Strategy
AI adoption strategy: a 90 day plan for a 10 to 250 person business
A week-by-week AI adoption plan for SMEs: diagnose, train, ship one workflow and set up governance inside a single quarter.
The short answer
A workable AI adoption strategy for an SME runs in one quarter and in this order: diagnose where AI pays in weeks one to three, train managers and teams in weeks three to six, build and evaluate one workflow in weeks four to ten, put it into production with monitoring by week twelve, and report a single number to the board. Sequencing matters more than tooling. Businesses that buy licences before building capability stall at low usage and no measurable result.
Key takeaways
- Diagnose, train, deploy, operate. In that order, in one quarter.
- Pick one workflow. Multiple pilots dilute attention and prove nothing.
- Name an internal owner in week one. Programmes without one decay.
- Baseline your metrics before anything changes or you cannot show the gain.
- Governance should fit on two pages at this company size.
Adoption strategies written for enterprises assume a data team, a change function and an eighteen-month horizon. None of that exists in a 60-person business. This is the version that fits, and it is deliberately a quarter long, because a quarter is how long attention lasts.
Before week one: two decisions
Name the internal owner. Not a committee. One person with enough authority to change a process, ideally an operations or commercial leader rather than whoever is most technical. Second, agree the single business number this programme is meant to move, and write it down. Cost to serve, response time, gross margin, quote turnaround. Everything downstream is judged against it.
Weeks 1 to 3: diagnose
- Observe a week of real work in two or three departments. Watch it, do not just ask about it.
- Log recurring tasks with frequency and minutes per instance, then convert to annual hours and cost.
- Map which systems each candidate workflow touches, and confirm API access and who administers each one.
- Score workforce capability by department. Be honest, because this determines the sequence.
- Rank use cases by payback, and cost the top five including running cost.
- Baseline the metrics you will report on. This step is skipped constantly and regretted later.
Output: a ranked, costed shortlist and one chosen first workflow. Choose something frequent, structured and cheap to get wrong.
Weeks 3 to 6: build capability
Training runs in parallel with the early build, not after it. Managers first, because they specify the work and can block it. Then the frontline team that will live with the first workflow.
- Leadership session: business cases, risk, data boundaries, what good sponsorship looks like.
- Manager sessions: mapping their team's week, writing an automation brief, deciding what stays human.
- Frontline sessions: hands on with their own live work, plus verification habits.
- Pick two internal champions and involve them in build reviews from here on.
Weeks 4 to 10: build and evaluate one workflow
- Write the specification: inputs, rules, exceptions, escalation and definition of done.
- Assemble an evaluation set of 30 to 100 real historic cases with known correct outcomes.
- Build against the specification, integrating with the systems the humans already use.
- Score against the evaluation set and set an acceptance threshold before launch, not after.
- Shadow run: the system handles real cases while a human checks every output.
- Fix the top failure patterns, rescore, then reduce review to sampling.
Weeks 10 to 12: production and governance
- Monitoring in place: volume, containment or completion rate, error rate, escalations, cost.
- A named owner for the live workflow and a weekly ten-minute review of sampled outputs.
- A two-page policy: approved tools, data that must not leave your tenancy, when a human must review, how to report a problem, who signs off new use cases.
- A short internal write-up of what changed, shared with everyone, including what did not work.
- The next two use cases queued, with the roadmap updated using what you now know about your own integration reality.
What to report to the board
| Metric | Why it earns a place |
|---|---|
| Movement in the one chosen business number | The whole point of the quarter |
| Hours recovered on named tasks | Concrete and easy to challenge, which is good |
| Completion or containment rate of the live workflow | Shows reliability, not enthusiasm |
| Total cost: build plus running plus internal review time | Prevents a favourable comparison built on omission |
| Use cases proposed from inside the business | The leading indicator of whether capability is real |
Five traps
- Buying licences first. Seat spend without capability produces low usage and a bad taste.
- Running six pilots. Attention divides, none reaches production, the programme is judged a failure.
- Automating the visible work rather than the expensive work. The noisy task is rarely the costly one.
- Leading with headcount reduction. Adoption depends on the people who would be reducing themselves.
- No owner after launch. Every deployed workflow degrades without someone reviewing it.
What quarter two looks like
With one workflow live, a trained management layer and a real cost baseline, quarter two is where the compounding starts: two or three more workflows, deeper integration, and the beginning of the cross-system work that carries the larger returns. The reason it is possible is that quarter one built the capability rather than just buying software.
Frequently asked questions
- What is an AI adoption strategy?
- A plan that sequences diagnosis, capability building, deployment and operation against a named business outcome. For an SME it should fit in a quarter and end with one workflow in production, an owner accountable for it, and a baseline you can measure future work against.
- How long does AI adoption take for a small or medium business?
- One quarter to a first production workflow is realistic for a business of 10 to 250 people: three weeks to diagnose, training in parallel, six weeks to build and evaluate, and two weeks to move into production with monitoring. Cross-system workflows take longer.
- Should we start with training or with building AI agents?
- Run them in parallel, with manager training slightly ahead of the build. Managers who can write a clear automation brief make the build better, and teams who understand the tools adopt the result instead of working around it.
- Who should own AI adoption in an SME?
- One named person with the authority to change a process, usually an operations or commercial leader rather than the most technical person available. Committees stall, and unowned workflows degrade quietly once the launch attention fades.
- How do we measure whether AI adoption is working?
- Baseline before you start, then report movement in the one chosen business number, hours recovered on named tasks, completion rate of the live workflow, total cost including running and internal review time, and the number of use cases proposed from inside the business.
Work through this with us
brep runs AI operating reviews, workforce training and agent deployment for businesses of 10 to 250 people in the UK and US. One diagnosis, one workflow, someone accountable for keeping it working.