What Should a Kenyan SME Budget for AI?
"So how much is this AI thing going to cost us?" Wanjiru asked, pen hovering over the January budget like a hawk over a chicken. Otieno, who had suggested AI in the first place, did what all of us do in that situation. He said "It depends" and then went to make chai, which is how most conversations about an AI budget for SMEs in Kenya begin.
He was not wrong, though. It does depend. But "it depends" is not a budget line, and KRA does not accept vibes. So let us break the question into pieces small enough to put a number next to.
Start with the problem, not the tool
The most expensive AI budget is the one that starts with "we should be using AI" and ends with five subscriptions nobody opens after February. Before you spend a shilling, write down two or three specific jobs that eat your team's time. Good candidates look like this:
Replying to the same twenty customer questions every day
Turning messy meeting notes into something readable
Drafting quotations, proposals and follow-up emails
Summarising long documents nobody has time to read
Cleaning up spreadsheets that three people have edited in three styles
If you cannot name the job, you are not ready to budget for the tool. That is fine. Spend January watching where the hours actually go instead.
The three buckets of an AI budget for SMEs in Kenya
Almost every AI cost for a small business falls into one of three buckets. Thinking in buckets stops you from forgetting the ones that are not on the invoice.
Bucket one: subscriptions. The general-purpose AI assistants, and the AI features now bolted onto email, office suites and accounting tools, are usually priced per person, per month, and usually in dollars. That last part matters. Your card, your bank's foreign transaction charges and the exchange rate all get a vote. Budget per seat, and only for the people who will genuinely use it.
Bucket two: time and training. This is the bucket everyone forgets. Someone has to learn the tool, write the instructions, test the outputs and show colleagues how to use it. That is real salary going into the project, even if no money leaves the account. Pencil in a few hours a week for a champion during the first couple of months.
Bucket three: integration and custom work. If you want AI inside your own systems, for example answering customer questions using your actual price list, or reading invoices straight into your accounting, you are paying for development. This is where costs can grow, so it is the bucket to approach last, once the cheap experiments have proved their worth.
The hidden costs nobody mentions at the launch event
AI tools look free at the demo. Then real life arrives. A few costs to keep in mind:
Data clean-up: AI working from a messy customer list produces messy results, just faster
Internet: if the whole team is suddenly uploading documents, your bundle will notice
Usage-based billing: some tools charge by how much you use them, which is lovely until someone loops a big job overnight
Review time: every AI draft still needs a human to read it before it goes out
None of these are reasons to avoid AI. They are reasons to stop the finance person from being surprised in March.
Run a small, cheap pilot first
Do not budget for the whole company on day one. Pick one team, one job and a fixed period, say six to eight weeks. Give two or three people a paid subscription, agree what success looks like before you start, and keep a simple log of time saved and mistakes caught.
Success should be boring and measurable. "Kiprop now drafts quotations in half the time" is a result. "Everyone agrees it is very exciting" is a WhatsApp group, not a result.
At the end of the pilot you will know three things you cannot know today: whether people actually use it, what it really costs once usage settles, and whether the output is good enough for your customers. Then you can budget for the rest of the year with real numbers instead of hopes.
A simple rule of thumb
If you want one sentence to take into the budget meeting, try this: spend a little on subscriptions now, protect real time for learning, and hold custom development until a pilot proves the value. Review the spend every quarter, cancel what nobody opens, and double down on what works.
And if your pilot shows that a custom tool would genuinely pay for itself, that is the point where it makes sense to look at custom software development services to build it properly. Until then, the best AI investment many SMEs can make is a quiet afternoon, a cup of chai and an honest list of what is wasting everyone's time.
Your next step: block out one hour this week, list three jobs that waste your team's time and pick one for a small pilot. When you are ready to cost the custom bits, talk to Philsoft and we will help you size it honestly, with no whiteboard question marks.