AI Strategy for SMEs: A 90-Day Roadmap
An SME can build a workable AI strategy in 90 days by splitting the work into three blocks: 30 days to frame it (audit, use cases, rules), 30 days to pilot one or two real use cases, and 30 days to decide what to roll out and what to drop. You do not need a data scientist, but you do need a sponsor on the management side and success criteria set on day one.
This roadmap is part of our complete guide to AI in business for SMEs. It is written for managers and HR or operations leads whose teams already use ChatGPT or Copilot on their own, and who want to move from scattered individual use to a managed approach.
The context argues for acting now. According to the France Num 2026 barometer (French Ministry of the Economy, published late September 2026, more than 9,000 companies surveyed), 40% of very small businesses and SMEs now use AI solutions, up 14 points in one year, and 53% of SMEs do. Only 19% of very small businesses and SMEs use paid solutions (32% of SMEs). Our reading: many companies are trying, few are structuring. That is the gap this roadmap is meant to fill.
Why 90 days, and what this roadmap does not cover
90 days is long enough to measure a real result and short enough for management to keep AI on the agenda. Beyond that, SME AI projects get lost between two urgent matters. Below that, you only measure the novelty effect.
This plan does not cover a full company transformation or building a custom model. It aims at a more modest and more realistic goal: by day 90, one or two use cases running, written rules, a trained team, and a reasoned decision on what comes next. If you are starting from zero, begin by assessing your AI maturity: our article How to assess a company's AI maturity gives a simple grid.
One constraint to set from the start: available time. Plan about half a day per week for the sponsor and one to two days per week for the designated AI lead. Without that time, the calendar slips.
Days 1 to 30: frame before you buy
The first month is not for choosing a tool, it is for choosing a problem. SMEs that buy licenses before identifying a use case end up with subscriptions nobody uses.
Here are the four workstreams of the first month.
- Week 1: name a sponsor and an AI lead. The sponsor (owner or director) makes the calls and frees up time. The lead coordinates day to day. In a company of 20 to 50 people, this is often the person already handling digital tools.
- Week 2: take stock of actual usage. Simply ask each team which AI tools they already use, and with what kind of data. You will almost always find undeclared uses, as described in our article on shadow AI.
- Week 3: list 10 to 15 pain points and score them. Each one gets three scores: time lost, data sensitivity, ease of implementation. The top 2 or 3 become your pilot candidates. The detailed method is in our article on the AI audit for SMEs.
- Week 4: write the ground rules. One page is enough: which tools are allowed, which data must never go in, who approves. Our guide to AI policy in business offers a template.
By day 30, you should have one page stating: the sponsor, the lead, the two chosen use cases, the expected value of each, and the decision rule (see below).
Days 31 to 60: pilot one or two use cases, not ten
The second month means running a small number of use cases with a small group, measuring before and after. A successful pilot is limited to 5 to 10 users, one precise task and one numeric indicator.
Take a typical example, purely illustrative (this is not a client case). A 40-person SME chooses to pilot AI on meeting minutes and first drafts of quote replies. Six people from administration and sales take part. Before the pilot, each one logs for a week the time spent on these tasks. During the pilot, they use an AI assistant with shared instructions. If the measured gain is 2 hours per week per person, that is 12 hours per week for the group. This kind of simple calculation then feeds the AI ROI calculation.
Three points determine success in this phase.
- Training before the tool. A pilot without training produces disappointing results and a bad image of AI. Half a day of hands-on practice on your own tasks is enough to start. See the guide to AI training in business.
- A 30-minute weekly check-in. The lead collects what works, what blocks, and the cases where the AI got it wrong. These notes feed the usage rules.
- One indicator per use case. Time saved, response time, share of output needing rework. Avoid vague indicators such as "overall satisfaction".
If you want reference points on what to track beyond the pilot, the glossary entry on KPIs covers the basic vocabulary.
Days 61 to 90: decide, roll out or stop
The third month is for deciding: roll out a use case, fix it, or stop it. That decision must rest on criteria set on day 30, not on the enthusiasm of the moment.
A simple decision rule works well in an SME:
- Roll out if the measured gain reaches the target and at least two thirds of the pilot group keep using it without reminders.
- Fix if the gain is real but the use remains fragile (unstable prompts, poorly prepared data, ill-suited tool). Extend by 30 days with one targeted change.
- Stop if the gain is smaller than the review effort, or if the use creates a confidentiality risk you cannot remove.
Stopping a use case is not a failure, it is the normal outcome of a well-designed pilot. What costs money is rolling out something that has not proven itself.
