COM-B can help diagnose an AI adoption barrier by asking what a specific person needs to do, then examining their capability, opportunity and motivation for that action. Start with the behavior that is holding up useful work. It may belong to the learning lead buying training, the specialist reviewing a request or the manager arranging cover for practice.
That changes the investigation. If staff cannot get suitable training, asking why they have not learned enough starts too far downstream. Find out what prevents the person responsible for arranging it from completing that work.
This article focuses on those enabling actions. For a barrier during an employee's own task, use the guide to distinguishing inability, reluctance and lack of opportunity.
Name the behavior before diagnosing the barrier
“Improve AI readiness” is too broad for a useful COM-B assessment. “The learning lead compares two training proposals against the customer-service team's needs and sends a recommendation to the budget owner” gives you a person, an action and an output to examine.
COM-B stands for capability, opportunity, motivation and behavior. In the original 2011 paper, Susan Michie, Maartje van Stralen and Robert West describe an interacting system. Capability includes physical and psychological capacity, including knowledge and skills. Opportunity concerns external conditions. Motivation includes deliberate evaluations, emotions and habits. The model does not assign one category universal priority.
Use the framework to organize evidence about a named action. A category is a question to investigate, not a verdict about the person. Several conditions can contribute, and acting can change them in turn. The original paper developed an intervention framework and tested classification reliability in public health; it did not test an enterprise AI adoption program.
Work backward from the blocked task until you find a consequential enabling action. Keep the scope small enough to inspect:
- A learning lead gathers requirements and recommends suitable training.
- An IT specialist reviews an approved-access request and returns a usable decision.
- A team manager changes the rota so employees can attend practice within working hours.
Ask the action's owner to show a recent attempt and identify the people they depend on. A request waiting for specialist advice needs a different response from a recommendation awaiting budget approval. The broader people and change guide helps you connect this investigation to the conversations with employees who are affected.
A training lead can need enabling support too
The UK Department for Transport's June 2025 AI readiness report describes a COM-B application to procuring AI training. Its learning and capability lead lacked the expertise to evaluate training options and had limited time, information and access to networks. Internal experts were supportive but busy. The lead was motivated to take responsibility, yet overextended.
The report describes support that included an external learning-needs analysis, dedicated funding, a streamlined approval pathway, internal advisors and regular check-ins. The intervention addressed the person enabling learning, rather than starting with another invitation for staff to train.
This is a reported departmental application, not a measured comparison showing that COM-B caused better adoption. The underlying systems map covered the central department, and its authors identify incomplete influences and a need for further validation. Its practical value here is the change in focus: a willing training owner may need expertise, available advice and a workable approval route.
Examine the owner and the handoff
Suppose Claire Bennett leads learning at a furniture retailer. Customer-service colleagues need to practise checking AI-drafted delivery replies against order and carrier records. Claire has two course proposals, a description of the approved tool and examples of replies with unconfirmed delivery promises. She must recommend a course the service manager can use with the team.
Ask Claire to walk through how she would compare the proposals. The following questions help keep the diagnosis attached to that work:
| COM-B condition | Question for this action | Evidence worth checking |
|---|---|---|
Capability | Can Claire judge whether the exercises teach staff to verify delivery claims? | Her comparison of the proposals against an actual reply and its source records. |
Opportunity | Can she obtain service and technical advice within the purchasing timetable? | Available reviewers, scheduled time, permitted materials and the approval route. |
Motivation | What makes this recommendation worth prioritizing, and what concerns compete with it? | Her account of responsibilities, competing deadlines and the consequences she expects. |
Keep actual skill separate from confidence. The model authors' FAQ distinguishes belief in one's ability from the ability itself. An unsure lead may produce a sound comparison; a confident lead may overlook the checking exercise the team needs.
Cloud and AI value consultant Dan W. makes a related observation in a January 2026 LinkedIn article about FinOps, the practice of managing cloud spending. He describes how cost dashboards can coexist with missing time, ownership and legitimate decision forums. This is his practitioner reflection, not a measured AI rollout result. It is a useful prompt to look beyond another information pack when the owner already understands the request.
