On this page
TL;DR:
- A Kenyan company with fewer than 100 staff can adopt AI in about six months by passing six named checkpoints: the Honest Picture, the House Rules, the Proof Task, the Team Habit, the Money Decision and the Six-Month Review. The timing is our suggestion.
- Choose the first task with care. In a field experiment with 640 Kenyan entrepreneurs, AI advice showed no gain on average. Weak performers did worse, and strong performers may have gained.
- Give the work to one named person. A company this size rarely needs a committee.
- Data protection law still applies if you are small enough to skip registration, and many sectors must register however small they are.
- Buy paid seats only after a task has proved itself on free or already-included tools.
Who this guide is for
This guide is for owners, general managers and operations leads in Kenyan companies with roughly 5 to 99 people. You probably have no IT department and no spare budget, and the person leading this also has a day job.
Kenya's Revised MSME Policy, dated April 2026, sets these bands:
| Category | Annual turnover | People employed |
|---|---|---|
| Micro | Up to KES 500,000 | Up to 10 |
| Small | KES 500,000 to 5 million | 11 to 50 |
| Medium | KES 5 million to 100 million | 51 to 250 |
We use "SME" loosely here to mean any company below 100 staff. Above that size you will need a more formal plan, with a working group and training in waves.
What does research on Kenyan small businesses tell us?
The strongest evidence we found comes from Kenya itself, and it suggests that AI can help some firms and harm others, depending on which of its suggestions they act on.
A research team including academics from Harvard Business School and UC Berkeley ran a five-month experiment in 2023 with 640 Kenyan entrepreneurs, including fast food sellers, poultry farmers and cybershop owners. Some were given a GPT-4 business mentor on WhatsApp. Others received a standard business guide. The paper was later published in the journal Management Science, and its results shifted between versions, so we use the published figures.
| Finding | Detail |
|---|---|
| Average effect | No measurable gain across the whole group |
| Strong performers before the trial | May have gained over 15%. The authors call this evidence suggestive. Earlier versions said just over 15% and just over 20% |
| Weak performers before the trial | Did nearly 10% worse. Earlier versions said about 8% and roughly 10% |
| Why | The published paper found no evidence that the gap came from the questions owners asked or the advice the AI gave. Strong performers found changes tailored to their business. Weak performers more often adopted generic moves, such as lowering prices and advertising more. An earlier version and the Berkeley Haas report said weak performers took harder problems to the AI, so that explanation is less settled |
The study used an older model and one tool, so treat it as a warning. It still shapes three rules in this roadmap:
- Start with work you can check quickly. This is our advice and goes beyond what the study tested. Drafting a quotation is a good fit. A cash crisis is a poor one, because you cannot check the advice until the money is spent.
- Keep a person's judgement over the advice. The tool can suggest ten things, and someone must know which two make sense. Be wary of generic suggestions.
- Train alongside the tool. One of the researchers suggested that owners also need complementary skills and resources.
Your staff may be ahead of you too. The Star reported on 28 April 2026, citing a Digital 2026 report, that 97.5% of Kenyan internet users aged 16 and over use AI tools each month. That counts internet users and says nothing about work, but it is a good reason to ask before you assume.
The six checkpoints at a glance
A checkpoint is a point where you stop and answer one question with evidence. If you cannot answer yes, stay there or change course. No study proves this timing is right, so stretch it if you need to.
| Checkpoint | By when | The question | You pass when |
|---|---|---|---|
| 1. The Honest Picture | Week 2 | What is really happening, and which tasks could AI help with? | You have a list of tools in use and three candidate tasks |
| 2. The House Rules | Week 3 | Does everyone know what is allowed? | One page is shared and an owner is named |
| 3. The Proof Task | Week 8 | Does AI improve one real task once checking is counted? | Your before and after numbers show a real saving |
| 4. The Team Habit | Week 12 | Do people keep using it without being chased? | A second task is running and the method is written down |
| 5. The Money Decision | Month 4 | Should we spend anything? | You have decided to buy, hold or stop, with reasons |
| 6. The Six-Month Review | Month 6 | Keep, widen or end? | The owner reports results and sets the next review date |
Checkpoint 1: The Honest Picture (by week 2)
The aim is to find out what is already happening and which tasks are worth testing.
Ask what people already use. Do this in a staff meeting or a short anonymous form. Promise that nobody will be punished for an honest answer, and keep that promise. Ask which tools they use, for what, and what kind of information goes in.
