From Pilot to Rollout: Why Only a Quarter of AI Programmes Get Past the Pilot

Most AI pilots look great in the demo and quietly die in month three. A team of enthusiasts gets access, builds a few impressive workflows, and then the momentum flattens.

The reason is almost never the model. It is the absence of a rollout system.¹1By "system" we mean owners, routines, and a baseline measure — not more software. Every item in it is organisational; none of it is a procurement decision.

The stall has a measured size

This is the most thoroughly documented failure in enterprise AI, and four independent sources put it in roughly the same place.

BCG's October 2024 survey of 1,000 executives across 59 markets found that 74% of companies had yet to show tangible value from AI, that only 26% had moved beyond proofs of concept, and that just 4% were consistently generating significant value¹.^◦ Deloitte's State of Generative AI in the Enterprise Wave 4, published January 2025 from 2,773 director-to-C-suite leaders, found more than two-thirds saying that 30% or fewer of their current gen-AI experiments would be fully scaled within three to six months².SourceBCG, "Where's the Value in AI?", October 2024By the numbers26%of companies get past proof of concept to tangible value — BCG, 2024, 1,000 executives in 59 marketsSourceDeloitte, "State of Generative AI in the Enterprise", Wave 4, January 2025

Gartner predicted in July 2024 that at least 30% of gen-AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs and unclear business value³.² And MIT's Project NANDA study in 2025 found around 95% of enterprise gen-AI pilots delivering zero measurable P&L return⁴.³SourceGartner press release, 29 July 20242That Gartner figure is a forward-looking prediction, not a measured outcome — cite it as an expectation the market has priced in, not as history.SourceMIT Project NANDA, "The GenAI Divide: State of AI in Business 2025" (via Fortune)3The MIT number needs its caveats. It rests on 150 leader interviews, a 350-employee survey and 300 public deployments, and it is not peer-reviewed. "Zero return" means no measurable impact on the P&L — not that the tools were useless. Individual productivity gains were common in the same report.

The Thai picture sharpens it further. AWS's 2026 Thailand study, run with Strand Partners across 2,000 respondents, found continuous AI use rising to 43% of organisations from 32% the year before — real, fast adoption. But 74% remained at basic use, meaning chatbots, off-the-shelf tools and incremental process tweaks, and only 9% had reached advanced use⁵.^◦ Adoption is not the Thai bottleneck. Depth is.⁴SourceAWS with Strand Partners, "Unlocking Thailand's AI Potential 2026" — vendor-commissioned and self-reported (via Forbes Thailand)By the numbers9%of Thai organisations have reached advanced AI use; 74% remain at basic use — AWS with Strand Partners, 20264Vendor-commissioned and self-reported — AWS sells the infrastructure the study encourages buying. Directional, not audited.

The pilot trap

A pilot proves a tool can work. It does not prove your organisation can absorb it. Those are different problems:

• Pilots select for motivated early adopters. Rollouts have to reach the sceptical majority.
• Pilots run on enthusiasm. Rollouts need routines, ownership, and measurement.
• Pilots tolerate rough edges. Rollouts expose every gap in training and process.

BCG has measured that second population twice, and the movement between the two readings matters more than either one. In 2025, across more than 10,600 employees in 11 countries, regular gen-AI use among leaders and managers ran above 75% while frontline use sat at 51%⁶. By the 2026 edition — 11,749 workers across 14 markets — frontline regular use had reached 74%, up 23 percentage points in a year^7◦.SourceBCG, "AI at Work 2025"SourceBCG, "AI at Work 2026: Strategy Matters More Than Tools", June 2026 (via BCG press release)By the numbers51% → 74%frontline regular AI use across BCG's 2025 and 2026 AI at Work editions — the access gap closed inside a year

So the frontline caught up on access. What did not follow was the work itself: in that same 2026 survey, 47% now spend more time directing AI than doing the task, and 66% get little or no guidance on what to do with the time they save. The enthusiasts in your pilot are still not a small version of your organisation — but the tail you have to reach is no longer people without access. It is people with access and no redesigned job to put it in.

The question is not "did the pilot work?" It is "what has to be true for the 400th person to use this without you in the room?"

Why scaling the tool is the wrong move

The instinctive next step after a good pilot is to buy more licences and widen access. Every serious study of what actually moves the needle says the licences are not the variable.

McKinsey's March 2025 State of AI survey of 1,491 organisations tested around 25 attributes against bottom-line impact. The one with the biggest effect was fundamental workflow redesign — and only about 21% of adopters had redesigned any workflows at all⁸.^◦ The same survey found more than 80% of organisations seeing no tangible enterprise-level EBIT impact from gen AI, and fewer than one in five tracking well-defined KPIs for their gen-AI solutions.⁵SourceMcKinsey, "The State of AI", March 2025By the numbers~21%of adopters had redesigned any workflow, though redesign was the strongest predictor of EBIT impact — McKinsey, 20255McKinsey's EBIT findings are correlations with self-reported impact, not causal measurements. The direction is well supported; the magnitude is not.

