Most of the AI advice aimed at businesses is written for companies that don’t look like yours. It assumes a data warehouse, a machine-learning team, and a budget measured in millions. For the typical Kenyan SME, NGO, or mid-size firm, that advice is noise.

The good news: the highest-value AI projects today need none of that. They need a clear problem, a willingness to change a workflow, and tools that already exist. Here’s where we tell clients to start — and what to ignore.

Start where the work is repetitive and text-shaped

AI is best, right now, at reading and writing language at scale. So the first wins are almost always in places where your team already moves a lot of text by hand:

  • Drafting and replying. Supplier emails, customer support responses, proposal first drafts, social copy. A well-prompted assistant turns a 30-minute task into a 5-minute review.
  • Summarising and extracting. Turning long documents — contracts, reports, meeting notes — into structured summaries or pulling specific fields out of messy PDFs.
  • Translating and localising. Moving between English and Swahili (and back) at a quality that’s genuinely usable, instantly.

These are unglamorous. They’re also where the hours actually go, which is exactly why the return shows up in the first week.

Three projects worth running first

If you want a concrete sequence, this is the one we recommend most often:

  1. An internal assistant grounded in your own documents. Point a chat assistant at your policies, price lists, and FAQs so staff get correct answers instantly instead of interrupting a colleague. Low risk, immediate payoff, and it teaches your team how to work with AI.
  2. A document workflow. Pick one painful paper-shaped process — invoice intake, application screening, claims — and automate the reading-and-routing step. You keep the human in the loop for the decision; the machine does the tedious extraction.
  3. A drafting layer for a customer-facing team. Give sales or support a tool that drafts on-brand replies they edit and send. Measure the time saved per ticket; it compounds fast.

Notice what’s not on this list: a custom model, a chatbot on your homepage that no one asked for, or a “data strategy” project that takes a year to produce a slide deck.

What to skip (for now)

  • Training your own model. Almost never the right first move. Off-the-shelf models with good prompting and your own documents will outperform a from-scratch model you can’t afford to maintain.
  • Tools looking for a problem. If you can’t name the task, the hours it costs today, and who owns it, you’re not ready to automate it.
  • “AI strategy” as a deliverable. Strategy that doesn’t ship a working tool in a few weeks is procrastination with a budget line.

The part everyone underestimates: adoption

The technology is the easy half. The hard half is getting people to actually use it, trust it, and fold it into how they already work. That’s why we pair every implementation with hands-on training for the team that will use it — not a lecture, but a workshop where staff build something with their own data and leave able to do it again.

A tool no one adopts is a cost. A team that’s genuinely AI-native is a compounding advantage. The gap between those two outcomes is rarely the model — it’s whether anyone bothered to bring the people along.

Start small, ship, then widen

Pick one repetitive, text-shaped task. Ship a working tool in weeks, not quarters. Measure the hours it returns. Then use that proof — and the confidence your team gained — to widen the circle. That’s how AI adoption actually sticks in a real business, in Kenya or anywhere else.

If you’d like a grounded second opinion on where AI fits in your operation, that first conversation with us is free.