Keep a one-week task diary. Ask each team to note repeated tasks and how long they take. A notebook is fine. FuKazee's free AI readiness assessment, which takes about three minutes, can give you a rough starting score. It is a self-check, and we do not claim it is a validated research tool.
Filter your list with five yes or no questions:
- Do we do this at least every week?
- Can someone check the result in a few minutes?
- If it is wrong, is the worst outcome embarrassing and fixable?
- Can it be done without pasting in customer or staff details?
- Is it a drafting, tidying or explaining job, and not a hard business problem?
A task needs five yes answers to go on the shortlist. Question 5 is our own caution after reading the Kenyan experiment.
| Team | Good first tasks | Leave for later, or keep away from AI |
|---|---|---|
| Sales | Quotation cover notes, follow-up messages | Deciding discounts for named customers |
| Customer care | Standard WhatsApp reply templates | Pasting in real complaints with names and numbers |
| Marketing | Post ideas, first drafts of adverts | Claims about results you cannot prove |
| Admin and HR | Job adverts, first drafts of procedures | Screening CVs or ranking staff |
| Finance | Explaining Excel formulas on sample data | Tax positions, payroll, M-PESA statements with names |
| Owner | Testing a plan by asking for objections | Deciding what to do in a cash crisis |
Our guide to free AI tools for small businesses in Kenya shows what the free plans can do on a phone.
Checkpoint 2: The House Rules (by week 3)
The aim is one page that everybody has read, and one person who owns it.
Pick the owner now. This person chooses tasks, keeps the numbers and reports back. We suggest allowing about two hours a week, and giving them the authority to say no. A careful operations lead or office manager often suits the job better than the most excited AI fan.
Adapt this template:
How we use AI here 1. Owner: [name]. Ask this person when unsure. 2. Tools we use: [list], on work accounts where we have them. 3. Never paste in: names with phone numbers, ID numbers, KRA PINs, M-PESA statements, payroll, staff records, health details or client contracts. 4. A person reads and checks everything before it leaves the company. 5. AI does not decide who we hire, pay, discipline or give credit to. 6. If something goes wrong, tell [name] the same day.
Our guide to an AI usage policy for Kenyan organisations goes further if you need a longer version.
Think twice before banning AI outright. In the University of Melbourne and KPMG global study (data from late 2024 to early 2025), 67% of employees whose organisation had banned generative AI reported behaviour such as uploading sensitive company information or copyrighted material to public AI tools. The figure was 56% where a policy guided use, and 33% where there was no policy. Kenya was not in that survey, and it shows a link only. The report says policies alone do not guarantee compliance, so back your page of rules with clear guidance and training.
Do you need to register with the data commissioner?
Check this now, because size does not exempt everyone. Under regulation 13 of the 2021 registration regulations, you skip mandatory registration only if turnover is below KES 5 million and you have fewer than ten employees.
Even then, the Third Schedule requires registration for certain purposes. Those include:
- Running a school or other educational institution
- Health care
- Hospitality, excluding tour guides
- Property management and selling land
- Financial services
- Transport services, including ride-hailing
- Businesses wholly or mainly in direct marketing
- Security CCTV
The exemption only removes the duty to register. Regulation 13(3) says exempt firms must still comply with Parts IV and VI of the Act. Part VI covers sending personal data outside Kenya, which matters because many popular AI tools are run by companies abroad.
Under section 43 of the Data Protection Act, a breach that poses a real risk of harm must be reported to the regulator within 72 hours of becoming aware of it. Under section 63, a penalty notice can reach KES 5 million or, for a business, 1% of the previous year's turnover, whichever is lower.
The ODPC's guidelines page also lists a guidance note on processing by micro, small and medium enterprises. Digital Policy Alert, a policy tracker, records it as adopted on 6 November 2025. We could not confirm its formal status from the ODPC itself, so ask the regulator before relying on it.
If a use looks like the list in regulation 49 (automated decisions with significant effects, sensitive data, large-scale processing and others), an impact assessment may be needed before you start. Under regulation 52, if you hear nothing within sixty days of submitting it, you may begin. For most small companies, the simple answer is to keep those uses off the roadmap for now and take legal advice if you truly need them. Our guide to the Data Protection Act and AI explains more.
Which tool should you use?
