AI Consultant vs Agency vs In-House Engineer for Kenya

If you are weighing an AI consultant in Kenya against an AI agency or an in-house hire, the choice turns less on job title and more on who will run the system after launch. A consultant is usually the right first step when the problem itself is unclear. An agency suits a defined build with agreed acceptance criteria, and an in-house engineer makes sense when the work will change every month and someone internal must own it.

Three options, defined

An AI consultant is a senior person or small team hired to decide what to build, how to measure it, and which tools or vendors fit. Their output is usually a plan, a pilot, or advice on a build someone else carries out. Some consultants also build, so ask directly what is in scope.

An AI agency is a firm that designs and delivers the system under contract. Some agencies also run the system after launch, for a monthly fee.

An in-house engineer is an employee. They hold the system's context day to day, but they cost money whether or not a project is active, and they can leave.

Hybrid setups are common. HatchWorks, a nearshore software vendor, describes a hybrid model in which a small in-house core works alongside outsourced specialists, in its in-house versus outsourcing guide. That is a vendor's view, but it matches the trade-offs below.

The Nairobi AI consulting market

For a shortlist, Clutch's Kenya machine learning listing lists 13 Kenyan firms when it was fetched. Listed minimum project sizes run from $1,000 to $5,000 and above, and hourly rates range from under $25 to $300 and above. Only four of the 13 display client ratings. Clutch says its rankings weigh verified reviews most heavily and notes that it may earn a fee for some placements. Use the list to build a long list, then check each firm's recent client work and references yourself.

Scope: what each option is built for

Scope is the first split. A consultant fits a problem you cannot yet describe in one sentence: which workflow is worth automating, what data exists, and what the payback range might be. An agency fits a build you can describe precisely, with inputs, outputs, and a test for "done". An in-house engineer fits work that keeps changing after launch.

HatchWorks says outcome-based contracts are best for "self-contained projects with clear acceptance criteria." The same principle applies to consultants: if you cannot write acceptance criteria, pay for discovery first.

A practical test: write down the workflow, the people who use it, the decision it supports, and the cost of a wrong answer. If you can fill those four lines, an agency build is a realistic option. If you cannot, start with a consultant or a short diagnostic.

Speed

HatchWorks, which sells the partner model, gives its own estimates. It says building equivalent in-house AI capability "typically takes four to six months," while renting it from a partner can reach capability in "2–3 weeks." These are vendor estimates, not independent data.

A Kenyan AI consultancy website, aiconsultant.co.ke, publishes indicative timelines for common builds in its AI automation cost guide: an AI opportunity audit at 1 to 3 weeks, a WhatsApp AI agent at 3 to 8 weeks, and a custom AI system at 2 to 6 months.

Pawanax's published package durations vary with scope: the Discovery Diagnostic runs two weeks, the Clinical AI Pilot six to eight weeks, the WhatsApp Agent Factory five to seven weeks, and the Enterprise Agent OS 16 to 24 weeks. Those durations describe scope and planning, not a promise about any one client's timeline. Timelines depend heavily on data access, approvals, and how quickly your staff can review outputs.

Knowledge retention

This is where the in-house option is strongest and where outsourcing most often goes wrong. HatchWorks argues that in-house teams build "a repository of business context that no contract can replicate," and that business context is "effectively impossible to rent." It also models in-house turnover at about 15 per cent a year, which is its own assumption.

The mitigations that matter, whichever option you pick, are practical:

• The code, prompts, configuration, and test cases live in a repository your company controls.
• Every decision that changed the system is written down where the next person will look.
• At least one named person inside your company reviews outputs and understands why the system behaves as it does.
• Handover sessions are scheduled and recorded as deliverables, not left to goodwill.

HatchWorks summarises the principle as "Project context lives in the repo, not in someone's head." That is a good test to apply to any proposal you receive.

Cost structure

Each option has a different cost shape, and comparing headline day rates misses most of it.

For in-house hiring, HatchWorks says recruitment costs run 15 to 25 per cent of first-year salary and that a senior engineer in the United States "exceeds $180,000 per year" before recruitment costs. Those are United States figures from a vendor. Get local Kenyan salary offers before comparing.

For project pricing, the aiconsultant.co.ke guide gives ranges it describes as what "competent delivery in the Kenyan market" costs. Examples include an AI opportunity audit at KES 75,000 to 400,000, a WhatsApp AI agent at KES 250,000 to 1,200,000, and a custom AI system at KES 600,000 to 4,000,000 or more. The same page says maintenance should be budgeted at "roughly 15 to 25 per cent of build cost per year," and that monthly running costs "commonly land in the low tens of thousands of shillings."

