Mauritius now has a National AI Strategy and FAIR Guidelines. That is significant. It gives the country a public direction on skills, innovation, governance and adoption. It does not, however, tell a small business which process to change next Tuesday morning, who should own an AI tool, or whether the information in its shared drive is ready to be used safely.
That distinction matters. A national strategy can create momentum, convene institutions and set expectations. A small firm still has to make its own operating choices. The useful question for an owner is not, “What does the strategy promise?” It is, “What should be different inside my business because the environment is changing?”
The official launch materials are worth reading directly, particularly because the FAIR Guidelines are described as a public-sector framework. They should not be presented as a new private-sector rulebook for every SME. The Ministry’s announcement is a better starting point than a chain of secondhand summaries. Existing obligations, including data-protection responsibilities, remain a separate question.
Start with one work problem
The first move is deliberately unglamorous. Name one recurring piece of work that is slow, inconsistent or hard to supervise. It might be turning meeting notes into follow-up actions, answering the same customer questions, preparing first drafts of proposals, or identifying incomplete files before a deadline.
Avoid beginning with a tool. “We need ChatGPT” is not a business problem. “Our account managers spend four hours each week preparing routine client updates, and quality varies by person” is. The second statement gives you something to test, measure and improve. The first gives you a subscription.
The National AI Strategy should make this discipline more urgent, not less. More choice will enter the market. More vendors will use the language of AI readiness. A clear problem statement is one of the few reliable ways to avoid buying a polished answer to the wrong question.
Make a simple inventory before you experiment
Most small firms already use AI in some form. Staff may be drafting text with public tools, recording meetings through an assistant, using automated recommendations inside a business platform, or pasting customer information into a service they signed up for individually.
Ask three questions:
- Which AI-enabled tools are already being used?
- What information goes into them?
- Who is accountable when the output is wrong, misleading or shared too widely?
This is not bureaucracy. It is basic visibility. A firm cannot govern what it cannot see, and it cannot make a sensible investment decision while informal use is happening outside the discussion.
Draw the boundary around sensitive information
The next step is to define information that must not be entered into a public AI service without explicit approval. For many SMEs, that list will include personal data, client contracts, financial records, unpublished pricing, employee matters and information supplied in confidence.
The wording can be short. What matters is that it is specific enough for a person under time pressure to use. “Be careful with data” is not a boundary. “Do not paste a client’s payroll file, passport copy or signed contract into a public AI tool” is.
Where an AI system makes, rather than assists with, a decision about a person, the risk increases sharply. Hiring, eligibility, pricing, credit and disciplinary decisions should not be treated as ordinary productivity experiments. The earlier guide to the National AI Strategy explains why existing data-protection questions need to be considered separately from the new policy direction.
Choose a pilot that teaches you something
Pick one contained workflow with a named owner. A good first pilot has a visible before-and-after measure, limited consequences if it fails, and a real user who can say whether it improved the work. It should use representative material, not a carefully cleaned demonstration dataset that bears little resemblance to normal operations.
That last point is where many pilots become misleading. A tool can look impressive during a presentation and still fail in ordinary work because the inputs are incomplete, the output needs too much correction, or nobody has time to oversee it. A pilot is useful when it reveals those conditions early.
Build capability around the work
Training should be attached to the workflow being changed. People need to know what a useful prompt looks like in their context, how to check output, when to stop using the tool and where to raise a concern. A general demonstration may create enthusiasm. It rarely changes a routine.
Give teams permission to report where the tool creates rework. This is valuable evidence, not resistance. An organisation that only celebrates positive stories will keep scaling weak use cases because nobody wants to be the person who says the output is unreliable.
Review the pilot as a management decision
After four to six weeks, review the work with the owner and the users. Did the task become quicker or more consistent? Where did review time increase? Did the output create a new risk? Has the time saved been redirected to useful work, or simply disappeared into another queue?
The decision at this point may be to scale, adapt or stop. Stopping is a valid outcome if the use case does not carry its weight. What matters is that the decision is explicit and based on what happened in the business, rather than on a general feeling that AI must be useful somewhere.
Mauritius’s strategy creates a favourable moment to learn. The firms that benefit will not be the ones that react with the longest tool list. They will be the ones that get better at diagnosing work, setting boundaries and learning from a small number of honest experiments.