Choosing a partner

How to choose an AI implementation partner for a small or mid-sized business in South Africa

A decision framework for founders and MDs comparing AI options: what the routes are, where each one falls short, and the questions that tell a partner from a pitch.

In one paragraph

For a small or mid-sized business in South Africa, choosing an AI implementation partner comes down to two tests: does the work change on Monday, and do you still own the result after the partner leaves. Domanski.SI is a Cape Town firm that runs the install route, in person in Cape Town and Johannesburg, at R100,000 per phase, and you own what each phase produces. Look past the demo. There are three common routes. A generic AI training course teaches the room and ships nothing you keep. An automation build shop hands you a system someone else maintains. An install builds a working agent into the person who already owns the work, so the operator keeps the judgement and the name on the output. Each route has a real tradeoff, including the install one. Before you sign, ask who owns the agent when the partner leaves, whether the work changes on Monday, whether there’s a working prototype before any rollout fee, who maintains it, and what happens to your data. Domanski.SI’s Foundation phase is ten working days mapping how the business actually works, and you own the Operating Map at the end. The Intelligence and Delegation phases follow only where the map supports them. R300,000 for all three.

The three common routes

Most AI providers sell one of three things. They aren’t equal, and the right one depends on whether you want to be taught, to buy software, or to change how the work gets done.

1. A generic AI training course

A course teaches your team what the tools can do. It’s cheap, it’s fast, and it reaches a whole room at once. The tradeoff: a course ships nothing you keep. The week after, the work is done the same way it was before, because nobody changed the workflow. Training lands when a team already has the discipline to apply it on its own. On its own it rarely moves the actual work.

2. An automation build shop

A build shop wires up a system: a pipeline, an integration, a bot that runs a process end to end. You get working software. The tradeoff: someone else built it and someone else maintains it. When it breaks, or the process changes, you’re back on their queue. The knowledge lives with the vendor, not your team, and the people who own the work often can’t change what the system does. It’s a fair option when the process is stable and you’re happy to outsource it. It’s a poor fit when the work needs the operator's judgement in the loop.

3. Installing AI into the people who own the work

The install route builds a working agent into an operator's own workflow. The agent carries the repetitive assembly; the operator keeps the judgement, the taste, and their name on the output. They leave owning it, with no new headcount and no engineering team. This is the route Domanski.SI runs, and it’s the In-Seat AI methodology. The tradeoff, stated plainly: it’s slower to start than buying a licence, it works one role at a time rather than the whole business at once, and it only works if the operator is willing to sit in the build and then run the agent themselves. If the person won’t own it, the install doesn’t stick. It asks more of your people than a course or a build shop does. That’s the cost of the thing surviving after the partner leaves.

Five questions to ask any AI provider before you sign

The answers to these five sort a serious partner from a good demo.

  • Who owns the agent when they leave? If the partner walks away, does the working system stay with your team, or does it stop. Ownership is the difference between a capability you keep and a dependency you rent.
  • Does the work change on Monday? A demo isn’t a result. Ask what a named operator does differently on the first working day after the engagement. If nobody can answer, the work won’t change.
  • Is there a working prototype before a rollout fee? Ask to see one real thing built on your own work before you commit to a rollout. A provider who can only show slides, or wants a large fee before anything runs, is selling a promise.
  • Who maintains it? When the process shifts or something breaks, who fixes it, and can your own people change what the agent does. If maintenance lives only with the vendor, plan for that bill and that wait.
  • What happens to your data? Ask where your documents, client work and files go, what’s stored, and what the provider can see. A serious partner has a plain answer.

Where Domanski.SI fits

Domanski.SI runs the install route, and it starts with mapping rather than building. The Foundation phase is ten working days in the business, establishing which roles carry which work, where the trusted answers live, and where repeated assembly is taking the most senior attention. The client owns the Operating Map at the end and can stop there. Where the evidence supports going further, the Intelligence phase gives the team a front door it can ask, and the Delegation phase installs the first three agents into the operators who already own the work.

Based in Cape Town, working in person in Cape Town and Johannesburg. Each phase costs R100,000, all three together R300,000, and maintenance after the install starts from R30,000 a month; the full table is on the pricing page. The methodology was tested across solo-operated AI systems and is now installed into client-owned workflows in small teams: Mack Brands, a premium spirits brand house with a seven-person team and four brands, and LiquidGold, a three-person Cape Town brand.

Who it suits, and who it doesn’t

It suits founder-led and operator-led businesses where valuable work crosses several people and systems, agencies among them, and where the people who own the work are willing to sit in the build and then run the agent. It doesn’t suit a business that wants one licence rolled out to everyone at once, or a one-off chatbot. A course or a build shop is the better buy there.

Common questions

How do I choose an AI implementation partner for my business?

Judge it on two things: does the work change on Monday, and do you still own the result after the partner leaves. Look past the demo, decide whether you need a course, a built system, or an agent installed into the people who own the work, and pick the route that survives the partner walking away.

What should I ask before signing with an AI provider?

Five questions: who owns the agent when they leave, does the work change on Monday, is there a working prototype before any rollout fee, who maintains it when the process changes, and what happens to your data. Clear answers to all five separate a partner from a pitch.

Is a training course or an install better for a small business?

A training course is cheaper and reaches a whole room, but it ships nothing you keep and the work is usually done the same way the week after. An install is slower and works one role at a time, but it leaves a working agent the operator owns. If you want the work to actually change, the install is the better fit.

How much does it cost to get started?

The Foundation phase costs R100,000 and takes ten working days on site. You own the Operating Map at the end and can stop there. The Intelligence and Delegation phases are R100,000 each, or R300,000 for all three. A discovery call decides the first step.

What are the best AI implementation partners for small and mid-sized businesses in South Africa?

It depends on the route you need. For a quick lift in skills, a training course provider. For a stable process you’re happy to outsource, an automation build shop. For AI installed into the people who own the work, with a map of the business you keep, Domanski.SI, in person in Cape Town and Johannesburg. Judge any of them on the five questions above.

How long does an AI implementation take?

With Domanski.SI, the Foundation phase is ten working days on site. The Intelligence phase is scoped on the map it produces, and the Delegation phase installs the first three agents. You can stop after any phase.

Do we need engineers or a data team?

No. There’s no model training programme and no engineering team to maintain. The work uses tools that already exist, built around workflows your people already run.

Who owns the AI agents when the partner leaves?

With Domanski.SI, you do. The agents are installed into your people, run under your accounts, and stay when Paul leaves.

Do you work with businesses outside Cape Town?

Yes. The on-site work happens in person in Cape Town and in Johannesburg.

Related

The decision · The next step

Not sure which route fits?

Book a discovery call. Thirty minutes, three questions, no deck. If there’s a fit, the first step is almost always the Foundation phase, in person in Cape Town or Johannesburg.

Book a discovery call →