AI questions Kenyan businesses actually ask

An AI consultant in Kenya helps an organisation identify where artificial intelligence and automation will genuinely save money or win revenue, then designs, builds and deploys those systems into the tools the business already uses. Practically, the work covers four things: auditing how the business currently operates; choosing the small number of processes worth automating first; building them and integrating with local systems such as M-Pesa, WhatsApp Business and the accounting package; and training staff to run them safely and lawfully.

You are ready if three things are true: there is a repetitive process costing you real hours every week; the data that process needs is recorded somewhere, even imperfectly; and one person internally will own the result after it goes live. You are not ready if the honest answer to any of those is no — and in that case the right first step is usually a short data or process piece, not an AI build.

Two options give the fastest return. Either a half-day team training session plus a written AI acceptable-use policy — cheap, immediate, and it stops confidential data leaking into public chatbots — or automating whichever single process is most obviously bleeding hours, most often payment reconciliation or answering the same WhatsApp questions all day. Do not start with a twelve-month transformation programme.

No. A focused training day starts around KES 120,000 and a first automation around KES 250,000. What genuinely constrains most projects is not budget but internal time and access — if nobody can free two to four hours a week during the build, the project will stall regardless of how much you spend.

An AI opportunity audit typically starts around KES 75,000. A first working automation or AI agent typically starts around KES 250,000, rising with the number of systems it must integrate with. A team training day starts around KES 120,000. Ongoing retainers start around KES 150,000 per month. On top of the build there are running costs — model usage, WhatsApp conversation fees and hosting — which are usually modest but should be modelled at your actual volumes. See our pricing page for the full breakdown.

For the automations we recommend first — reconciliation, document capture, customer response — payback is usually between three and twelve months. If our own modelling says payback is longer than about eighteen months, we will tell you not to do it, because the technology and your business will both have changed by then.

Three that quotes routinely omit. Running costs, which continue forever: model usage, WhatsApp Business Platform conversation fees, hosting. Maintenance, at roughly 15 to 25 per cent of build cost per year, because model providers deprecate versions and your business changes. And your own team’s time during the build, which is the cost most likely to be underestimated and most likely to sink the project.

Sometimes, and we will say so when it is true. A capable operations analyst can eliminate a surprising amount of waste without any AI at all. The case for automation is strongest when the work is high volume, low judgement, and needs to happen at times when nobody is at a desk — nights, weekends, month end. If the work is genuinely judgement-heavy, hire the person.

In the projects we run, the work moves rather than disappears. Automation takes the copying, checking and chasing; people move to judgement, relationships and exceptions. We design deliberately for this: anything touching money, contracts or a complaint routes to a person for approval. The honest general point is that AI replaces tasks rather than roles, and organisations that adopt early usually grow into the freed capacity instead of shrinking.

That describes most businesses we meet and it is not a blocker. Modern document AI reads scanned invoices, delivery notes and even handwritten forms well enough to be useful, and converting paper into structured data is frequently the highest-return automation available. What we do first is decide what genuinely needs cleaning versus what can be worked around.

The system is designed on the assumption that it will. Every workflow has a confidence threshold: above it the system acts, below it the item goes to a human with the reason attached. Money-touching actions always require human approval regardless of confidence. Every action is logged with its inputs and outputs so an error can be traced and reversed rather than quietly compounding.

Almost never. We build around QuickBooks, Sage, Odoo, Zoho, Excel, Google Workspace, Microsoft 365 and the custom system a developer built years ago. Connecting systems is usually cheap and fast; replacing them is expensive, slow and politically painful. If something genuinely cannot be integrated we will tell you and price the workaround honestly.

We design for it, because in Kenya it will. Work is queued and retried rather than lost, capture can happen offline and sync later, and critical notifications fall back to SMS or WhatsApp. Systems that only function on a good fibre connection are not systems we are willing to hand over.

Yes, when done properly. The Kenya Data Protection Act, 2019 governs how personal data is collected, processed and stored, and it applies to AI systems exactly as to any other system. That means having a lawful basis for processing, collecting only what you need, keeping it only as long as necessary, controlling access, and being able to explain what happened. Organisations acting as data controllers or processors may also need to register with the Office of the Data Protection Commissioner. We design against these requirements from the first design session rather than retrofitting them.

That depends on choices you get to make explicitly. Business and enterprise tiers of the major model providers generally do not train on customer data by default — but we verify the contractual position for your specific plan in writing rather than assuming it. Where data must not leave Kenya or your premises, we deploy open-weight models on infrastructure you control; that costs more and performs somewhat below the frontier models, and we will show you the trade-off with real numbers.

It can prepare, check, summarise, flag and recommend. We strongly advise against letting it decide, and in regulated sectors it is usually not permissible anyway. Credit, insurance, employment and similar decisions materially affect people’s lives, and a named human should be accountable for them. This is both good practice and consistent with the direction of Kenyan and international regulation.

If your staff are already using AI tools — and they almost certainly are — then yes, and it is urgent. The realistic risk is not runaway automation; it is an employee pasting a customer list, a draft contract or patient information into a consumer chatbot. A workable policy states what may go into which tools, who approves new tools, how output must be checked, and how this relates to your Data Protection Act obligations.

We sign an NDA before the first working session if you want one. We work inside your accounts wherever possible rather than copying data to ours. Where a copy is unavoidable, we agree in writing what is copied, where it lives, who can access it and when it is deleted.

An audit takes one to two weeks. A first automation takes two to six weeks from kickoff to production. Larger programmes run three to six months but are delivered as a sequence of small live releases rather than a single launch. Access and permissions are almost always the bottleneck, not engineering.

Yes. We are Nairobi-based and meet clients there in person, but delivery is remote across Kenya — Mombasa, Kisumu, Nakuru, Eldoret, Thika, Naivasha — and elsewhere in East Africa. We travel when being in the room genuinely matters, which usually means process discovery and staff training.

Entirely. Code, prompts, evaluation sets, documentation and cloud accounts are yours from the first commit, in your repository under your account. You can dismiss us at any point and another team can pick it up. We think being replaceable is the only honest way to earn a renewal.

Yes, and we do it regularly. If a KES 4,000 per month off-the-shelf tool, a corrected spreadsheet or one changed procedure solves your problem, that is what we will recommend, and we will not charge you for the conversation. A short honest engagement is better business for us than a long disappointing one.

No. Fixed-scope projects end on delivery. Retainers run month to month with 30 days’ notice either way and a quarterly review. Because you own the code and accounts, leaving is always practical.