How to choose an AI consultancy (or agency): the honest guide

In brief. "Which is the best AI agency" is the wrong question: anyone giving you an answer is almost always selling themselves or a sponsor. The question that counts is a different one: who takes responsibility for the number this project is supposed to improve? From there you pick the right type of supplier and the real criteria. At the end, the 7 questions to ask before signing, useful with any supplier, not just with us.

"Which is the best AI agency?"

You have probably asked yourself this too, in roughly those words, in a search box. It is the most natural question in the world when you have to hand an important piece of your company to somebody you do not know. The problem is that it has no honest answer.

Whoever compiles that ranking has an interest in the result: either they appear in it, or somebody pays them to appear in it. There is no third party measuring AI agencies the way, say, an authority certifies a set of accounts. What you are reading is marketing dressed as information, and it is that regardless of who writes it, us included if we ever published a similar list.

So the first honest thing to do is not to look for a better answer to the same question. It is to change the question.

The question that really counts

Not "who is best", but: who takes responsibility for the number this project is supposed to improve?

"Ask them, before you sign, how they will let themselves be judged. If the answer is vague, you already have your answer."

Revenue lost in a slow process, a team's hours consumed by repetitive work, margin eroded by a tool nobody really uses: whatever the problem, there is a number behind it. The right question is not about the supplier's CV. It is about who, when things do not go to plan, answers for that number, and how they demonstrate it before you even start working together.

That shifts the choice from "who has the nicest website or the glossiest case studies" to "who agrees to be measured". And it is a criterion you can apply to any supplier in front of you, not just to us: ask them, before you sign, how they will let themselves be judged. If the answer is vague, you already have your answer.

The types of AI supplier (and when each one makes sense)

Under the word "AI" hide very different service models, and they are not interchangeable:

• Freelance / single consultant. Fast and cheap for an isolated task. Risk: you depend on one person, little structure, no continuity if that person changes project.
• AI agency / dev shop. Builds what you ask, per project. Good for a defined, circumscribed need. Risk: they deliver and leave, adoption and maintenance stay with you.
• System integrator. Puts existing technologies together at large scale. Suited to big companies with a structured IT function. Risk: too slow and too expensive for a smaller company.
• Software house / SaaS. Sells you a ready-made product. Fast to activate, but identical for every customer and designed to sit outside your specific processes.
• Operating Partner. A team that goes into the processes, builds the systems and stays, with responsibility for the result measured over time, month after month.

None of these is "the best" in the abstract. They are answers to different questions, and the question is: how strategic is the problem, and who do you want answering for it?

How to read this. It is not a ranking: it is a map. The right supplier depends on the problem in front of you, not on the most visible brand.

From the workbench: how to structure accountability, not just declare it

It is worth describing how we do it, not to sell you the model, but because it is a concrete example of what "taking responsibility" means when it stops being a line on a slide.

We never enter a company with an open-ended project and an annual contract signed blind. We enter with a Pilot: a bounded scope, a precise problem, and objective acceptance criteria defined before a line of code is written. If the pilot does not deliver the agreed result, the risk of the first step was ours, not the client's. Only afterwards, if the pilot works, do we move to the Retainer: we stay, we integrate, we scale.

And every month, not just at the start, we produce a Value Report: the document by which we agree to be judged ourselves, in euros. It is not a vanity-metrics report on how many automations we built. It is the number we promised to move, set against what actually happened.

I am not writing this because we think we are the only ones who can do it. I am writing it because it is a replicable format: if another supplier offers you something equivalent, under another name, you have found a serious partner. If nobody can tell you how they intend to be judged, the problem is not the technology they are proposing. It is that they have not yet decided to be responsible for your number.

The 7 questions to ask before signing

Take them to the first call, with any supplier you are evaluating, not just with us. The answers tell you almost everything you need to know:

1. How will you measure the result of this work, and in what unit (euros, hours, both)?
2. Am I paying you for hours worked or for the value you generate?
3. Who is directly responsible if the project does not work?
4. Does the system genuinely reach production, or does it stay a pilot nobody will use in six months?
5. What is left with me, in skills and in tools, if we stop after six months?
6. Who on my team will be able to run it without you?
7. Will you show me a real case with starting and ending numbers, not just a logo?

If a supplier answers all seven clearly, they have almost always done this exercise with other clients before you. If they slip on two or three of them, that does not automatically make them dishonest: they may simply be the wrong type of supplier for this specific problem. But it is a signal worth exploring before, not after, you sign.

Warning signs, whoever you are talking to

• They give you a quote before understanding the problem.
• They only talk about technology and never about process or margin.
• They promise a guaranteed result with no conditions attached: no serious supplier does that, because the result also depends on what happens inside your company.
• They compare their price with that of a generic piece of software: those are two different things, and whoever compares them is confusing you on purpose.
• They cannot tell you clearly what is left with you at the end of the contract.

Where we sit, transparently

We are an Operating Partner. We do not sell a tool: we go into your processes, find where you are losing margin, build the systems and agree to be measured on the value generated, every month, with the Value Report described above. It is the right model when the problem is strategic and you want the value to compound over time. It is not the right model for an isolated task: for that, the honest answer is often a freelance or a ready-made tool, and we would tell you so in the first call anyway.

It is not the best choice in absolute terms. That does not exist. It is the best choice when the question you have to answer is who takes care of your number, not just who promises it to you.

In short

Stop looking for "the best AI agency": that list does not exist in any honest form, and whoever offers it to you has an interest in the result. Look instead for whoever agrees to be measured on the number you want to improve, choose the type of supplier that fits the size of the problem, and ask the seven questions above before signing anything.