An AI Adoption Plan for a 10-100 Person Company

An AI adoption plan for a 10-100 person company fits on one page and inside 90 days. Four phases. First, map what your team already does with AI and pick the three use cases worth doing properly, days 1 to 14. Second, set the foundations: company accounts on a tool with defensible data terms, a one-page usage policy, and a training hour, days 15 to 30. Third, run one supervised pilot per use case against a baseline you measured before starting, days 31 to 75. Fourth, a scale-or-stop gate: keep what beat its baseline, kill what did not, write down both, days 76 to 90. That is the entire plan. The rest of this note is how to run each phase, and why skipping the boring middle phase is the reason most adoptions stall.

Why do most small companies stall?

Because they experiment instead of adopting. A May 2026 SAS and IDC study of more than 1,600 SMB leaders found nearly 70 percent of small and midsize businesses remain in the experimental or opportunistic stages of AI maturity: tools in use, no plan around them. In practice the experimental stage looks like the situation I described in Shadow ChatGPT: Your Team Already Uses AI. Now What?: real usage, personal accounts, zero visibility, and nobody able to say what is working. A plan is not about adding AI. It is about converting usage you already have into something measured, safe, and deliberate.

Phase 1: map and choose (days 1-14)

Run the amnesty hour: the team shows what they already use AI for, no judgment. Inventory the tasks, then choose three use cases by one formula: hours consumed per week times how easily the person who owns the task can check output quality. For most companies the three land close to the list in The First Three AI Use Cases Worth Automating: drafting, routine admin chains, and first-response triage. Resist choosing analytics or autonomous agents in round one; both need foundations you do not have yet.

Phase 2: foundations (days 15-30)

Three deliverables, all small. Company accounts on one general assistant whose business plan contractually keeps your data out of training. The one-page policy: approved tools, banned data, company accounts only, a named owner, a review date, exactly as laid out in The One-Page AI Usage Policy Your Company Actually Needs. And one training hour that doubles as quality school: proofing, editing, and fact-checking AI output before it leaves the building.

For contrast, the consultant-grade version of this phase runs committees and cohorts: Pertama Partners' 90-day roadmap assigns weeks to governance committee formation, tool evaluation, training cohort design, and a CEO-endorsed final policy. That structure is right for a few hundred employees. At 10-100 people, the same outcomes compress into an owner, a page, and an afternoon, and the saved weeks go into piloting.

Phase 3: supervised pilots (days 31-75)

One pilot per use case, sequenced, not parallel: tool overload is the primary failure mode of small-company adoption, and three pilots run one at a time beat three run at once. Each pilot needs a baseline measured before launch (how many hours the task took, what quality looked like), a human reviewing everything that leaves the building, and a weekly fifteen-minute check on three numbers: time saved, quality versus the baseline, and total cost including subscriptions and the humans' review time.

Phase 4: the gate (days 76-90)

Hold the meeting most companies skip. For each pilot: did it beat its baseline? Keep and standardize what did. Kill what did not, in writing, so it does not zombie back next quarter. Update the policy's approved-tools list with what survived. Then, and only then, pick the next three use cases. If one of the survivors has grown into a customer-facing AI feature, your questions change from adoption to trust, and that is a different discipline with different instruments; my explainer The Agent Observability Gap is the map for that territory.

Running phases 1 and 2 for you, with the pilots scoped and the baselines set, is precisely my AI Adoption Sprint: the usage mapping, compliant tool selection and setup, the one-page policy plus team training, in 2 days + follow-up at EUR 1,800 fixed.

FAQ

Do we need an AI strategy document first? No. At this size the plan above is the strategy. Write a longer document when you have pilot results worth generalizing from, not before.

Who should own the plan? One named person with authority over tools and budget, usually an ops lead or a founder. The owner role from the policy and the plan owner should be the same human.

What if a pilot fails? Then the plan worked: you spent thirty days and a subscription fee to learn a real answer, instead of a year of vague dissatisfaction. Kill it at the gate, document why, and promote the next candidate use case.

Is 90 days realistic alongside day jobs? Yes, because the plan's total new work is small: an hour of mapping, an afternoon of setup, one training hour, and fifteen minutes a week per pilot. The calendar time is for habits to form and numbers to accumulate, not for effort.