Start with what comes with the email and documents system you already pay for. As of 10 November 2026, this is what each provider's own page says. We do not promote one tool, and all of this changes often.
| Tool | Included or free | Data settings | Paid business option |
|---|---|---|---|
| Microsoft 365 | Copilot Chat is included at no additional cost with eligible plans | Microsoft says prompts and responses are not used to train its foundation models under enterprise data protection | Copilot Business add-on: US$21 per user per month paid yearly, with a US$18 first-year offer for eligible existing customers until 31 December 2026 |
| Google Workspace | Business Starter includes Gemini chat and Gemini in Gmail. Standard adds Docs, Sheets, Slides, Meet and more | Google says chats on core-service editions, such as Business Starter, will not be reviewed by humans. Additional-service editions differ | Priced by location. Google's admin help page, updated 26 October 2026, lists US$7 for Starter and US$14 for Standard per user per month on an annual plan |
| ChatGPT | A free plan exists | The comparison table lists "opt-out available" for training on Free, Go, Plus and Pro. Find the setting and use it | Business starts at 2 users. The copy of the price page we could open showed no figure, so check it yourself |
| Claude | A free plan exists | The comparison table lists "opt-out" for individual plans. Team has no model training on your content by default | Team standard seat: US$20 per seat per month billed annually, or US$25 monthly, for 2 to 150 people |
These are not Kenyan quotes, and tax may apply.
Checkpoint 3: The Proof Task (by week 8)
The aim is to learn whether AI improves one real task once checking time is counted.
Take the best task from your shortlist and give it to two to four people. Teach them the basics first. We use the four Cs: Clarity (say exactly what you want), Context (give the background), Constraints (set limits) and Customisation (shape it for the reader).
Then measure in three steps:
- Before. Time ten examples of the task done the normal way.
- After. Time ten more with AI, from opening the tool to the moment the work is ready to send. Include checking and fixing.
- Errors. Write down every mistake the AI made and who caught it.
Our guide to evaluating ROI on AI training explains how to keep this honest.
You pass if the saving is real after checking, quality held, no customer or staff data went where it should not, and the people doing the work want to continue.
If you fail, change the task once and run it again. If the second attempt also fails, stop and write down why. That is still a useful result, and it costs far less than a year of unused licences.
Checkpoint 4: The Team Habit (by week 12)
The aim is for the work to continue without the owner chasing anybody.
- Add a second task from your shortlist, ideally in another team.
- Keep working prompts in one shared document or folder.
- Write the method down as a short procedure. Our guide to writing SOPs for AI shows how.
- Spend ten minutes at an existing team meeting on what worked and what went wrong. Avoid creating a new meeting.
You pass if people are still using the method without reminders, and a new person could follow your written procedure.
If use collapses as soon as the owner stops asking, find out why. The task may not be worth it, staff may have no time to check properly, or their manager may not value it.
Checkpoint 5: The Money Decision (month 4)
The aim is to decide whether to spend anything, using your own numbers.
| Option | Choose it when |
|---|---|
| Stay on free or included tools | The tasks work and nobody is hitting limits |
| Buy a few paid seats | A named person has a named task, and the free version clearly falls short |
| Pay for training | People are stuck on skills, checking or data handling |
| Hold | The numbers are unclear, so run another month |
| Stop | Nothing has shown a real saving |
A fair test for each seat is this: who will use it, for which task, and what can they not do on the free version? For wider costs, see how to budget for AI training.
A made-up example: 35 staff
This company is invented to show the method and the arithmetic.
A 35-person distribution company in Nairobi has a sales team that sends about 15 quotations a week.
- Checkpoint 1. The diary shows quotation cover notes are repeated, quick to check and need no personal data.
- Checkpoint 2. The operations manager becomes the owner. The rules go on one page.
- Checkpoint 3. Ten cover notes done the old way average 40 minutes. Ten with AI, including checking, average 28 minutes. That saves 12 minutes a quotation.
- The sum. 12 minutes × 15 quotations is 180 minutes, or three hours a week across the team.
- Checkpoint 4. The admin team adds drafting job adverts. The sales method becomes a one-page procedure.
- Checkpoint 5. Four people have hit limits on the free version. Using the US$21 Copilot Business rate from the table as one example, 4 × 21 × 12 is US$1,008 a year. Use the rate of whichever tool you are considering. The owner compares that with three hours a week of staff time and with what is still free, then decides.
Your numbers will differ, so run the same sums with your own.
