AI Adoption Plan

Pick a few use cases, redesign the workflow around each one so that an old step actually stops, put most of the effort into people and process, and set the support and review points before launch.

Practice, not research: adoption advice here comes from consulting-firm surveys, self-reported and published by firms that sell AI advice (McKinsey, 2026; BCG, 2024; Bain, 2025). Peer-reviewed economics offers a reason why measured value lags adoption (Brynjolfsson, Rock & Syverson, 2021). The 10-20-70 split is a rule of thumb.

Use it when:

• Your team has moved past scattered personal use and you want a shared, deliberate way of working.
• A tool rollout has stalled: licenses are bought, but little has changed in the work.
• You are about to introduce an approved tool and want the first months planned.

Skip it when:

• You do not yet know which tasks suit AI: have each person fill in 10.7 AI Task Map first.
• The team has no shared rules yet: write 10.10 Team AI Policy before you scale up use.

This page is consulting practice, not research. The large surveys of AI use come from firms that sell advice on it, and they rest on what respondents say. With that caveat, they share a pattern: use is broad, measured business value is narrower, and the organizations that report most value also tend to say they redesigned how work flows rather than handing out access (McKinsey, 2026, not checked against the original page). The surveys cannot show which comes first.

An independent argument points the same way. Economists modeling general purpose technologies argue that the payoff depends on large complementary investments, much of it in new processes and skills, so measured productivity can lag before it rises (Brynjolfsson, Rock & Syverson, 2021). That is a model of whole economies, not a test of any team's plan.

So this page asks for few use cases, a workflow where something stops, and most of the effort on people and process, following BCG's split of attention (BCG, 2024). The split has no published derivation. Treat it as a reminder, not a ratio.

Use the team's 10.7 maps. Choose tasks that are frequent, tested within the tool's reach, and visibly painful today. Each use case brings training, a workflow change and new checking habits, so few beats many.

Strong answer: "First draft of replies to routine claims questions. Done today by all eight handlers. Evidence: two handlers tested it on invented cases in March and the drafts needed light edits. Expected gain: faster replies on simple cases, more time for complex ones."

Illustrative example: the team, the company and all numbers are made up.

The lead of a claims team at a mutual insurer in Norway plans the team's first deliberate use of the company's approved AI tool.

Section 1: two use cases from the aggregated task maps: first drafts of replies to routine claims questions, and summaries of long case files for handovers. A third, drafting decision letters, is left out: the tests showed errors on policy terms.

Section 2: AI drafts before the handler reviews, with level 2 checks for replies and level 3 when amounts are mentioned. The old boilerplate library is retired. Saved time goes to the complex claims backlog.

Section 3: two hands-on sessions using invented cases, the lead using the tool in the weekly meeting, and productivity targets eased for two weeks.

Section 4: two champions with four hours a week each, a Thursday drop-in, and a start moved to August because a new case system goes live in May. The month-two review shows faster replies on simple cases and one near miss caught at the quality gate, which the lead presents as the system working.