AI implementation for small business: how a project really runs

What AI implementation is — and what it is not

AI implementation means putting AI into an existing work process, so that afterwards the work is actually done differently. It starts with one piece of work that comes back every week, and it is only finished when the people who used to do that work lean on it without thinking.

That is something else than the four things often sold under the same name:

• Buying a tool. A licence is not an implementation. The software is switched on, nobody uses it, and a year later it is still on the books.
• A report. An advisory document with opportunities and a roadmap changes nothing as long as nobody executes it. The party that writes it should also be able to build it.
• A pilot that never ends. A trial without an agreed moment to say yes or no quietly turns into a hobby alongside the real work.
• A chatbot on your website. It can be useful, but it rarely touches the process where you are losing money and time.

What it looks like in practice

Concretely, AI implementation at a wholesaler means packing slips and order confirmations from suppliers landing in the system on their own, instead of someone retyping them. At an installation company it means job sheets from the engineer no longer being keyed in by hand on Friday afternoon. At an accounting firm it means incoming documents being recognised, sorted and prepared per client.

In all three cases the work itself changes. The system is the smaller half of the job; the larger half is getting the process around it right and making sure your people know what they still have to check.

The four phases: scan, core session, build, embed

We work in four phases. You can stop after any of them, and after phase two we will tell you ourselves whether carrying on is sensible.

1. 01Phase 1 — AI scan (free, five minutes online). Adaptive questions about your sector and your processes. You get a score, a short list of concrete opportunities and a first price indication. It is meant to decide whether a conversation is worth having, not to base a plan on.
2. 02Phase 2 — Core session on your shop floor (half-day of 4 hours, €596 excluding VAT and travel). We come to you, follow one or two processes end to end, talk to the people doing the work and count the handling steps. You get a core report: priorities, what each opportunity is worth, what it needs in terms of data and integrations, and a go or no-go per item. The report is yours and stands on its own, even if you do nothing further with us.
3. 03Phase 3 — Building (per hour: €110 for straightforward work, €165 for complex development). The same people who ran the session build it out: integrations, automation, custom software. We work with a fixed number of hours per month, paid up front as an AI budget or build deposit, so there is never an open end. Every month you see what was built and what was used.
4. 04Phase 4 — Embedding (from the same hours budget). Explaining it to the people who work with it, a one-page working instruction, looking over their shoulder in the first weeks and removing the errors that only surface in real use. After that, down to a small maintenance rhythm, or stop.

What AI implementation costs

Our rates are above and do not change per client. What does differ per business is the number of hours — which is exactly why we will not put a fixed project price on paper before we have watched the work. A fixed price given before anyone has seen the shop floor always contains a margin for the unknown, and you are the one paying for it.

The bill has three kinds of cost, and the second and third are the ones most often forgotten:

• Build cost: our hours, €110 or €165 per hour, drawn each month from the budget in advance. Predictable, because you decide how many hours go into a month.
• Running cost: what the AI models and connected services cost per month once it is live. That moves with your volume, so more orders means a higher bill. We work this through with you up front using your own numbers.
• Preparation on your side: cleaning up data, merging duplicate customer records, agreeing how something will be recorded from now on. This almost always takes more time than expected and appears in no demo.
• Maintenance: a supplier changes an integration, a form gets an extra field, a rule no longer holds. Count on a few hours a month once something is genuinely running.

How long it takes

The scan is a matter of minutes. We usually schedule the core session within a few weeks, and the core report follows shortly after; we agree the date when we book the session.

For the build: a first process that genuinely runs should be finished in weeks, not quarters. If a first project takes longer than that, it was almost always scoped too large, or it is stuck on data that still needs cleaning up.

What decides the lead time in practice is rarely our planning. It is access to your systems, the availability of the colleague who knows the work best, and how fast your software supplier answers a request for an integration. That is why we put those three things on paper at the start.

Who you need inside your own business

An implementation that only lives at the agency comes to nothing. You need three roles on your side, and two people may cover them:

• An owner. Someone with authority to decide who treats this as their subject, not as a job that landed on top of everything else. Usually the director or the operations manager.
• The person who does the work. The planner, the admin assistant, the work preparer. They know where the exceptions are, and the exceptions decide whether it works.
• Someone with the keys. Access to your system, your mail environment and your supplier accounts. It sounds trivial; in practice it is often why a project sits still for two weeks.

Where AI implementations go wrong

Four causes keep coming back. Not one of them has anything to do with AI.

• No owner. Everyone finds it interesting, nobody is responsible. The project does not die, it simply disappears behind the daily rush. This is the most common cause by far.
• Data that does not add up. Three article numbers for the same product, customers listed twice, addresses in a free text field. A system cannot guess what a person automatically fills in. Cleaning that up is dull manual work up front, and it is often the largest part of the first weeks.
• The wrong first process. Starting with invoicing, or with the time registration your payroll hangs on, is tempting because that is where the hours are. But if something breaks there, you pay it back in trust, and trust does not come back quickly. Pick something where a mistake is visible and repairable.
• A supplier who keeps their integration closed. Some packages have no open API, or only release one for a substantial fee or in a more expensive subscription. That is a real risk and it never appears in a sales pitch. We check this before anything is built; sometimes the outcome is that a different process goes first.

Buying a tool, hiring a large consultancy, or this

There are three ways to implement AI, and none of them is always the right answer.

Buying a tool yourself works fine as long as you stay inside one package. If you mainly want to answer email faster, write minutes or draft text, you do not need an agency: a subscription and an afternoon of practice with your team will get you further than we can. And if your current package already has an AI feature you are not using, switch that on first. It breaks down the moment two systems have to talk to each other and will not do so on their own.

A large consultancy delivers a solid analysis and a roadmap. The problem is the seam: the people writing the report are not the people building it, and half the context is lost between them. On top of that you pay a rate built for enterprise scale, which a small business rarely earns back.

Our model sits in between. We spend a half-day on your shop floor before we build anything, we give an honest go or no-go — including “do not do this” — and the same people who advise also build. Fixed hourly rates, paid in advance from a budget you set, so there is no open end. And if the core session shows you are better off with a standard package, we say so. That is a short conversation, but an honest one.

Beyond that: we carry professional indemnity and business liability insurance, and a subsidy partner routinely checks whether a grant scheme applies to your project. That is not a promise that money is available, only that someone looks before you start.

The first step

Start with the free AI scan. Five minutes, no sales call: you get a score, a few concrete opportunities and a price indication. That tells you whether a conversation is worth having.

If the answer is yes, we schedule the core session: a half-day of 4 hours at your location, €596 excluding VAT and travel. A full day is possible too (€1,192) if several departments or sites are involved. Our advisory rate is €149 per hour.

After that session you have a core report with priorities and a go or no-go per opportunity. What you do next — build with us, carry on yourself, or nothing — is up to you.

Frequently asked questions

What we do for this

Further reading

Other topics

Start with the free AI scan

Five minutes online, no sales call: you get a score, concrete opportunities and a price indication. Prefer to talk it through first, book an intro call and we will tell you whether a core session on your shop floor makes sense.