Diagnosing adoption opportunity and work context · Lesson 4

Analyze affected groups and incentives

6 min read

Lesson objectives

  • Diagnose a specific adoption barrier from observed behavior and incentives rather than labeling every objection resistance, then select a testable intervention.
Reading tools

A project manager may call employees “resistant to AI” because only twenty percent have used an assistant. That rate describes behavior; it does not explain it. Employees may not know when use is permitted, may know but fear being blamed for a tool error, or may have tried and found that review does not fit into their shift. Begin with groups affected in different ways: the worker performing the task, the supervisor assessing performance, the customer receiving an output, the data team managing access, and the risk owner able to stop a harmful pilot. Do not just list supporters and opponents. Ask what each group could gain or lose, what decision it controls, and what evidence might change its view. A worker who raises a real data concern provides design information, not a communication problem to be soothed by a slogan.

Prosci describes ADKAR as five elements of individual change: awareness of why change is needed, desire to take part, knowledge of what to do, ability to perform, and reinforcement that helps behavior last. It is a diagnostic lens, not a campaign for persuading everyone or proof that a tool fits the task. If a worker heard an explanation but cannot restate the reason for change, awareness may be the barrier. If they understand the reason and procedure but avoid a trial because rewards depend on response volume rather than accuracy, repeating the slide deck will not change the incentive. If they want to try and attended training but fail on exception cases, practice and a better interface or review allowance may be needed more than another motivational message. Seek independent observations for each hypothesis to avoid circular diagnosis.

A brief task-specific interview, observation of real work, and cross-role measures help. Ask, “Show me the last case where you could have used the assistant but did not; what happened?” rather than “Why do you refuse change?” Compare the answer with actual system access. Perhaps a review team cannot open the original document even though the program assumes it checks that document. Separate technical ability from willingness; both can produce the same nonuse. Beware of hearing only leaders: affected workers may be reluctant to say that a new productivity target pushes them to skip review. Gather feedback without turning the interview into an individual performance rating, and explain how it will improve the design.

How can we tell whether those interventions address the cause? Decide on tests before changing the workflow: detection of an omitted condition in ten new cases of balanced difficulty, review time, and the share of summaries needing correction. If safe and feasible, compare a team able to open the source within the screen with one using the previous method, while maintaining review for both. Watch who adopts and who still avoids the tool; shifts, languages, or case types may differ. If usage rises but error detection stays poor, do not expand; redesign the task. If detection improves but resolution slows, check whether fewer reopened cases compensate for review time. Do not ask workers to rush a necessary check solely to improve the manager’s dashboard.

OECD’s analysis of skills and work in the AI era discusses changed tasks and training needs for employees and managers. It does not establish an adoption rate for this fictional center or dictate its pay policy. Prosci explains elements of individual change; its sequence is not an automatic individual diagnosis from one survey. Labelling an objection as “lack of desire” and punishing nonuse can conceal a serious design flaw. A more defensible approach considers competing explanations: skill, incentive, access, tool fit, or a justified objection. Each needs evidence that could support or disconfirm it and a different intervention, followed by another observation.

A stakeholder matrix should guide action rather than decorate a presentation. For each group, record the task changing, a plausible benefit and burden, its ability to influence the decision, a feedback route, and an initial action. A risk owner may have high formal authority but little daily tool use; frontline staff have high exposure and limited formal authority. The former needs risk evidence and a stop point; the latter needs practice time and a way to challenge outputs. Customers do not attend training, but answer accuracy and correction rights affect them. One message that “AI raises productivity” would leave their distinct questions unanswered. Revisit the matrix when the use scope changes, because the affected population may change as well.

Lesson summary

Diagnose a specific adoption barrier from observed behavior and incentives rather than labeling every objection resistance, then select a testable intervention.