Where AI pilots stall on the way to production

Before the pilotStaff are already using AI, on accounts the company does not manage

Only 40% of companies in the MIT study had bought an official large language model subscription. At more than 90% of them, staff said they used personal AI tools for work.¹

Two conclusions follow, and they pull in different directions. Demand already exists, because employees have found their own uses for these tools. The approved tool, when it arrives, competes with whatever people do on their phones, and it loses if it cannot handle the same tasks.

There is a security side as well. Work material is moving through accounts nobody in IT can see. A ban without a better approved option tends to push that use further out of sight, which is the opposite of what the ban was for.

Inside the pilotThe sample was cleaned by hand

A pilot team can fix its data manually. Nobody can do that across a whole business. Gartner surveyed 1,203 data management leaders and found that 63% of organisations either lack the data management practices AI needs or cannot say whether they have them. It expects organisations to abandon 60% of AI projects through 2026 where the data is not AI-ready.²

Other surveys say the same thing in their own terms. In Informatica’s 2026 study of 600 data leaders, 57% named data reliability as a key barrier to moving AI from pilot to production.³ Fivetran found that 42% of enterprises said more than half of their AI projects had been delayed, had underperformed or had failed because of data readiness.⁴

Gartner’s advice runs to five steps: align data to specific use cases, work out the governance requirements AI adds, make metadata active, prepare pipelines for training and for production, and keep testing the data once it is live.² The first real test of a pilot, in our reading, is whether its last-mile data preparation can be automated at all.

Inside the pilotCorrections that vanish

MIT names learning capability as the main barrier it found. The tools in its sample often failed to keep feedback, adapt to a team’s context or improve with use. One mid-market manufacturing COO quoted in the study said that LinkedIn makes it sound as though everything has changed, while day-to-day operations look much the same.¹

This is easy to test during a pilot and rarely tested. Ask five users to correct the system on the same task over two weeks, then check how many corrections still hold at the end without being repeated. A figure close to zero means the pilot is a demonstration, and in production it would repeat the same mistakes at full volume.

The approval meetingRunning costs show up after the decision

Gartner gives three reasons for its forecast that more than 40% of agentic AI projects will be cancelled by the end of 2027: escalating costs, unclear business value and inadequate risk controls.⁵ In McKinsey’s 2026 survey, one respondent in five said their organisation was limiting AI use because of operating costs.⁶

Costs come late because pilots are small. Usage is low, vendors sometimes discount it, and the people checking outputs are the enthusiasts who volunteered. The full bill arrives when the system opens to everyone.

After approvalLarge organisations need longer, and finish stronger

MIT reports that mid-market firms took about 90 days from pilot to full implementation, while large enterprises took nine months or more.¹ IBM’s survey of 2,000 CEOs points to one reason: half said rapid investment had left them with disconnected, piecemeal technology.⁷

McKinsey’s 2026 figures show the other half of the story. Among respondents from large organisations, 40% now say they are scaling AI agents, against 22% at smaller ones.⁶ Size slows the start and seems to help later, perhaps because the platform, approval and ownership work that large firms do early pays back once projects are ready to scale.

For a sponsor, the practical point is the timeline. Nine months is normal for a large enterprise in this evidence. A plan written for three will report a delay in month four while the project is on schedule.

What the survivors shareThey change the work before they widen access

McKinsey found that nearly three-quarters of its high performers had fundamentally redesigned workflows because of AI, up from 55% a year earlier. Among other respondents the figure was about a quarter.⁶ BCG reports that 33% of its "future-built" companies use AI agents and spend about 15% of their AI budget on them, against 12% of scalers and almost no laggards.⁸

Across these studies the order of events is consistent. The organisations that get through pick a process, give it a business owner, agree what a good result looks like and only then put the model inside it. Wider access comes last.

For your next reviewSix symptoms and where to look first

Each row rests on the evidence above, and each needs your own data before it means anything for a particular project.