Sense Behavior, Not Usage: The Change Model AI Adaptation Needs
Edmundo Barrientos · 2026-09-09 · 3 min read
TL;DR: Deloitte's change model for AI adaptation is a continuous loop: sense behavior, intervene, then design experiences. You cannot buy that loop in three parts. You have to build it, and sensing comes first.
The ticket nobody looked at
A support agent runs the AI assistant on every ticket. On one, the assistant states the refund policy wrongly. The agent sends it without challenge. Usage sits at 100%, judgment at zero, and the license dashboard files the month as a win.
Deloitte's July 2026 analysis names that blind spot: "Adoption tells you someone opened the door. It tells you nothing about whether they changed how they work, how they think, or what they're capable of on the other side." Deloitte reports that fewer than 60% of people with AI access use it daily, and 84% of organizations have not redesigned jobs or workflows.
Deloitte's answer, read as a build specification
Their change model has three moves:
- Sense actual behavior, not usage.
- Intervene on real-time data, in the flow of work.
- Design experiences that build behavior over time, not compliance in the moment.
Deloitte calls these a continuous adaptation loop, whose parts reinforce one another.
That is the target state. Hold it against the one bad ticket, and two problems appear that Deloitte does not solve.
Bought in parts, the loop never starts
Most teams buy the three moves separately: a survey to sense, a nudge app to intervene, a catalog to deliver experiences. Each tool covers one move, and the loop still breaks, because what the survey catches never reaches next quarter's catalog. The ticket goes untouched.
Sensing is the foundation the rest stands on. No survey would have found this ticket, because a survey senses what people say about their behavior. Ask that agent's cohort whether they verify AI output, and they will say yes, because they know which answer is expected. The behavior appeared only where the work happened: in the ticket, on the call, in the draft a manager sends back a third time. That is also the only place it can be scored rather than guessed.
So sensing has to sit inside the work, and the other two moves depend on it. Real-time intervention needs something watching the work as it happens. Practice that builds over time needs something to build on, and the ticket is it.
What closes the loop
One category is built this way: autonomous skill building in the flow of work. It builds skills inside the work, and it proves them by measured skill transfer, not completions. Your docs, calls, and recordings already contain the training, including the refund policy. Practice built from them adapts to what each person needs, because the system saw the answer nobody challenged.
The catalog keeps its job here. It simply cannot see. Compliance training fades in a week for the same reason: nothing watched what happened after the quiz, so nothing adapted. No loop survives being bought in three parts. Teams that try will report adoption again next year.
Related reading
- Course Completion Is Not Skill
- Five Communication Behaviors You Can Actually Score
- Autonomous Skill Building in the Flow of Work
References
- Deloitte. AI adoption to AI adaptation, July 2026.
About the publisher: this article is published by Grasply (grasply.ai), a capability builder. Grasply turns what a company already knows, its docs, calls and recordings, into skills people use at work, for employees, partners and customers. Practice turns knowing into doing, and Grasply shows proof that the work changed.