AI Adoption in Regulated Organisations: Why Most Projects Fail First

There's a line doing the rounds at the moment that sums up where AI adoption in regulated organisations keeps going wrong: AI isn't a new operating system. It's an accelerator. Apply it to an outdated organisation, and it simply lets that organisation do outdated things faster.

It's a sharp way of putting something borne out by the data, too. In UK social housing, the NHS, and legal and compliance teams, AI tools are being adopted at pace, but the underlying processes they're bolted onto often aren't ready for them. The result isn't transformation. It's the same broken process, running at higher speed, with the same risks baked in.

The numbers behind failed AI adoption

This isn't a hunch, it's one of the best-documented problems in enterprise technology right now. RAND Corporation interviewed 65 data scientists and engineers with years of hands-on AI/ML experience and found that more than 80% of AI projects fail, roughly double the failure rate of ordinary IT projects. Gartner's research points the same way: it now expects over 40% of agentic AI projects to be cancelled by the end of 2027, driven by escalating costs, unclear business value, and inadequate risk controls.

Note what isn't on that list: the models themselves. RAND's researchers were explicit that AI projects most commonly fail because organisations misunderstand or miscommunicate the problem they're actually trying to solve, not because the underlying technology doesn't work.

The NHS provides a close-to-home example of this pattern. A Royal College of Physicians survey found that 68% of doctors believe the NHS lacks the digital infrastructure to introduce AI effectively, with 70% citing poor integration with existing systems like electronic patient records as the main barrier. The AI isn't the problem. The fragmented, non-standardised process underneath it is.

Why this matters more for regulated organisations

For a retail business, a failed AI pilot is a wasted budget line. For social housing providers, NHS trusts, and legal or compliance teams, the stakes are different: a process failure isn't just inefficient, it's a compliance risk.

Take a housing provider managing repairs correspondence, damp and mould reports, and Awaab's Law compliance deadlines across email threads and shared drives. Add an AI summarisation tool on top, and case reviews might get quicker, but the underlying process is still manual, deadlines are still tracked by memory, and the accountability trail is still fragmented. Speed was never the bottleneck. The process was.

The same logic applies to accessible communication. An AI tool that converts a tenancy agreement or hospital consent form into Easy Read format faster is only useful if the organisation already has a clear process for identifying who needs it, when it's triggered, and how it reaches the person in time. Without that, faster output just means more output, produced at the same rate of missed cases.

What successful AI adoption in regulated organisations looks like

The organisations bucking the failure statistics aren't the ones with the most advanced models, they're the ones who fixed the process first, or used the AI adoption moment as the reason to finally fix it. In practice, that distinction looks like:

• Contract and obligation management: an AI tool bolted onto a manual contract review process, versus a platform like Contract intELIEgence, which restructures how obligations, renewal dates, and compliance risk get tracked from the start, so the AI has a coherent process to accelerate, rather than a spreadsheet to summarise.
• Accessible document conversion: a generic AI writing tool producing Easy Read content with no grounding in accessibility standards, versus askVERA, built specifically around the compliance requirements regulated organisations are held to, so the output is usable and auditable, not just fast.

In both cases, the AI isn't standing in for a functioning process. It's sitting on top of one that was designed properly in the first place, which is exactly the distinction RAND's researchers, Gartner's analysts, and the RCP's survey of NHS clinicians all independently land on: the technology is rarely the constraint. The process is.

The question to ask before adopting AI

Before bringing in any AI tool, the useful question isn't “what can this do?” It's: what exactly are we accelerating? If the honest answer is “a process we haven't looked at properly in years,” the AI will get you to the wrong outcome faster, with a more convincing paper trail. If the answer is “a process we've deliberately rebuilt to work,” AI adoption will genuinely change how that team operates.

For social housing, NHS, and legal and compliance teams specifically, that question isn't optional. Getting it wrong doesn't just waste a budget cycle, it produces compliance failures at speed, which is a considerably worse outcome than the slow version.

If you're weighing up where AI actually fits into your organisation's existing processes, rather than where it looks impressive in a demo, get in touch and we can talk through where the gaps genuinely are.