The 88/12 AI readiness gap: why deployment is not adoption

Most boards have already declared victory on AI. They were premature.

Eighty-eight per cent of organisations report regular use of AI in at least one business function, up from 78% a year earlier, according to McKinsey's 2025 global survey. Only about a third have begun to scale beyond pilots.¹ At the level of the individual worker the picture is starker: a 2026 Gallup survey of more than 22,000 employees found roughly 12% use AI daily,² and PwC's independent figure of 14% daily use of generative AI, 19% among office workers against 5% among manual ones, barely moved from the year before.³ Deployment has saturated. Adoption has not.

The pairing mixes its denominators on purpose, and the gap it exposes is real. The 88% counts organisations that have switched something on. The 12% counts people who reach for the tool without being told to. The distance between the two is where most AI budgets quietly disappear, and it has nothing to do with how many licences were bought.

A learning gap, not a tooling gap

MIT's "GenAI Divide" research, built on 150 interviews, a 350-person survey and analysis of 300 deployments, found that 95% of enterprise generative-AI pilots delivered no measurable impact on profit and loss. The cause it identified was not model quality but a learning gap: tools that fail to adapt to real workflows, and organisations that fail to adapt around the tools.⁴ New capabilities now arrive switched on by default, with no rollout moment to mark them, so the readiness gap reopens every quarter.²

Gallup's February 2026 study of 23,717 employees names the behavioural levers that separate the adopters from the holdouts. Where workers strongly agree their AI tools integrate well with existing systems, 88% are frequent users, against 55% who do not. Manager support shows a similar split, 78% against 44%, with support for experimentation and clear policy close behind.⁵ Adoption is a question of workflow and leadership before it is a question of skill.

That reframing matters because the obvious counter-argument is sound as far as it goes: if integration does most of the work, the tooling is where leaders should spend, and the people can be left alone. The trouble is that knowing which workflows are broken, and which managers have gone quiet, is itself a measurement problem. Deployment quality and individual readiness are not rivals. One is how you find where the other has failed.

The adoption map nobody has

Two distortions sit inside the standard picture. The first is unsanctioned use. MIT found that while only 40% of companies hold official large-language-model subscriptions, 90% of the workers it surveyed reported using personal AI tools for their jobs.⁶ The populations and methods differ from Gallup's, so the figures should not be set against each other as a contradiction; read together they say something more uncomfortable, which is that the instruments disagree wildly about a behaviour every organisation claims to manage. A readiness programme built on sanctioned-usage surveys may be aiming at a gap that has already moved.

The second distortion is stratification. Adoption concentrates at the top: 67% of leaders are frequent users against 46% of individual contributors, with a quarter of contributors not using AI at all.⁵ The people closest to high-volume, repeatable work are the least-supported adopters. Meanwhile the market has begun pricing the skill directly. PwC's AI Jobs Barometer finds workers with AI skills command a 56% wage premium over otherwise-identical colleagues, more than double the previous year's gap.³ Individual readiness has become a priced asset and an attrition risk, no longer a soft metric. IDC, for its part, puts the cost of the wider skills shortage at up to $5.5 trillion by 2026, while only a third of organisations describe themselves as fully ready for AI-driven ways of working.⁷⁸

What follows is a measurement discipline rather than a training budget. A credible readiness view tracks assessed proficiency by role family and observed usage behaviour rather than course completions, reviewed monthly as a small set of stable indicators.⁷ Course-completion theatre tells you who sat through the webinar. It does not tell you who changed how they work.

The harder part is doing this without breaking the trust the behaviour depends on. Gallup's own levers, manager support and safe experimentation, only operate where people feel supported rather than watched. Readiness measured as team-level patterns and behavioural signals can guide intervention; the same data collected as individual keystroke counts will simply teach people to perform compliance, the very failure mode the 12% figure already hides. So the question for any board that has declared AI done is a plainer one: do you know which of your people have actually changed how they work, or only which of them have been handed a licence?

Footnotes

1. McKinsey & Company. (2025). The state of AI: Global survey 2025. McKinsey & Company. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai ↩

2. Chief Learning Officer. (2026, March 16). From AI access to workforce readiness. Chief Learning Officer. https://www.chieflearningofficer.com/2026/03/16/from-ai-access-to-workforce-readiness/ ↩ ↩²

3. PwC. (2025). Global workforce hopes and fears survey 2025. PwC. https://www.pwc.com/gx/en/issues/workforce/hopes-and-fears.html ↩ ↩²

4. Estrada, S. (2025, August 18). MIT report: 95% of generative AI pilots at companies are failing. Fortune. https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo ↩

5. Gallup. (2026). AI in the workplace: What separates adopters and holdouts. Gallup. https://www.gallup.com/workplace/704252/workplace-separates-adopters-holdouts.aspx ↩ ↩²

6. Legal.io. (2025). MIT report finds 95% of AI pilots fail to deliver ROI, exposing 'GenAI divide'. Legal.io. https://www.legal.io/blog/5719519/MIT-Report-Finds-95-of-AI-Pilots-Fail-to-Deliver-ROI-Exposing-GenAI-Divide ↩

7. Sarder, R. (2026, April 10). Building AI-ready workforce analytics dashboards. Forbes Technology Council. https://www.forbes.com/councils/forbestechcouncil/2026/04/10/building-ai-ready-workforce-analytics-dashboards/ ↩ ↩²

8. Workera. (n.d.). The $5.5 trillion skills gap: What IDC's new report reveals about AI workforce readiness. Workera. https://www.workera.ai/blog/the-5-5-trillion-skills-gap-what-idcs-new-report-reveals-about-ai-workforce-readiness ↩