Check whether the requested support fits the work
The diagnosis can also reveal a problem in how an enabling action was designed. A lead might know how to buy a course without knowing whether colleagues need that course.
In a December 2025 Reddit discussion, a self-described talent employee at a technology scaleup with more than 150 employees said they had been assigned learning and AI training work. They proposed workshops to break down sales and post-sales tasks. A colleague warned that this felt like exposing jobs to automation. The poster acknowledged the concern and later clarified that they wanted discussion, including more customer time and less administration.
An eLearning designer in the replies questioned whose priorities were being served and whether AI was the right solution. The account does not establish the company's eventual staffing decisions or training outcomes. It shows a question the training owner needed to investigate before treating the planned workshop as useful support.
For Claire, that means asking the service manager and colleagues what they need help completing, what they can share safely and which proposed changes concern them. A course emphasizing faster reply generation could miss the team's immediate need to avoid unsupported delivery promises. Revise the requirement before scoring suppliers against it.
COM-B should leave room for that result. If a proposed use is unsuitable, unauthorized or unable to meet the work's quality standard, changing or stopping it can be the appropriate decision.
Change one condition and inspect what reaches the team
Choose support that follows from the evidence. If Claire cannot assess the checking exercises, arrange a joint review with the service manager using the approved example. If she can assess them but advice repeatedly arrives too late, assign a reviewer and a scheduled decision point. If several urgent assignments compete, the manager responsible for her workload must decide which takes priority.
Record the change without converting tentative explanations into employee scores:
- The action and evidence. Claire's recommendation must explain how each course teaches delivery-claim checking. Record the missing expertise or delayed advice observed.
- The support and owner. Name who will provide the joint review, time or decision, and when it will be available.
- The next output and recipient. Specify the recommendation the budget owner receives and the practice material the service manager needs.
- The follow-up. Inspect whether those outputs arrived and were usable. Keep any remaining barrier visible.
Suppose the joint review reveals that neither proposal includes checking against source records. Claire now has a reason to request a revised exercise before recommending either course. That is a useful procurement decision even though no purchase has been completed.
When a course is selected, inspect the whole handoff. Can the service manager schedule practice, access approved examples and arrange help for uncertain answers? A signed contract establishes a purchasing milestone. It does not establish that colleagues can use the learning, that replies improved or that AI adoption increased.
Test those later outcomes separately. The guide to running a credible AI pilot explains how to choose a comparison and keep failed or missing work in the result. COM-B can inform the intervention you test; it cannot supply the result in advance.
Questions and answers
Whose behavior should a COM-B assessment examine?
Examine the specific person's action that currently blocks useful AI work. That could be an employee checking a reply, but it could also be a learning lead procuring training or an IT specialist returning an access decision. Define the action, output and recipient before choosing questions. Use the enabling-action examples to trace the dependency without treating low employee usage as the diagnosis.
Is COM-B an AI adoption maturity score?
COM-B is a framework for understanding behavior through capability, opportunity and motivation. It does not turn those categories into a validated company AI maturity score. Collect evidence about an actual action and retain competing explanations rather than averaging employee ratings into a verdict. The owner-and-handoff comparison shows questions for one training recommendation, with no scoring threshold.
What if the lead knows what to do but approval is missing?
For an AI training purchase, a missing approval is a condition around the lead's action. Check whether the request contains what the approver needs, who has authority and when a decision can be returned. Resolve that dependency before prescribing more training for the lead. If expertise is also missing, address both conditions and inspect the next output to see what remains unresolved.
Does every COM-B diagnosis lead to more AI training?
No. A COM-B investigation can point to advice, time, access, workload priorities or a changed requirement. It can also reveal that the proposed AI use does not suit the work. For an employee struggling during a task, compare skill, reluctance and missing opportunity before choosing support. For the training owner, first verify that the course addresses something colleagues actually need to do.
