A 90-day plan for AI adoption
By Sascha Cutura

More UK businesses are using AI than a year ago. The government's AI adoption survey for 2026 puts adoption at 32%, double the 16% recorded in 2025. The barriers that remain are familiar. Among businesses that had already adopted AI, 67% said limited skills, expertise or knowledge was a barrier. Among those planning to adopt, the figure was 78%. And 65% of businesses not yet using AI cited a lack of a clear business case or proven return.
Those are management problems, and they can be worked on in a single quarter. A 90-day plan is long enough to name an owner, prepare the people who will use the system, set a few rules, and run one guarded pilot you can measure. It is short enough that the work has to stay focused.
Diagnose first
The first month is for a clear picture. Tool choice comes later.
Name one accountable owner, with the budget authority for the 90 days and the right to stop a use case. A steering group can advise. One person still has to own the work. Write the name on the brief before you start the scan.
Then inventory what is already running. Official tools, unofficial ones, and AI features sitting inside software you already pay for. You cannot govern what you have not listed.
From there, shortlist three to five use cases. Score each one for value, feasibility, risk, data readiness and the skills required. Sit with the people who do the work and record how it actually runs, including the workarounds. Then set a baseline the organisation already tracks: hours, cycle time, error rates, first-contact resolution, whatever number already has a life in the business. A new metric invented for the pilot is harder to defend at day 90.
The output of month one is a one-page shortlist and a 90-day brief, with owners and baselines.
Enable the people and the rules
The second month is about readiness. A tool rolled out to unprepared teams creates shadow use and weak checks.
Build a role-based skills plan. Start with the owner and the people who will run the first pilot. A whole-organisation programme can follow. Keep the sessions practical. People should leave knowing what they may use, what they must check, and where to escalate.
Write a short AI use policy that people can actually follow. Start a use-case register and assign reviewers. Check data protection, access and residual risk for the shortlisted work.
The output is a first cohort trained, a one-page policy, and a live register.
Deliver one guarded pilot
The third month is for a thin version of a single use case, piloted with the trained team, measured against the baseline.
At day 90 you should be able to say whether to scale, iterate or stop. If the figure has not moved, or the checks are not good enough to reach a fair test, say so. A stop or review point agreed before the pilot starts makes that conversation much easier.
Hand over ownership, run arrangements and any further training as part of the decision, so the work does not sit in a project folder once the 90 days end.
Four rules that keep the quarter honest
One named owner, with the authority to stop work. Every shortlisted use case has a baseline the organisation already tracks. The people who will use the system are trained before it goes wider. A stop or review point is agreed before the pilot starts.
If those four are in place, the rest of the plan is much easier to run. If they are missing, another tool will not fix the quarter.
This is the structure we use in The 90-Day AI Adoption Plan, a short guide for leaders who have been asked to get AI working in their organisation. It includes an opportunity canvas, a skills plan and a checklist you can use as you go.



