AI1 October 2026
Where AI agents fit in a growing Rwandan business
Most growing companies don't need an AI that does everything. They need one agent that takes a single repetitive job off the team, like answering booking questions or confirming mobile-money payments. Here is how to choose that first job, keep a person in the loop and know within a month whether it is paying off.
The pitch for AI agents often sounds like replacing whole departments. For a growing company in Kigali, the useful version is much smaller: one agent that handles one repetitive task from start to finish, inside the tools your team already uses.
Start with the task, not the technology
Good first candidates share three traits. They happen dozens of times a week, they follow rules you could write on a single page, and the information needed to handle them already sits in one of your systems: a booking database, a spreadsheet or a CRM. Questions like “Is my order ready?” or “Has my mobile-money payment arrived?” are typical.
Chatbot, agent or simple integration?
If the answer is always the same, a clear FAQ or a quick-reply template may be enough. If data just needs to move from one system to another, a plain integration is cheaper and more reliable than AI. An agent earns its place when it has to understand free-form messages, look something up and then take an action.
Keep a person in the loop
Decide upfront which actions the agent may take on its own and which need approval. Refunds, cancellations and anything involving money should go to a person at first. Give customers an obvious way to reach a human, and make sure staff see the whole conversation when they take over.
Test in your customers' languages
Customers write in English, French, Kinyarwanda and Swahili, often mixing them in a single message. Before launch, test the agent on real messages from your own inbox, in every language your customers use.
Measure the first 30 days
Track how many requests the agent resolves without help, how often it hands over to a person, how long customers wait and what each conversation costs. Read a sample of conversations every week. If the numbers or the transcripts look wrong, narrow the agent's job before you widen it.
Once one agent works reliably, the second is far easier to build: the data connections, monitoring and review habits are already in place.