AI Consulting in Kenya: What an Enterprise AI Strategy Actually Costs

Every few weeks a finance lead, an operations director, or a founder sends the same brief to three or four firms and waits for three or four wildly different quotes. One comes back at a number that fits a quarterly line item. Another is ten times that. Neither quote explains what actually changed between them. The gap rarely comes down to greed or generosity. It comes down to scope: what is being assessed, what is being built, and who carries the work after the engagement ends.

This piece is about the cost drivers behind an enterprise AI strategy in Kenya, the shapes those engagements take, and the questions worth asking before anyone signs anything.

What a strategy engagement actually includes

An enterprise AI strategy is not a slide deck about the future. A credible engagement runs through four stages, and each one carries a different weight.

Use-case triage comes first. The consultancy sits with department heads, maps where decisions are made, where documents pile up, where forecasts go wrong, and filters those candidates against feasibility and business value. Most companies arrive with a list of eight to twelve ideas and leave with two or three worth pursuing. The value of the stage is the elimination, not the ideation.

Data audit follows. Someone has to look at the systems where records live, check whether the data is clean enough to train anything on, and document what exists alongside what does not. For many Kenyan enterprises, this is the stage that rewrites the plan. Customer records split across an old ERP and a spreadsheet, or years of transactions with no consistent naming, will change what is possible in year one.

Roadmap turns the surviving use cases into a sequence: what gets piloted, what gets bought off the shelf, what waits, and what infrastructure has to exist first. Then a pilot proves or disproves one use case under real conditions with real data.

A quote that covers only the first and third stages will always look cheaper than one covering all four. The cheaper quote is not the same work.

What drives the cost

The main variables are knowable in advance.

• Discovery depth. A two-week assessment across one department costs a fraction of an eight-week review spanning finance, operations, and customer service. Deeper discovery surfaces more, but it also consumes more senior consultant time.


• Data readiness. This is the single largest swing factor. If your data is already centralised, labelled, and accessible, the work is analysis. If it is scattered across systems that were never designed to talk to each other, a share of the budget goes to plumbing before any model is trained.


• Build versus buy. Adapting an existing model to your workflow is cheaper than building custom systems from scratch. Plenty of problems do not need custom work at all, and a good consultant will say so even when it shrinks the invoice.


• Integration complexity. An AI system that cannot write back to your accounting software, CRM, or stock system is a demo. Getting it to work inside existing systems, with permissions and audit trails intact, is engineering work, and it is priced accordingly.


• Team training. Models that nobody in the organisation can operate, correct, or question become expensive shelfware. Budget for handover, documentation, and time with your staff.


• Ongoing maintenance. Data drifts, business rules change, models degrade. Someone has to monitor and retrain. If that is not scoped into the engagement, it becomes a surprise in month six.

The shape of the pricing

Consulting fees in East Africa usually arrive in one of three shapes, and the shape matters more than the headline figure.

Phased engagements break the work into stages with a decision point at the end of each. You pay for discovery first, review the findings, then decide whether to fund the pilot. This costs more per hour than buying everything upfront, but it caps downside. If the pilot shows a use case is not viable, you stop.

Fixed-scope projects price a defined deliverable: one chatbot, one forecasting system, one document-processing pipeline. Fixed scope works when the inputs are stable and the requirements are clear. It breaks down when the data turns out to be messier than the brief suggested, and the variation clauses decide who absorbs that.

Retainer models pay a monthly fee for ongoing access to a team. This suits organisations with a roadmap already in motion who need capacity rather than direction. It is the wrong shape for a first engagement, because you end up paying for time you cannot yet direct.

A rough order of magnitude: a focused single-use-case engagement sits at one end of the scale, and a multi-department enterprise programme with custom integration sits several multiples above it. Any firm quoting the low end for the high end's scope has either misunderstood the brief or is planning to renegotiate midway. Ask which.

Timelines, honestly

Assume that the first conversation and a signed engagement are three to six weeks apart in a large organisation, mostly procurement and legal rather than anyone's hesitation. Discovery runs two to eight weeks depending on breadth. A pilot, realistically, is one to three months after that, and that assumes the data audit went well. Enterprise-wide rollout after a successful pilot is measured in quarters, not weeks.

Organisations that compress these timelines usually discover why the phases existed. The pattern is predictable: a pilot ships fast, works on demo data, then stalls for two months while someone untangles the connection to the production database.

How to evaluate consultants

Three tests separate credible firms from polished ones.

Ask for working demos, not case-study slides. A pilot running on someone else's data tells you far more about engineering quality than a deck does. Insist on seeing the thing operate.

Local presence matters more than it sounds. An AI strategy has to survive Kenyan realities: connectivity, compliance with the Data Protection Act, integration with the systems actually deployed here, and meetings that happen in person when something breaks. A firm that can sit in the room with your operations team holds an advantage offshore-only providers struggle to match.

Finally, ask about post-handover support. What happens three months after go-live, when accuracy drops or the workflow changes? Get the answer in writing, with response times. GashoTech, a Nairobi-based AI consulting company, structures engagements around that handover instead of treating it as an afterthought, and the same principle applies to whoever you hire.

For organisations that want automation rather than a full strategy, GashoTech's AI automation services cover the narrower problem of connecting workflows without the longer assessment.

When consulting is the wrong purchase

Consulting is the wrong purchase more often than firms admit. If you cannot name the business problem in one sentence, the problem is upstream of any consultant. Not "we need AI", but "our claims processing takes eleven days and that costs us customers." Write the problem first.

If no one senior owns the outcome, wait. AI projects that sit with a curious middle manager and no executive sponsor stall at the first internal disagreement about data access.

If your data is not accessible even to your own team, fix that first. A modern data stack and clear access policies cost less than an engagement that will spend half its budget discovering the same thing.

And if a standard tool solves it, buy the tool. Chatbots, transcription, and document extraction are commodities now. A useful test: if the use case is one a competitor could buy off the shelf next Tuesday, it is not a strategy question.

FAQ

Do we need a data team first?

No, but someone must own the data. A consultancy can run the audit and set up the pipelines, yet someone in your organisation has to make decisions about access, quality, and who fixes problems when they appear. Hiring a full data team before the strategy exists is usually premature.

How long before we see results?

A well-scoped pilot can show measurable signal within a few months of kickoff. Meaningful, sustained impact across a department typically takes two to three quarters, most of which goes to integration and adoption rather than modelling.

Should we buy a platform instead of hiring consultants?

When the problem is standard, yes. Consulting earns its fee when the problem is specific to your operations, spans multiple systems, or requires decisions about sequencing and build-versus-buy that a vendor has no incentive to give you honestly.

What should we prepare before the first meeting?

Three things: one clear business problem with a rough cost attached, a list of the systems where the relevant data lives, and a named person who can grant access to them. That preparation can shorten discovery noticeably.