Every AI journey starts with the same sentence: 'We should be doing something with AI.' What happens in the following 90 days decides whether that sentence becomes a shipped, measured win — or a recurring meeting that produces slides. The difference is rarely talent or budget. It's sequencing: knowing what to do first, what to postpone, and what to refuse to do at all. Here is the sequence, week by week.
Days 1–15 — Look before you leap
Map your processes, discover the AI already in unofficial use, check what data is accessible, and publish a one-page usage policy.
Days 16–30 — Score the ideas
Collect candidates from the people who do the work, then score each on volume, clarity of rules, error tolerance, and measurability.
Days 31–60 — Run one pilot
A single workflow, a named owner, a human in the loop, and a success metric agreed before you start.
Days 61–85 — Measure honestly
Compare against the baseline, count the review time, and listen to the people using it.
Days 86–90 — Decide and tell the story
Ship it, fix it, or kill it — and publish the numbers either way. All three outcomes beat limbo.
#Days 1–15: audit before you automate
Resist the urge to buy anything yet. Spend two weeks learning three things: which repetitive, high-volume processes eat the most hours (ask the people who run them, not the org chart); what AI your teams are already using unofficially — there's always some, and it tells you where demand is real; and whether the data each candidate process needs is actually accessible. Close the fortnight by publishing a one-page AI policy — approved tools, what never gets pasted where — so experimentation is safe instead of secret.
#Days 16–30: from every idea to one pilot
Now collect ideas widely — you'll get twenty or forty, from 'automate our quotes' to 'build a robot lawyer.' The funnel is your friend:
Everything anyone suggests — 20 to 40 is healthy.
Ranked by volume, rule-clarity, error tolerance, measurability.
Sanity-checked for data access and a willing owner.
The one with a clean baseline you can measure against.
#Days 31–60: the pilot
- Keep the scope embarrassingly small — one workflow, one team, off-the-shelf tools wherever possible.
- Name one owner — a person, not a committee, responsible for making it work and empowered to change course.
- Keep a human in the loop — the AI drafts, suggests, or triages; a person approves. Autonomy is earned later.
- Log everything — every input, output, correction, and complaint is data you'll want at decision time.
- Meet weekly, decide fast — a pilot is for learning, and learning that isn't acted on within a week goes stale.
#Days 61–90: measure, decide, tell the story
Compare against the baseline honestly — including the review time humans now spend checking the AI, which enthusiastic teams love to forget. Gather what the users say, not just what the dashboard says. Then make the call: ship it wider, fix a specific weakness and re-test, or kill it and take the next candidate off your shortlist. Whichever you choose, publish the numbers internally. A transparent kill builds more credibility for the next project than a quiet, unmeasured 'success' — and credibility is the real currency of your second 90 days.
“The goal of your first 90 days isn't to transform the company. It's to earn the right to the next 90 — with one measured win, one honest number, and a team that now believes this is real.”
Ninety days from now, you can be a company with a working AI system, a measured result, and a scored backlog of what to automate next — or a company with a slide deck. The path to the first one starts unglamorously: this week, list your ten most repetitive processes and ask the people who run them which one they'd hand to a machine tomorrow. That list is your roadmap. The rest is sequencing.
