How to hire an AI consultant in Kenya

What an AI consultant actually does

The title covers a wide range of practice, which is precisely why buyers get burned. At one end sits a strategist who produces a document. At the other sits an engineer who ships a system. Both call themselves AI consultants; only one of them changes anything about your Tuesday.

A full-service AI consultant in Kenya should be able to do four things, and you should establish which of them a firm actually does before the second meeting:

1. Audit. Map how your business currently works, measure what each recurring process costs in hours and shillings, and identify where AI would genuinely pay.
2. Decide. Turn that into a sequenced, costed plan — including an explicit list of what not to do, which is usually the most valuable page.
3. Build. Develop and deploy the systems, integrated with what you already run: your accounting package, your M-Pesa till, your WhatsApp line, your CRM.
4. Transfer. Train your staff, write the runbook, and hand over the code and accounts so the thing survives their departure.

Firms that only do the first two produce strategies that never get executed. Firms that only do the third build things nobody asked for. The fourth is the one most often skipped, and skipping it is how organisations end up with a working system that quietly dies eight months later because the only person who understood it has moved on.

The four kinds of firm you will meet

In the Kenyan market you will encounter roughly four species, each with a legitimate use case and a characteristic failure mode.

The global consultancy

Strong at board-level engagement, governance frameworks and change management at scale. Expensive, and delivery is often subcontracted — sometimes to the very local firms you could have hired directly. Right choice if you are a large regulated institution running a multi-year programme with a matching budget. Wrong choice if you want one process automated by March.

The software vendor

Sells a product and configures it. Fast and comparatively cheap when your need is standard — helpdesk, scheduling, transcription, document storage. The failure mode is the honest one: they cannot tell you their product is wrong for you, because that is the only thing they have. Buy from a vendor when you have already decided a product category fits.

The independent consultancy

Small, senior, builds as well as advises. Best fit when the value sits in your own data and processes and you want to end up owning the result. The risk is capacity: a small firm can be genuinely busy, and you should ask directly when they could actually start rather than when they say they can.

The freelancer

Often excellent value for a small, well-defined piece of work. The risk is continuity — one person who may take another contract, fall ill, or relocate. Acceptable for a contained project; risky for anything your operations will depend on. If you go this route, insist on documentation as a deliverable, not a promise.

What it should cost in Kenya

Prices in this market vary more than they should, largely because buyers cannot compare like with like. These are defensible benchmarks for competent work delivered in Kenya, in Kenyan shillings.

Two warnings about these numbers. A quote dramatically below the range usually means the firm has misunderstood the scope, and you will meet the difference later as change requests. A quote dramatically above it should come with a specific justification — regulatory audit requirements, an unusual number of integrations, on-premise deployment — and if the justification is vague, that is your answer.

Then there are the costs that are not in anyone's proposal. Model usage charges, which continue forever and scale with volume. WhatsApp Business Platform conversation fees charged by Meta plus a platform fee from whichever Business Solution Provider you use. Hosting. Maintenance, realistically 15 to 25 per cent of build cost per year. And your own team's time during the build, which is the cost most often underestimated and the one most likely to sink the project. We cover all of these in detail in our guide to AI automation costs in Kenya.

Eight questions that expose a weak firm

Ask these in the first meeting. You are not testing technical knowledge — you are testing honesty and operating experience.

1. “What have you built that is still running today, and who maintains it?”

Anyone can show a demo. Far fewer can point to a system that has survived two years, a model deprecation and a staff change. The second half of the question matters most: if the honest answer is “we do, and the client could not,” you have found a firm that builds dependency.

2. “What did you last talk a client out of?”

A firm that has never advised against a project either has not been asked for anything foolish, or does not say no. Both are bad. The best answer here is specific, slightly awkward, and clearly cost them money.

3. “How would you measure success on our project, before starting?”

The answer should involve numbers you already track or could track within a week — hours, error rates, response times, cost per transaction. If the answer is about “transformation”, “innovation” or “capability building”, you will have no way to tell whether you got what you paid for.

4. “What could go wrong, and what happens then?”

