Before you sign an AI proposal: the questions to ask, from the side of whoever pays

You have a proposal on the table to bring AI into your company. It looks serious, but you have nothing to compare it with.

You do not know if it is disproportionate to what you need. You do not know if anyone has done the same thing in a company like yours. And you will have to explain the spend to whoever keeps the books.

These questions apply to any provider, including us. You judge by the answers, not by the technology.

Has anyone already solved a problem like mine?

This is the first check, even before the price: has the provider worked with companies like mine? It is not distrust. It is the normal way to reduce risk when you cannot judge the technology.

Ask for a comparable case and ask to speak with someone who lived it. Then separate two things: the result claimed by whoever tells the story and the result measured with data you can see.

If the provider has no similar case, they should say so. A clear “we have never done this” is worth more than an inflated reference.

You can read two client cases: Fabio Armellini and Pietro Landri.

How much AI do I really need, and how do I avoid buying too much?

A real fear is not paying too little. It is paying for a system bigger than the one you will use.

“I wouldn’t want a system that is oversized compared to what I actually use” — Fabio Armellini, call of 22 October 2026 (translated from Italian)

How to answer this fear before you sign:

• Start with a part you can stop. If the first step does not work, you must be able to leave without having bought everything else.
• Ask for modules with an end point. Pietro Landri put it this way:

“every module should end with a full stop and not with a comma” — Pietro Landri, diagnostic session of 28 September 2026

• Ask for the bare minimum. If the provider cannot tell you, the proposal is built around their catalogue, not around your problem.
• Decide who builds and who learns. Built for you or built together: there is no right answer. It depends on who will use the system every day.

How do I explain this spend to whoever pays?

Sometimes the person who decides and the person who keeps the books are two different people. Maybe it is you, and before signing you want to show the numbers to your accountant, your finance office, the budget.

You need the proposal in writing, but that is not enough. You need a measure that holds up in front of a third party.

“I trust the result, but then you have to explain it to whoever sees the result” — Fabio Armellini, same call

Ask for three things:

1. Where I start from. Today’s measure, before anything changes: times, hours lost, errors.
2. What should change after three months. Not a generic “more efficiency”, but something visible, like a rule the team really applies.
3. What is claimed and what is measured. And who took the measurement.

No serious proposal promises a guaranteed return. If one does, ask which cases it is based on.

What do I have left when the contract ends?

Here are the checks that apply to any provider. They tell you whether you are being left a system or a dependency.

Who answers for the result, by name?

Ask who will be the person who decides and signs off on AI choices in your company. If the answer is “the team” or “our platform”, nobody answers.

What do they leave me in writing?

A serious engagement produces documents that last: a register of tools, rules of use by role, who validates outputs. If all you have at the end is a presentation, you paid for work that cannot be repeated.

A useful reference on what a policy should contain is the operational checklist.

How do they handle my company’s data?

Ask which tools they will use, with which accounts, and where your documents end up. If they run the test on your documents using their personal accounts, you have just created a case of shadow AI with a provider.

What happens when the contract ends?

If the AI stops without them, you do not have a system: you have a dependency. Ask how they pass knowledge to your team and who in the company becomes the reference point.

Who checks the results before they go out?

A result must be checked before it leaves the company, and someone must have decided that beforehand.

As Fabio Armellini said in the same call: “I do, I verify and I check”. A serious proposal answers with a method, not with reassurance. It says who checks outputs before they leave the company. It says on which cases the system is tested before real use.

Which signs say the proposal does not hold up?

• It talks only about tools, never about rules and responsibilities.
• It promises results without saying how they are measured.
• It asks you to sign everything at once, with no part you can stop.
• It has no similar case for you to verify, and does not say so.
• It does not mention the AI Act or Italian law, not even to say they do not apply to you. For the updated picture, see AI Act, the delay only touched high risk.
• It refuses to put in writing what it delivers.
• It does not tell you who, inside your company, has to do what.

A project or an ongoing role?

There is no single answer. A bounded project, such as an assessment or a policy, can be handed to someone who delivers it and leaves. Governing AI over time, instead, needs a person present on an ongoing basis. We discuss this in The Fractional Manager is not an AI plug-in.

Where do I start before signing?

Whichever provider you choose, you need to know your starting point. The SME diagnostic gives you a first starting measure in a few minutes, to bring to whoever checks the accounts. If you want to see how we work, there is the For Companies page.