Why 73% of AI tools bought by healthcare practices go unused — and how to avoid being in that majority

Three quarters of the AI tools purchased by healthcare organisations collect digital dust within six months. That figure — 73%, from Gartner's Healthcare AI Adoption Survey (2025) — tends to surprise people when they first hear it. It shouldn't. The more interesting question is what separates the 27% that actually stick.

The answer is not better technology. The tools that succeed and the tools that fail are often technically comparable. The gap is almost always in what happened before and immediately after purchase. Specifically: whether anyone did a proper analysis of fit before signing the contract.

This piece sets out what the evidence says about why clinical AI adoption fails in independent practices, and what the practices that do get a return on their investment actually do differently.

The three failure modes — and how common each one is

In the literature on technology adoption in healthcare settings, three failure modes account for the overwhelming majority of abandoned implementations. They're worth naming precisely, because each one has a different fix.

Failure mode 1: Wrong problem fit. The tool solves a problem the practice has, but not the most important one. A physiotherapy clinic buys AI-assisted scheduling when the bigger time drain is clinical documentation. The tool gets used, but the impact is small, and within three months the enthusiasm fades. This is the most common failure mode for independent practices — not because the tool is bad, but because no one mapped the practice's problems before selecting a solution.

Failure mode 2: Staff non-adoption. The owner buys the tool, the team was not involved in the decision, training was a 30-minute vendor onboarding session, and within weeks the staff have found workarounds that don't involve the new system. Research published in JMIR Medical Informatics (2024) found that clinician involvement in AI tool selection was the single strongest predictor of sustained adoption — stronger than the tool's usability score, stronger than the price point. The decision-making process matters more than the product.

Failure mode 3: Infrastructure mismatch. The tool requires data quality or system integration the practice doesn't have. A dental practice buys an AI recall messaging tool that requires structured patient data exports from their PMS — but their PMS is ten years old and doesn't export in the required format. Three months of troubleshooting later, the project dies. This is the readiness problem, and it's the most preventable failure mode of the three.

So what does the 27% do differently? In almost every case, they answered three questions before buying: does this tool address our highest-value problem? Is our team ready to change how they work? And does our infrastructure actually support it? Most practices answer none of them.

Why practices buy without auditing

The purchasing process for AI in independent healthcare is mostly broken. Practices get pitched at conferences by vendors who lead with outcomes — "saves 2 hours a day", "40% reduction in no-shows" — without context about what kind of practice those results came from, what their starting baseline was, or what implementation support they had.

Peer recommendation fills the gap. Someone on a professional forum says they use a particular clinical note tool and it's changed their practice. That's compelling, and it's not useless information — but it tells you what worked for them, not what will work for you. A physiotherapy practice in Leeds with a 4-day appointment cycle, one admin member, and EMIS Web is a completely different implementation context from a psychology group in Bristol with Cliniko, six therapists, and a practice manager dedicated to digital.

The vendors don't help. Their job is to sell the tool. Their ROI calculators are built to show you a number that justifies the purchase. An independent audit is not their business model.

The result: practices buy tools that are technically fine but contextually wrong. Six months later, they've spent money, burned staff goodwill on a failed implementation, and are more sceptical of AI than they were before.

The audit gap: what a structured pre-purchase analysis actually looks at

What distinguishes an evidence-led approach from buying on recommendation is the specificity of the pre-purchase analysis. At minimum, a proper fit assessment covers four things.

Workflow mapping. Where is time actually being lost? Not where you think it's being lost — where it demonstrably is, measured against a typical week. Clinical documentation, scheduling management, billing, referral letters, patient communications: each area has a different AI solution profile. The practice that's losing 90 minutes a day to clinical notes needs a different tool from the practice that's losing time to manual appointment reminders. A good analysis produces a ranked list of time-cost by workflow area, so you're solving the right problem first.

So what for you: before you look at any AI tool, spend one week logging where your admin time goes. Thirty minutes at the end of each day. The answer almost always surprises practice owners — and it almost always reorders their purchasing priorities.

Staff readiness assessment. This is the failure mode that kills the most implementations, and it's the one that gets skipped most often. Readiness is not whether your team is "up for it" in the abstract. It's whether they have the basic digital literacy to use the tool, whether their current workload has any capacity to absorb a change, whether the practice culture handles change well, and whether anyone has thought about who will champion the new system once you've moved on to the next problem.

The research is clear on this: staff involvement in the selection process is the strongest predictor of sustained adoption (JMIR Medical Informatics, 2024). That doesn't mean asking your receptionist to choose the tool. It means involving them in the problem definition, showing them the options, and addressing their concerns — which are usually legitimate — before you buy.

So what for you: before evaluating any AI tool, run a 30-minute session with the staff who will use it. Ask them what's slowing them down. Ask them what they'd need to feel confident using a new system. Their answers will tell you more about implementation risk than any vendor demo.

Infrastructure check. Every AI tool has dependencies that vendors understate in their sales materials. Data format requirements. Integration APIs with your PMS. Internet connectivity at the point of care. Device compatibility. GDPR compliance and data processing agreements. The infrastructure mismatch failure mode is almost entirely preventable with a 30-minute technical checklist run before signing a contract.

For UK practices, there's an additional regulatory dimension. The Information Commissioner's Office (ICO) requires a Data Protection Impact Assessment (DPIA) for any AI tool processing patient data. The tool must have a data processing agreement in place. If the AI makes clinical recommendations rather than just administrative ones, it may fall under MHRA's AI as a Medical Device classification. These aren't bureaucratic obstacles — they're genuine patient safety and liability questions that belong in the pre-purchase process, not after you've committed.

So what for you: ask every vendor for their UK data processing agreement before the demo ends. How they respond tells you a great deal about whether they understand the regulatory environment you're operating in.

Financial case. Vendor ROI calculators are not financial cases. They're marketing. A real financial case models the specific cost of your current problem (time × cost per hour × weeks per year), the realistic implementation cost (tool subscription + training time + transition period), and the conservative estimate of what changes — not the vendor's best-case scenario.

A clinical note AI tool that saves a clinician 20 minutes per appointment across 8 appointments a day is worth £X per year at their equivalent hourly rate. Calculate it yourself, with your numbers. If the payback period is under 12 months at conservative estimates, the economics are probably sound. If you need the vendor's optimistic projections to make it work, they probably aren't.

So what for you: the honest financial case takes about 45 minutes to build with a spreadsheet. If the numbers don't work at 50% of the vendor's claimed benefit, the tool is too expensive or the problem isn't big enough. Either is useful to know before you've signed a 12-month contract.

What this means for your next AI decision

The 73% figure is not an argument against AI in healthcare. It's an argument against buying AI the way most practices currently buy it — reactively, without analysis, on the strength of a vendor demo and a peer recommendation.

The practices in the 27% — the ones with tools that are still running 18 months after implementation, with staff who actually use them and outcomes that justify the cost — didn't get there by luck. They got there by doing the analysis first. The tool selection came second.

An independent audit of your practice covers all four dimensions above and produces a ranked list of where AI fits your specific workflows, with a financial case for each one and an honest readiness assessment. It costs significantly less than a year of a subscription for a tool that goes unused.

The maths — as the Gartner data makes clear — points firmly toward auditing first. The 73% that don't learn this lesson learn it through a wasted year and a more cynical team. The 27% that do get a return on their investment that compounds over time as they add more tools to a foundation that actually works.

Related: AI Readiness Score: where most independent practices actually sit in 2026 · The AI Opportunity Audit™ · Book a free discovery call