For the rollout, plan three things: training for the teams that did not take part in the pilot, an update of the rules page with the previous month's lessons, and a quarterly follow-up. AI change management then becomes the main topic, far more than the technology. Our article on AI change management in SMEs covers this.
| Period | Goal | Deliverable | Criterion to move on |
|---|---|---|---|
| Days 1 to 30 | Frame | Sponsor, lead, 2 use cases, one-page usage rules | Both use cases have a quantified expected value |
| Days 31 to 60 | Pilot | Before and after measure on 5 to 10 users | One indicator recorded every week |
| Days 61 to 90 | Decide | Roll out, fix or stop decision, deployment plan | Decision criteria applied as written |
Compliance: build it in from day 1, not day 90
In an AI roadmap, compliance runs alongside the pilot: Article 4 of the EU AI Act and the GDPR apply as soon as your teams use an AI tool. Many deployment plans leave this angle out.
Article 4 of Regulation (EU) 2024/1689 asks organizations using AI systems to ensure AI literacy among their staff. According to the European Commission's FAQ on the topic, the article has applied since 2 February 2025, and enforcement by national market surveillance authorities is announced from 2 August 2026. The Commission also mentions changes under way through the "Digital Omnibus" package, so the exact content of the obligation may evolve and needs to be monitored. Our article on Article 4 of the AI Act and AI training explains what this means in practice.
On personal data, the GDPR still applies: do not enter customer or employee data into a tool whose terms of use you have not checked. Record the forbidden data types in your rules page and keep a trace of the training delivered. That record will serve as evidence in case of an inspection, and as a basis for your funding file (see below).
Budget and funding: what to plan for
The budget for a 90-day roadmap has three lines: internal time, licenses and training. Internal time is the line most often forgotten, even though it usually weighs more than subscriptions.
For licenses, pay only for the pilot users during the first 60 days: you extend subscriptions only at day 90, if the decision is positive. For training, OPCO funds may cover all or part of the cost depending on your sector and headcount; our page on funding AI training through your OPCO explains the process. For overall orders of magnitude, see how much AI costs an SME.
If your organization has fewer than 10 people, our article on AI for micro-businesses on a limited budget adapts this logic to tighter resources.
The mistakes that derail the roadmap
The three most frequent mistakes are starting with the tool, running too many pilots at once, and setting no stop criterion. They show up in most projects that run out of steam before month three.
- Starting with the tool. A subscription bought on day 3 steers the whole project. Start with the problem.
- Multiplying pilots. Five parallel use cases scatter the lead and make every measure unreadable. Two is already a lot.
- Forgetting the stop criterion. Without a decision rule written on day 30, the day 90 discussion turns into opinion.
- Neglecting human review. AI-produced content still needs checking. Review time must enter your gain calculation.
The detail of these pitfalls and others is in our article on AI deployment mistakes in SMEs. For the underlying steps, the article on the 5 key steps of AI adoption in SMEs complements this plan.
Frequently asked questions
Can an SME really build an AI strategy in 90 days?
Yes, provided you limit the ambition: frame, pilot one or two use cases, decide. A full multi-year strategy is built afterwards, using the lessons of these first 90 days.
Do we need to hire a technical profile to start?
No. You need a sponsor and an internal AI lead, trained on the uses. A technical profile becomes useful later, for automations or integrations with your business tools.
How much time should we dedicate to the project each week?
Plan about half a day per week for the sponsor and one to two days for the lead. This is a field estimate to adjust to your company size and the number of pilot users.
Which use cases should we pick first?
Those that combine a large volume of time, low-sensitivity data and simple setup: meeting minutes, first drafts of replies, document summaries, message sorting. Avoid first uses involving health, payroll or identifiable customer data.
How do we measure success at day 90?
With one indicator per use case set on day 30: hours saved per week, response time, share needing rework. Add an adoption criterion: the share of the pilot group still using the tool without reminders.
Is team training mandatory in this plan?
Article 4 of the AI Act requires organizations using AI to ensure AI literacy among their staff, and its exact scope is evolving at European level. In any case, a pilot without training gives poor results: training is part of the plan, and it is documented.
How GrowthPerf helps
GrowthPerf is a Qualiopi-certified training organization specializing in AI, no-code and automation for SMEs and nonprofits. We work on the three stages of this roadmap: an AI audit to frame the use cases, hands-on training on your own tasks (see the operational AI and AI for business programs), then pilot follow-up through to the decision.
To place this approach in a broader view, read the complete guide to AI in business for SMEs.
Need a framework to get started? Request the free GrowthPerf roadmap and talk to us about your first use case.