MIT's study found the same thing from the other end: the driver of failure was a "learning gap" in workflow integration, not model quality. Notably, purchased or partnered solutions succeeded roughly 67% of the time, while internal builds succeeded about a third as often — which is the opposite of what most engineering-led AI programmes assume.

BCG's rule of thumb for where AI value actually comes from is 10-20-70: 10% algorithms, 20% technology and data, 70% people, process and culture⁹.^◦ Whether or not the exact split holds, it points at the same conclusion as the measured studies: you cannot buy your way out of the 70%.SourceBCG's 10-20-70 heuristic — a consulting rule of thumb, not a measured statisticDefinition10-20-70 BCG's heuristic for where AI value originates — 10% algorithms, 20% technology and data, 70% people, process and culture. A consulting rule of thumb from client experience, not a measured statistic.

Microsoft's 2026 Work Trend Index put a measurement next to the heuristic. Across 20,000 knowledge workers in ten markets, organisational factors — culture and manager support — accounted for 67% of reported AI impact, against 32% for individual mindset and behaviour: more than twice the weight^10◦. Microsoft's own summary of its findings is blunter than anything we would write: workers are ready, their organisations are not.⁶SourceMicrosoft, 2026 Work Trend Index Annual ReportBy the numbers67% vs 32%how much of reported AI impact comes from organisational factors versus individual mindset — Microsoft, 20266Microsoft sells the Copilot this study encourages buying, and the sample is knowledge workers who already use AI — so it cannot speak for non-adopters. Read it as a vendor study whose direction is corroborated by BCG's independent 2026 survey.

What a rollout system actually contains

1. Named owners per function — not a central AI team, but a person in each department accountable for adoption there. This is the single component most correlated with programmes that survive.^◦RelatedWhat changed when adoption got a named owner
2. Role-specific workflows — generic training produces generic non-use. People adopt what maps to their actual Tuesday.^◦FigureA rollout holds when a named owner carries it, department by department.
3. A measurement loop — baseline, target, and a monthly read on real usage, not licence counts. Fewer than one in five organisations has this, which is why so few can defend their AI budget.^◦RelatedPutting a defensible number on adoption
4. A visible feedback channel — where friction gets reported and fixed fast enough that people keep reporting.

Note what is not on that list: a better model, a bigger licence pool, or another all-hands training session.

The part that is harder than it sounds

Four honest warnings, because a rollout system is easy to draw and hard to keep alive.

Named owners get quietly absorbed by their day jobs. Nobody withdraws from the role; the role just stops happening. It survives when it is written into the person's objectives with a named backup, and it dies when it depends on enthusiasm. Assume you will need to re-appoint at least once in the first year.

The measurement loop decays faster than the workflow. The monthly read is the first thing dropped when a quarter gets busy, and by the time anyone notices, the baseline is stale and the win is unprovable. Put the read in a recurring calendar slot owned by someone whose job is not AI.

Wider access without judgment training spreads confident errors. The 2023 jagged-frontier study by Dell'Acqua and colleagues, across 758 BCG consultants, found large gains on tasks inside AI's capability — 25.1% faster, over 40% higher quality — but on a task outside it, consultants using AI were 19 percentage points less likely to be correct than those without¹¹.^◦ Scaling a workflow scales whatever judgment surrounds it, including its absence.^◦SourceDell'Acqua et al., "Navigating the Jagged Technological Frontier", 2023By the numbers−19pphow much less likely AI-assisted consultants were to be correct on a task outside AI's capability — Dell'Acqua et al., 2023RelatedMeasuring judgment before you widen access

"Basic use" is a comfortable place to stop. The AWS Thailand finding — 74% at basic use, 9% advanced — describes organisations that did roll out successfully and then plateaued. A rollout system gets you to competent, routine use. Getting past that is a different project with a different appetite for risk, and pretending otherwise is how a programme coasts for two years on a win from its first quarter.

Where to start

If you are mid-pilot right now, do not scale the tool. Scale the scaffolding.^◦ Pick one more department, install a named owner, define three role-specific workflows, and instrument usage before you widen access again.^◦DefinitionScaffolding The supporting system around a tool — owners, workflows, measurement, and feedback — the parts that make adoption hold.RelatedSee how one firm did exactly this

Adoption compounds when the system around the tool is deliberate. It stalls when the tool is the only thing you shipped.

What to steal

• Before widening access, write down the baseline you will compare against. If you cannot state it in a sentence, you do not have one.
• Appoint owners by function, in writing, with a backup — and review the appointment at six months.
• Redesign one workflow properly rather than introducing AI into ten. Redesign is the variable with the evidence behind it.
• Instrument real usage, not licences. A licence count has never told anyone anything.
• Buy before you build unless you have a specific reason not to. The success-rate gap is large and counter-intuitive.
• Set the date now for the conversation about getting past basic use, so the plateau is a decision rather than a default.