An AI Strategy in 90 Days: A Realistic Plan for Mid-Sized Companies

"We need an AI strategy" has become a standing item on board agendas, in Mauritius as everywhere else. The trap is treating it as either a slogan or a two-year consulting epic. Ninety days, seriously used, is enough to produce something better than both: a short written strategy, a prioritised list of use cases, one pilot in motion, and rules for how the company uses AI safely.

Here is a plan that mid-sized organisations can actually execute, structured as three 30-day blocks.

Days 1 to 30: See clearly

The first month is diagnosis, and it should be brutally honest rather than aspirational.

Start with decisions, not technology. Interview the leadership team and the two levels below them with one question: which recurring decisions or tasks consume the most time or cause the most expensive mistakes? Collect twenty to thirty candidates. You are hunting for pain, because pain funds projects.

In parallel, run a data readiness review. For each candidate problem, ask whether the data to address it exists, where it lives, who controls it, and whether it contains personal data covered by the Data Protection Act. Many candidates will die here. That is the point of doing it now rather than after budget approval.

Finally, take stock of what is already happening. In most companies, staff are already using AI tools unofficially. Survey it without blame: the goal is a map, not a disciplinary file. Shadow usage tells you where demand is real.

Deliverable for day 30: a one-page current-state summary and a long list of candidate use cases with a data feasibility note on each.

Days 31 to 60: Choose deliberately

Month two converts the long list into a strategy, and the key discipline is saying no.

Score every candidate on two axes: business value (time saved, revenue protected, risk reduced) and feasibility (data readiness, process clarity, regulatory sensitivity). Plot them. You are looking for the quadrant that is valuable and feasible, and you should be suspicious of anything that scores maximum on both, because it usually means the scoring was optimistic.

Select exactly one pilot, with two runners-up named in writing. One, because a mid-sized firm rarely has the attention to run two first pilots well, and named runners-up, because it defuses the internal politics of not being chosen first.

Good first pilots share a profile: a bounded process, measurable today, with a tolerant failure mode. Document drafting, customer email triage, demand forecasting for one product family, or first-pass invoice matching all fit. Anything customer-facing and irreversible does not.

Then write the strategy itself, and keep it to three or four pages: where AI will and will not be used, the pilot and its success metrics, the budget envelope, and who decides what. If your team lacks the experience to scope the pilot technically, this is the moment to bring in outside help; a specialist partner like nexus.mu can pressure-test the shortlist and the architecture before money is committed.

Days 61 to 90: Move and govern

Month three runs on two parallel tracks.

Track one launches the pilot. Assign a business owner (not IT alone), agree the baseline measurement before switching anything on, and timebox it: eight to twelve weeks to a keep, fix, or kill decision. Measure against the baseline weekly.

Track two builds the guardrails while enthusiasm is high:

• An acceptable use policy. One page: what data may never be pasted into public tools, which tools are approved, and who to ask when unsure.
• A human accountability rule. AI output that leaves the building or affects a customer gets a named human reviewer. No exceptions in year one.
• A review rhythm. A monthly 30-minute steering session covering pilot metrics, new use case requests, and incidents.

Deliverable for day 90: pilot live and measured, policy published, steering rhythm running, strategy document approved.

What ninety days will not do

Be clear with the board about what this is not. It will not transform the company, retrain the workforce, or produce dramatic savings by day 90. It produces something more valuable at this stage: an honest map, one real experiment, and the organisational muscles to run the next ten. Companies that skip to the transformation usually spend more and learn less.

The quiet advantage of the 90-day frame is momentum. Strategy documents age badly; operating rhythms compound. If on day 91 your company knows its data, runs one measured pilot, and meets monthly to decide the next one, you do not just have an AI strategy. You have an AI capability, and that is the thing competitors cannot copy from a slide deck.