A good 90-day AI plan has five parts, run in a sensible order: a baseline of how AI is already used, one or two small pilots, a short usage policy, training for the people involved, and a formal review at day 90. The goal is not to "transform" the company in three months. It is to learn what works in your context, put basic guardrails in place and decide, with evidence, what to scale next.
Days 1–15: set the baseline
Start by finding out what is already happening. Staff are often using AI tools before management has made a decision, so the first job is to ask, not to assume.
- Current use. Which tools are people using, for which tasks, with what kind of data?
- Candidate tasks. List repetitive, text-heavy work in each department: drafting, summarising, reporting, first-pass research, customer replies.
- Owner. Name one accountable person for the plan. Without an owner, a 90-day plan becomes a 9-month drift.
- Starting measures. For the tasks you may pilot, record how long they take today and who does them. You cannot show improvement later without a "before".
Our guide to where SMEs in Malaysia should start with AI helps with choosing candidate tasks.
Days 15–45: run one or two focused pilots
Pick pilots that are small, low-risk and easy to judge. A pilot with a clear owner, a defined group of users and a written success test teaches you more than a company-wide launch. Avoid tasks where a mistake would reach a customer unreviewed, and keep a person checking the output (see what human-in-the-loop means).
For each pilot, write down before you start: the task, the users, the data involved, who reviews the output, and what result would count as success. The full method is in how to run an AI pilot.
Days 20–50: put a simple policy in place
Do not wait for the pilots to finish before writing rules. A one-to-two page policy is enough at this stage: which tools are approved, what data must never be entered, who reviews AI output before it leaves the company, and who to ask when unsure. Involve your legal or compliance team, especially where personal data is involved. This article is general information, not legal advice. See what a company AI usage policy should include.
Days 30–75: train the people who will use it
Tools without skills produce uneven results. Train the pilot users first, because they will shape how the rest of the company works with AI. Focus on hands-on practice with real tasks from your own baseline list rather than general talks. Leaders also need their own session, so that decisions about budget, risk and priorities are made with a shared understanding. Agmo Academy's AI Forward CXO Program is built for that leadership group, and our AI training programs cover team-level skills. If you plan to claim training costs, HRD Corp claimability is subject to eligibility and HRD Corp approval, so check early.
Days 75–90: review and decide
Hold a review meeting with the plan owner, pilot users and a senior sponsor. Compare the pilot results to the success tests you wrote, look at the problems people hit, and decide for each pilot: scale, adjust or stop. Stopping a pilot that did not work is a good outcome — it cost little and taught you something.
What to report to the board
Boards want a few clear indicators, not a dashboard of everything. Use measures you actually collected in your own baseline, and report them honestly, including where the result was unclear.
- Adoption: how many people in the pilot used the tools regularly, from your own usage records.
- Time and quality: the before-and-after on the specific tasks you measured, and whether reviewers found errors.
- Risk: incidents, near-misses and policy questions that came up.
- Decision requested: what you propose for the next 90 days and what it needs in people and budget.
Do not borrow benchmarks from other companies or from articles; they rarely match your processes. For ways to measure results, see how to measure AI training ROI.
Common mistakes to avoid
- Starting with a tool purchase instead of a list of tasks.
- Running five pilots at once with no owner for any of them.
- Writing the policy only after an incident.
- Training once and never following up with the pilot users.
- Measuring nothing at the start, then arguing about results at the end.
Treat day 90 as the end of the first cycle, not the end of the project. The review should produce the next 90-day plan.