Every honest AI practitioner has a list: the data was worse than expected, the source system had no API, the model got a category of case wrong, the users refused to adopt it. What matters is whether the firm has thought about detection and rollback, or whether the question is treated as impolite.

5. “Who owns the code, the prompts and the accounts?”

The right answer is “you do, from the first commit, in your repository under your cloud account.” Anything else — a proprietary platform, a hosted environment you cannot access, credentials held by the vendor — is a commercial position dressed as a technical one.

6. “How will this handle Kenyan realities?”

Name them specifically and see what happens: M-Pesa partial payments with wrong reference numbers, customers who write in mixed English and Kiswahili, invoices photographed at a loading bay, a two-hour power cut, KRA eTIMS. A firm that has actually delivered here will have opinions immediately.

7. “How do you handle our data, and what does the Data Protection Act require of us here?”

They should be able to explain lawful basis, data minimisation, retention and cross-border transfer without looking it up, and should raise the question of whether you need to be registered with the Office of the Data Protection Commissioner. If compliance is treated as your problem alone, expect to inherit a system that creates one.

8. “What do you need from us, and what happens if we cannot provide it?”

A firm that has delivered real projects will answer this instantly: access, a decision-maker, and a few hours a week of your people's time. A firm that says “nothing, we will handle everything” has either not done this before or is telling you what you want to hear.

Warning signs

• A proposal with no numbers in it. If nothing is measured before, nothing can be proved after.
• Guaranteed percentage savings before any discovery. Nobody can know that in week one. It is a sales number.
• “Our proprietary AI platform.” Sometimes genuine, usually a wrapper around the same model APIs everyone else uses, sold as lock-in.
• Unwillingness to name the underlying models or tools. Discretion about client names is professional; secrecy about your own stack is not.
• No mention of the Data Protection Act. In Kenya, in 2026, this is either ignorance or avoidance.
• Only strategy, or only code. Strategy without delivery produces shelfware; delivery without strategy produces expensive solutions to unimportant problems.
• Pressure to sign before you have spoken to anyone else. Good firms expect to be compared.

What to insist on in the contract

You do not need an elaborate agreement, but a handful of clauses are worth defending.

• Scope with acceptance criteria. Written so that a reasonable third party could judge whether it was met.
• Fixed price or a capped estimate. Open-ended time and materials on a first project transfers all the risk to you.
• IP assignment. All code, prompts, configuration, evaluation sets and documentation assigned to you on payment.
• Accounts and credentials. Cloud, model provider, WhatsApp and repository accounts in your name from the start — not migrated “at the end”.
• A data processing agreement. If personal data is involved, this is not optional under the Data Protection Act, 2019.
• Support period. A defined hypercare window after go-live, included rather than quoted separately once you are dependent.
• Exit and handover. An obligation to deliver documentation and a knowledge-transfer session on termination, however the relationship ends.

When not to hire anyone

Sometimes the right answer is nobody, and a decent consultant will say so. Consider doing it yourself if your first project is small, internal and low risk; if you already employ a developer with capacity and interest; or if the honest problem is that nobody has yet written down how the process works. That last one is a management task, not a technology purchase.

Bring in outside help when the work touches money, personal data or customers; when it must integrate with several systems; when the project has already been attempted and stalled; or when you need the result to outlive the enthusiasm of the person driving it. Some of our best engagements sit alongside an internal developer rather than replacing one — we bring pattern recognition and they bring institutional knowledge, and the combination usually beats either alone.

A one-page checklist

Print this. Take it into the meetings.

• Can they point to a system still running, maintained by the client?
• Can they name something they advised a client not to do?
• Do they propose measuring a baseline before building?
• Is the first project small enough to fail cheaply?
• Do you own the code, prompts and accounts from day one?
• Do they raise the Data Protection Act without being prompted?
• Do they have specific opinions about M-Pesa, WhatsApp and Kenyan data realities?
• Is there a human review step wherever money moves?
• Is the price fixed, and is scope creep their risk?
• Is there a written handover obligation if it ends badly?

If a firm scores well on eight of those ten, you are in reasonable hands. If they score below five, keep looking — there are enough capable people in this market that you do not need to compromise.