Treasury’s plan and ask your risk lead whether they could point, right now, to how Consumer Duty and model risk management apply to your last AI deployment at the workflow level. If the honest answer is no, that was never a regulatory clarity problem. It’s a design problem, and it’s solvable faster than most boards assume.

If you’re ready to move from waiting on regulatory clarity to designing for it, talk to us about how we map AI accountability into your workflows.