Checkpoint 6: The Six-Month Review (month 6)
The aim is a clear decision by the owner and the head of the company. Sit down for an hour and go through:
- Which tasks stayed, and what did they save?
- What errors were caught, and did any get out?
- Did staff data or customer data stay protected?
- Are people more confident, or more worried?
- Do the House Rules need changing?
- What should we stop doing?
You pass if you can say keep, widen or end for each task, with a reason and a date for the next review.
Also check what has changed in the law. As of November 2026, Kenya has no AI-specific law in force. The Artificial Intelligence Bill, 2026 is a Bill: The Star reported in October that a Senate committee recommended not proceeding with it, and we found no report of a full Senate vote. As of November 2026, the regulator's AI guidance is a draft whose consultation closed on 17 August 2026. Our explainer on the AI Bill has the detail.
How does the roadmap change with company size?
The checkpoints stay the same, and how you run them changes. These are our suggestions.
| Staff | Who owns it | How to run it | Rough timing |
|---|---|---|---|
| 5 to 19 | The owner or a trusted senior | Combine Checkpoints 1 and 2 in one meeting. One Proof Task for everyone | About 10 weeks to Checkpoint 4 |
| 20 to 49 | One coordinator with some protected time | One Proof Task per department, one at a time | About 12 to 16 weeks to Checkpoint 4 |
| 50 to 99 | A coordinator plus one person from finance and one from HR | Two Proof Tasks in parallel, a slightly longer written policy | The full six months |
What usually goes wrong in smaller companies?
These are the risks we would plan for.
| What goes wrong | Why it happens | What to do |
|---|---|---|
| The owner buys licences after a demo | Enthusiasm came before evidence | Hold all spending until Checkpoint 5 |
| Customer lists end up in a chatbot | Nobody said not to | Share the House Rules before any training |
| The keenest person does everything | The knowledge never spreads | Require a written method at Checkpoint 4 |
| AI is asked to fix a business problem | It is tempting when things are hard | Use the fifth question in Checkpoint 1 |
| The plan dies in a busy month | It was extra work with no owner | Name the owner and avoid peak season for the Proof Task |
| Time savings are claimed with no record | Nobody timed the "before" | Time ten examples first |
| Staff assume jobs are going | Leaders said nothing | Say what you know, and do not promise what you cannot keep |
Three prompts for the person running this
Remove names and personal details before you paste anything.
1. Sort your task diary
I run a Kenyan company with [number] staff. Below is a list of repeated tasks from a one-week diary. Context: we want to choose a first task for AI. Constraints: for each task, answer yes or no to these five questions: [paste the five questions]. Show a table. Keep only my tasks and do not add new ones. Customisation: the reader is a busy owner, so end with the best two and one sentence on why.
Tasks: [paste]
2. Draft your House Rules
Write a one-page set of AI rules for a [sector] company in Kenya with [number] staff. Context: staff use [tools] and handle [kinds of information]. Constraints: under 200 words, simple English, numbered rules only, and include who to ask: [role]. Do not state any law or legal requirement. Customisation: it will be pinned up and shared on WhatsApp, and some readers are not confident in English.
Have someone who understands your legal duties read the result.
3. Check whether AI is the right tool
Here is a problem in my business: [describe without names]. Context: I am considering asking AI for help. Constraints: tell me honestly which parts AI could help with, such as drafting or organising information, and which parts need money, a change in the market or an experienced person. Then list what you would need to know before giving any advice. Do not give a final recommendation. Customisation: short bullet points in plain English.
When should you wait before starting?
Wait if nobody can give this two hours a week. Wait if the company is in a cash crisis or in its busiest season, because the Proof Task needs calm conditions. And be honest if your real problem is a broken process, because AI tends to speed up the work without repairing it.
Some firms should also get advice first, especially those whose main work involves health records, children's data, credit or large customer databases. Starting later is a fair decision if you make it deliberately.
Questions people ask
How long does AI adoption take for a small Kenyan company?
We plan for about six months to reach a firm decision. A team of under 20 can reach a working habit in about ten weeks. Busy seasons will stretch it.
Can a small company do this with only free tools?
Yes, for the first four checkpoints. Free plans have usage limits and different data terms, so keep personal and confidential information out of them. Our guide to free AI tools covers what each one gives you.
Do we need to register with the Office of the Data Protection Commissioner?
