The AI rollout is technically perfect, but the team is not using it. This is not an AI problem, it is an adaptability problem. Here is the metric that predicts which rollouts land.
AI does not fail at the model. It fails at the human in front of the model. And we are not measuring the human.
AQ before AI.
In the last twelve months, almost every senior leader I speak with has the same story. The AI capability is built. Copilot is rolled out, the bespoke agent is in production, the data pipeline is clean. Adoption is reported at sixty percent in the dashboard. Behaviour change in the work is closer to ten.
The reflex diagnosis is the tool, the training, the use case. It must not be good enough. The reflex remediation is more training. More prompts. More demos.
Both are likely wrong. The use case is likely fine. The training is more than likely sufficient. The variable that is failing is the adaptive capacity of the people in front of it. We have just not been measuring that variable.
AI adoption asks something specific of a person. It asks them to unlearn a method that previously worked.
The senior associate who built their career on writing perfect first drafts is now being asked to draft with an AI partner. The methods that got them promoted are the methods they need to drop.
The senior leader who built their reputation on briefing precision is now being asked to delegate to an agent that does not need a perfect brief. The senior analyst who used to spend Friday afternoons on reconciliation is now being asked to spend it on judgement.
In every case, the technical capability is not the bottleneck. The unlearning curve is. AI adoption is a behaviour-change programme dressed in a technology budget. The metric you need is not adoption. It is unlearning.
AI adoption is, at its heart, a behaviour-change ask. And behaviour change is what AQ measures. Two dimensions of the AQ Ability domain map directly onto the specific demand AI puts on a person.
Unlearning. The conscious capacity to let go of a method that previously worked. AI rollouts ask senior associates to retire the perfect-first-draft habit, ask senior leaders to retire the perfect-brief habit, ask analysts to retire the manual-reconciliation habit. Each of these is an unlearning event before it is an AI event.
Mental Flexibility. The cognitive range to hold the new approach alongside the existing one without rushing to choose. AI partnership requires the comfort of "I and the model" rather than "I or the model".
Together, these two dimensions describe an AI-readiness profile we can read at the individual and team level.
The supporting industry evidence is consistent: SnapLogic's AI at Work study (2025) and Archieapp (2025) show that organisations implementing AI-powered coaching report a 72% increase in productivity and up to 2x faster digital tool adoption. The numbers are external; the read is the same. AI value is unlocked by the human, not the model.
Start treating it as a behaviour-change programme that happens to use a tech artefact. The instrument you need is not the licence dashboard. It is an AQme profile of your most-impacted teams, with specific attention to unlearning and mental flexibility.
When you have that, three moves become available. You can predict adoption at the team level. You can target the development investment at the dimensions that actually move the curve. And you can re-measure in ninety days and prove the human capability shift, not just the licence usage.
First, run an AQme assessment on the most-impacted teams. Pay specific attention to unlearning and mental flexibility. The composite indicates how steep the adoption curve will be.
Second, identify the bottom-decile group on unlearning. These are not low-performers. They are often your most experienced staff, whose past methods are most rewarded. They need a different intervention than the broad training programme. They need permission, role-modelling, and a defined transition arc.
Third, re-measure at ninety days. If unlearning has moved by a band, the rollout is on track. If it has not, your training programme is the wrong shape.
By 2028, the AI question every board asks will not be "what is our adoption percentage?". It will be "what is our adaptive readiness?". The companies measuring AQ in 2026 will have a two-year head start in answering.
AI is the largest unlearning event of the last fifty years of work. The companies that recognise it as an unlearning event, not a licence event, will out-adapt the ones who do not.
The AI adoption question is not a technology question. It is a human behaviour question. Let go. The companies that crack it will be the ones that compound through the rest of the decade.
In most cases the model, the use case and the training are fine. AI adoption fails at the human in front of the model, whose adaptive capacity to unlearn old methods has not been measured or developed. AQai's view is that this is an adaptability problem, not a technology problem.
Adoption measures licence usage on a dashboard. Unlearning measures whether people can let go of the methods that used to work so they can adopt the new one. AQai argues unlearning is the metric that actually predicts whether an AI rollout changes behaviour in the work.
Two AQ Ability dimensions map directly onto what AI asks of a person: Unlearning, the conscious capacity to let go of a method that previously worked, and Mental Flexibility, the range to hold a new approach alongside the existing one. Together they describe an individual and team AI-readiness profile.
Run an AQme assessment on the most-impacted teams with attention to unlearning and mental flexibility, identify the bottom-decile group on unlearning, then re-measure at ninety days. If unlearning has moved by a band, the rollout is on track; if not, the training is the wrong shape.
The people with the most to unlearn are frequently the most experienced, because their past methods are the most rewarded. They are not low-performers. They need permission, role-modelling and a defined transition arc rather than the broad training programme.
Take your AQme and learn the AQ Essentials at essentials.aqai.io.
By Ross Thornley, Co-founder, AQai. Author of Decoding AQ: Your Greatest Superpower.
Copyright adaptai Ltd. All rights reserved. AQ® and AQai® are registered trademarks of adaptai Ltd.
.png){kind=small}
.png)
