Your First AI Project: A 30-Day Plan for Business Owners

Your first project should be small enough to finish in a month and useful enough that people notice. A good first project usually:

• happens every day or every week
• follows a predictable pattern
• annoys your team or slows down your customers
• is easy for a person to check
• does not involve your most sensitive data

Good examples: answering common customer questions on WhatsApp, sending appointment reminders, sorting incoming emails or preparing a weekly sales summary. Poor first choices: replacing your whole accounting process, or anything where a mistake could cause serious harm.

Sit with your team and ask a simple question: "What do we do over and over that we wish we did not have to?" Write down the answers, then choose one.

Next, describe how the task works today. Who does it? How often? How long does it take? What goes wrong? Then set one or two simple goals, such as "Customers get an answer to common questions within minutes, at any hour" or "No appointment goes without a reminder." Write down how things are today so you can compare later.

Finally, name one person as the owner of the project. Without an owner, it will drift.

AI is only as good as what you give it. This week, collect:

• real examples of the task, such as past customer messages, emails or reports, with personal details removed
• your best answers and the way you like things written
• the rules: what is allowed, what is not, and when a person must step in
• the facts it needs, such as opening hours, services and policies

Then choose your tool or partner and set it up for this one task only. Check the privacy terms before connecting anything to customer data, and keep Morocco's personal data protection law (Law 09-08, overseen by the CNDP) in mind whenever personal information is involved.

Do not launch to everyone at once. Test with a small group: two or three staff members, a few friendly customers, or one branch.

Ask testers to try normal cases and difficult ones. Write in Darija, French and Arabic. Ask strange questions. Try to confuse it. Every mistake you find now is one your customers will not see later.

Review the results every day or two. Fix wrong answers, fill in missing information and adjust the tone. Keep a person checking the important outputs throughout.

Go back to the goals from Week 1 and compare. Ask:

• Did it save time, and roughly how much each week?
• Were the results accurate enough?
• Did customers and staff find it helpful?
• What still needs a human?

Then make a clear decision:

1. Continue: it works, so roll it out more widely.
2. Adjust: it is promising but needs a few more weeks of improvement.
3. Stop: it did not help, so keep what you learned and try a different task.

Stopping is not failure. A small test that teaches you something costs far less than a big project that goes wrong.

A first AI project in Morocco follows the same 30-day plan, but a few local realities deserve attention from Week 1.

Test real Darija, not textbook Arabic. Your customers write in Darija, French, Modern Standard Arabic, and often a mix of all three in one message. Many also write Darija in Latin letters with numbers, like "3afak" or "wach kayn". AI tools handle French and standard Arabic well, but results in Darija can be less consistent. Use real past messages as test cases in Week 2 and Week 3, and decide early which language the assistant should reply in.

Start where your customers already are. For most Moroccan SMEs, that means WhatsApp. A good first project is often a WhatsApp assistant that answers common questions about prices, opening hours, delivery or stock, and hands over to a person for anything else. Make sure the hand-over is clear, so a customer is never stuck talking to a machine.

Take personal data seriously. Customer names, phone numbers and order histories are personal data under Law 09-08. Before connecting an AI tool, check what data it receives, where it is stored, and whether it leaves Morocco. Transfers of personal data abroad and certain types of processing can involve formalities with the CNDP. If you are unsure, ask a legal adviser before the pilot, not after.

Budget in dirhams. Many AI tools are billed monthly in dollars or euros, and some charge per message or per use. Estimate the cost in MAD based on your real volume of messages, and check how the cost would grow if you rolled it out to all customers.

Plan for your team, not just the tool. In a small family business, the person who answers WhatsApp today may worry about the change. Involve them as the owner of the project. They know the questions, the tone customers expect, and the cases that always need a human.

Keep it small. One task, one channel, one branch. That is enough to learn what works for your customers.

• Starting too big. One task done well beats five tasks half done.
• Skipping the examples. Without real examples, the assistant has to guess.
• No owner. Someone must be responsible for checking and improving it.
• Not telling the team. Staff should know what is changing and why.
• Removing the human too soon. Keep people involved until you trust the results.
• Forgetting it after launch. Your information changes, so the setup needs regular updates.

Africodex helps business owners in Morocco and across Africa choose a first AI project, set it up and run the pilot, from a WhatsApp assistant to automated reminders and reports. If you would like a second pair of eyes on your idea, we are happy to talk.