Possibly. You are exempt only with turnover below KES 5 million and fewer than ten employees, and sectors such as schools, health, hospitality, property and financial services must register anyway. The exemption leaves your other duties in place, so confirm your position with the regulator or a lawyer.
Do we need a written AI policy?
Yes, though one page is enough to begin. It should name an owner, list approved tools, say what must never be pasted in, and require a person to check every output.
Should we hire an AI specialist?
In our view, most companies under 100 staff do not need one for this kind of everyday use. A curious, careful staff member with protected time and some training can usually take you through the first six months.
What if our staff are worried about their jobs?
Talk about it early. Say what you know, say what you do not know, and make no promise you cannot keep. Starting with tasks people dislike may make the conversation easier.
What is the first step this week?
Ask your team what they already use, and start a one-week task diary. It costs nothing, and it gives you what you need for Checkpoint 1.
Sources
- State Department for MSMEs, Revised Micro, Small and Medium Enterprises Policy 2026 (April 2026), last checked 10 November 2026
- Otis, Clarke, Delecourt, Holtz and Koning, The Uneven Impact of Generative AI on Entrepreneurial Performance: Evidence from a Field Experiment in Kenya, Management Science (2026), open copy, last checked 10 November 2026
- Berkeley Haas, report on the Kenyan generative AI experiment (1 April 2024), last checked 10 November 2026
- Otis, Clarke, Delecourt, Holtz and Koning, The Uneven Impact of Generative AI on Entrepreneurial Performance, working paper summary (December 2023), last checked 10 November 2026
- EconPapers, abstract of another version of the same paper, last checked 10 November 2026
- The Star, "Countries with the highest use of AI tools" (28 April 2026), last checked 10 November 2026
- University of Melbourne and KPMG, Trust, attitudes and use of artificial intelligence: A global study 2025 (full report), last checked 10 November 2026
- The Data Protection (Registration of Data Controllers and Data Processors) Regulations, 2021, ODPC, regulation 13 and Third Schedule, last checked 10 November 2026
- The Data Protection (General) Regulations, 2021, ODPC, regulations 49 and 52, last checked 10 November 2026
- The Data Protection Act, 2019, text hosted by ICNL, sections 43 and 63 and Parts IV and VI, last checked 10 November 2026
- DLA Piper, Data Protection Laws of the World: Kenya, last checked 10 November 2026
- ODPC, Guidelines page, last checked 10 November 2026
- Digital Policy Alert, ODPC adopted guidance note on processing by MSMEs (6 November 2025), last checked 10 November 2026
- Digital Policy Alert, status of the ODPC draft Guidance Note on Artificial Intelligence, last checked 10 November 2026
- The Star, "Senate committee shoots down Nyamu's Bill on AI" (15 October 2026), last checked 10 November 2026
- Microsoft, Microsoft 365 Copilot pricing, last checked 10 November 2026
- Microsoft Learn, Enterprise data protection in Microsoft Copilot and Copilot Chat, last checked 10 November 2026
- Google Workspace, pricing, last checked 10 November 2026
- Google Workspace admin help, AI features and pricing updates for Business editions (updated 26 October 2026), last checked 10 November 2026
- Google, Use Gemini Apps with a work or school account, last checked 10 November 2026
- OpenAI, ChatGPT pricing, last checked 10 November 2026
- Anthropic, Claude pricing, last checked 10 November 2026
Where FuKazee fits in
FuKazee helps Kenyan companies with the people side of this roadmap. We can run a session on the House Rules, teach the team that will do the Proof Task, and help you read your before and after numbers honestly. Since 2022 we have trained more than 3,000 people.
We assume nobody knows AI, we avoid jargon, and we say plainly that nobody becomes an expert in a few days. We show the free version of every feature and never promote one tool, and Microsoft 365 and Copilot are part of what we teach. Your legal and data protection decisions stay with your own advisers.
See how this works on our page for AI training for small businesses in Kenya.
To talk through a roadmap for your company, email us at lead@fukazee.com
Want guided support instead?
Plan AI training around your team's real work. Answer a few quick questions about your sector, team size and goals, and we will recommend the right programme and format. Sessions are built on your workflows and include a usage policy.
Rodney has run FuKazee's training since 2022 and still teaches inside every programme line: corporate sessions, the postgraduate cohort and the KCSE examiner webinars. First Class Honours in Journalism and Mass Communication, University of Nairobi.
