Why most AI projects don't move the needle

The headline numbers on AI in UK business look like a success story. More than half of firms are now using it, up sharply in just two years. Spend is climbing. Almost every leadership team has it on the agenda.

Then you look at the return, and the story changes. The British Chambers of Commerce puts active AI use among UK firms at 54% in 2026 — though stricter measures put deliberate adoption far lower, nearer one in six. Either way, only around 31% of firms using AI report a clear positive return so far, and Microsoft with WPI Strategy estimate £78bn of growth sitting unrealised in the SME segment alone. A lot of activity. Not much of it yet landing.

It's rarely a technology problem

When an AI initiative underdelivers, the instinct is to question the model, the tool, or the vendor. In our experience the failure almost never sits there. The model usually works. What's missing is everything around it.

Government research found that the biggest barriers aren't financial at all: most firms simply haven't identified a clear use for AI, and the next most-cited blocker is a lack of skills. In other words, the constraint is direction and capability, not compute.

What the projects that work have in common

The engagements that move real numbers tend to share four habits. None of them are about the technology.

1. They start from value, not from the tool

Winning teams don't ask "where can we use AI?" They ask "where is this business losing time, money, or quality — and could AI change that?" The use case is chosen for its return, then the technology is selected to fit. Not the other way round.

2. They redesign the process, not just bolt AI on

Dropping AI into an unchanged process gets you an unchanged result, slightly faster. The gains come when the operating model is reshaped around the new capability — who does what, in what order, with what checks. That's operational work, not technical work.

3. They measure against the P&L from day one

If you can't say what success looks like in business terms before you build, you won't be able to prove it afterwards. The strongest programmes define the metric up front — hours saved, conversion lifted, error rate cut — and track it honestly, including when the answer is "this one isn't working."

4. They treat adoption as part of the build

A tool nobody trusts or uses returns nothing. Training, change, and clear ownership aren't a phase you bolt on at the end — they're how the value actually gets realised.

A short test before you start

Before committing to any AI initiative, four questions will tell you whether it's likely to pay:

• What specific operational outcome are we trying to change, in numbers?
• Who owns the process today, and are they ready to change how it runs?
• How will we know within weeks — not quarters — whether it's working?
• What happens to the people whose work this changes?

If you can answer those, you're already ahead of most. If you can't, that's the work to do first — and it's cheaper to do it now than after the build.

Sources: British Chambers of Commerce / Atos (2026); UK Government DSIT AI Adoption Research (2025); UK business AI ROI survey reported by TechRound (2025); Microsoft / WPI Strategy (2025). Figures vary by how "AI use" is defined; we've noted the measure beside each.