Last updated August 2026 · 5 min read · By Nico Bitzer, Co-Founder & CEO, Bots & People
The AI adoption gap is the distance between the number of organisations that have bought AI and the number whose people actually work differently because of it. McKinsey's November 2025 State of AI survey put the first figure at 88% and the second at roughly 6%. My argument is that the gap is not a technology problem and not a budget problem. It is the predictable result of handing people a tool and skipping the part where they learn to use it. If you run learning at a large organisation, that gap belongs to you, and closing it is worth more to your board than the next round of licences.
Two worlds, one day apart
A few weeks ago I moderated the AI Enterprise Summit Europe in Munich. Excellent speakers, several with doctorates, all of them on stage talking about autonomous agents and multimodal architectures.
The next morning I was on a video call with a client's HR team. The question on the table was whether their employees were allowed to paste something into ChatGPT. They honestly did not know the answer.
Those two worlds are one day apart. Everything in this article lives in the space between them.
How big is the AI adoption gap?
Wide, and now measurable. McKinsey's State of AI survey, published in November 2025 across 105 countries, found that 88% of organisations use AI in at least one business function. Only around 6% clear McKinsey's bar for high performers, meaning they attribute more than 5% of EBIT to it.
BCG arrived at the same place from a different angle. In a survey of roughly 2,700 executives, 26% had moved past pilots to value at scale.
Adoption is close to universal. Impact is rare. And the high performers are running the same models as everybody else, which rules out the technology as the explanation.
Accenture's research points at what does explain it. The organisations that redesign how work happens around AI, rather than layering it on top of existing processes, invest significantly more in change management and training. Those are the organisations reporting substantially higher revenue growth. The variable is not the model. It is whether anybody changed the work.
Why doesn't a paid AI licence become a daily habit?
Because three things sit between the licence and the habit, and procurement solves none of them.
1. Nobody showed them what good looks like. Salesforce found that 69% of workers had received no training from their employer on using AI at work. Microsoft put the share of AI users who had received company training at 39%. A tool you were never taught to use is a tool you open once.
2. They are afraid. PwC asked 56,000 workers worldwide and 37% worry about AI replacing their job. Fear is the most underrated line item in any adoption programme. Afraid people do not experiment, people who do not experiment do not learn, and people who do not learn do not use the tool. The licence goes quiet.
3. Nobody taught them judgement. This is the one that worries me most. Handing somebody a very fast car without a driving licence is dangerous, and that is close to what an untrained AI deployment does. I have lost count of the AI-generated documents I have read where nobody stopped to ask whether the output was any good before passing it on. Unreflective use is worse than no use, because it goes out with your organisation's name on it.
The tools people use are not the ones you deployed
While the official licence sits unopened, the unofficial tool runs without governance. Cyberhaven measured shadow AI usage growing 485% in a single year. Microsoft's Work Trend Index found that 78% of AI users bring their own tools to work, and Salesforce found that 55% use tools their company never approved.
This is the screensaver problem's twin. One expensive licence goes untouched on a company laptop while the person holding it does the same work in a browser tab nobody can see. Both failures have the same cause. The organisation bought a tool and skipped the learning.
What actually closes the gap?
Teaching people on the work they already have. Not generic prompting, not a one-off webinar, and not a SharePoint page full of links.
At Daimler Truck, the sessions ran on each department's own data, which made using AI in front of colleagues the expected behaviour rather than a confession. The programme reached 20,000 participations across six countries and Copilot Chat usage rose by 85%. That is an adoption number, not a completion number, and it is the difference between a board funding a second year and quietly not.
The same pattern held at PwC Germany, where people moved from everyday Copilot use up to building agents and automations. In both cases the enthusiasts were already there before we arrived. They had been using AI privately for months. What changed is that the company gave them a stage instead of a policy.
There is a reporting consequence worth naming, because it decides whether your programme survives its first budget review. IT reports deployment because licences are what IT can see. L&D reports completion because course records are what L&D can see. Neither of those describes a person doing their job differently on a Tuesday. Until somebody reports that, the 88% and the 6% will keep drifting apart on your watch.
An AI tool without a trained user is a very expensive screensaver. I use that line too often, and nobody has proven me wrong yet.
What to do next
- Measure the behaviour, not the licence. Ask your people anonymously how many times they opened an AI tool for a real work task in the past five working days, then compare that against your licence dashboard. The distance between those two numbers is your actual adoption rate.
- Fix supply before you write policy. If fewer than half your people have access to a sanctioned tool, the policy is not your constraint.
- Run one department on its own data this quarter. Two hundred people trained on their real documents will teach you more about your programme design than another round of stakeholder alignment.
If you want the full argument, including the measurement framework and the three-stage capability model we use, it is in our whitepaper How to Make AI Work for People.
Frequently Asked Questions
What is the AI adoption gap? It is the distance between organisations that have deployed AI and organisations whose people work differently because of it. McKinsey put deployment at 88% of organisations in November 2025 and meaningful EBIT impact at around 6%. The gap is behavioural rather than technical, because both groups run the same models.
Why do 88% of organisations use AI but only 6% get value? Because buying a licence and changing a working habit are different projects. The 6% invest in redesigning how work happens, according to Accenture, rather than layering AI on top of existing processes. The rest deploy the tool and expect the behaviour to follow on its own.
Is shadow AI a security problem or a training problem? Both, but the training problem comes first. People reach for personal accounts when the sanctioned route is missing, slower or unclear. A policy removes the visibility rather than the behaviour, so supply and training have to move before governance can work.
How do you measure AI adoption rather than training completion? Ask a time-bounded behavioural question. How many times did someone open an AI tool for a real work task in the past five working days, and was the tool company-provided? That produces a number describing behaviour, which is what a board will fund twice.
Does fear really affect AI adoption rates? Yes, and it is routinely left out of programme design. PwC surveyed 56,000 workers and 37% worry about AI replacing their job. People who feel at risk do not experiment in front of colleagues, and adoption depends on exactly that kind of visible experimentation.
Nico Co-Founder & CEO, Bots & People