Pawanax's published fixed fees are:

• Discovery Diagnostic: $28,000 (KES 3,612,000), two weeks.
• Clinical AI Pilot: $65,000 (KES 8,385,000), six to eight weeks.
• WhatsApp Agent Factory: $48,000 (KES 6,192,000), five to seven weeks.
• Enterprise Agent OS: $185,000 (KES 23,865,000), 16 to 24 weeks.
• Sovereign AI Program: from $420,000 (KES 54,180,000), a custom program.

Retainers are $8,500, $18,000, and $35,000 per month, and the project minimum is $15,000. The published delivery engineer day rate is $700.

Illustrative assumption: a 60-day delivery engineer engagement at $700 per day is $42,000 before tools, run costs, and your own staff time. The number is a worked example for comparison, not a quote.

Risk and exit terms

The main risks differ by option. A consultant can give advice that no one implements. An agency can deliver a system your team cannot operate. An in-house hire can leave, taking the context with them. HatchWorks calls vendor selection the "primary risk" of renting capability, which is a fair starting point for any outsourced work.

Exit terms are where most buyers are weakest. The HatchWorks guide does not cover termination, IP ownership, or service levels, so the list below reflects common contracting practice rather than a single published standard. Ask for written terms on:

• Who owns the code, prompts, configuration, and any trained or fine-tuned artefacts.
• What documentation is delivered and when.
• A handover period with named people and a defined set of sessions.
• Access to repositories and accounts that your company controls from day one.
• What happens to your data at the end of the engagement, including return or deletion.
• Which sub-contractors or third-party tools touch your data.
• Response times for critical incidents, if you need them.

If the system will handle personal data, ask a lawyer to confirm the data-processing terms required under Kenya's Data Protection Act 2019 before signing.

Pawanax's published Operating Partner retainer lists an SLA on critical incidents and budget ownership support. Whether any given engagement includes these terms depends on the signed contract. Evidence class for this offer: Proposed (published terms, not verified delivery results).

When Pawanax is not the right choice

Pawanax is not the right choice in several common situations:

• The work is a small, one-off task below the $15,000 project minimum.
• You want a tool you can configure yourself, and an off-the-shelf product covers the workflow.
• You already have an in-house team that can build and run the system, and you want to keep the whole stack internal.
• Your data cannot leave your premises under any contract you would accept.
• No one in your company can review outputs or own the system after launch.
• You want a build without any discovery. Pawanax's entry point is a paid diagnostic, so a company that already knows its use case may find that step unnecessary.

Evidence class for these judgements: Proposed. They reflect how the published packages are structured, not measured customer outcomes.

Decision table by company size and workflow criticality

Criticality here means the cost of a wrong or delayed output: lost revenue, harm to customers or patients, or regulatory exposure. The table is a decision aid, not a benchmark.

Whatever the row, the same rule applies: if no one inside your company can explain how the system reaches its outputs, the decision is not finished.

Frequently asked questions

Is an AI consultant cheaper than an AI agency in Kenya?

Not necessarily. A consultant is often cheaper for a diagnostic, because the scope is narrower. A build by an agency or a consultant who also delivers can cost more, and the total depends on the workflow, the data, and the run costs. Compare quotes on the same scope, with the same acceptance criteria and the same exit terms.

Can a small company hire an AI engineer in Kenya without a full team?

Yes, but the risk is concentration. One engineer holds the context, so documentation, repository access, and a backup plan matter more. Many small companies start with a consultant or agency for the build and hire an internal owner once the workflow is stable.

What should an exit clause cover?

At minimum: ownership of code and data, documentation and handover deliverables, access to repositories and accounts, data return or deletion, the names of sub-contractors, and the notice period. Have counsel review the wording, especially where personal data is involved.

Does Pawanax offer a discovery before a build?

Yes. The published Discovery Diagnostic is a fixed-fee, two-week engagement at $28,000 (KES 3,612,000). Its listed deliverables are a diagnostic brief, a use-case scorecard, and an executive readout. Treat it as one option among several, and compare it with a consultant or internal review before you commit.

Next step

If you are not yet sure which workflow to automate, or whether an agency build is ready to scope, compare your options on the Pawanax consultancy page. Bring the workflow, the people who use it, the decision it supports, and the cost of a wrong answer. That is enough to tell whether you need a consultant, an agency, an in-house hire, or a mix of them.

Sources

• Clutch: Top Machine Learning Companies in Kenya
• AI Consultant Kenya: How Much Does AI Automation Cost in Kenya?
• HatchWorks: In-House Software Development vs Outsourcing

Continue exploring

Building Production AI Agents: Architecture and Oversight. WhatsApp AI Agents in Kenya: Architecture, Costs and M-Pesa.

Cover: Photo: Stephenwanjau, via Wikimedia Commons. Licensed CC BY-SA 3.0 (https://creativecommons.org/licenses/by-sa/3.0). Illustrative image; not a photograph of a Pawanax system or client. Credit: Stephenwanjau (Wikimedia Commons user; uploaded as own work). CC BY-SA 3.0. Original source.