Before signing with an AI consultant, ask these key questions: who will work on your project, what experience do they have in your sector, how do they measure ROI, who owns the data and code, and what happens if the project fails. The answers separate the experts from the salespeople.
Why these 15 questions are worth more than 3 discovery meetings
Most Italian SMEs that select an AI consultant get the evaluation phase wrong. Not from lack of time, but from lack of the right questions. They end up asking the consultant things they can answer from memory ("what are your services?", "what technologies do you use?") and neglecting the questions that actually put them on the spot.
Questions about experience and track record
1. How many AI projects have you taken to production in the last 24 months?
Not "how many clients do you have," not "how many projects have you started": how many are in real production today, with users actively using them. A serious consultant has precise numbers.
2. Can I speak with two clients who have projects similar to mine?
Two, not one. And not chosen by the consultant alone: ask for a shortlist you can choose from. Anyone who refuses reference calls is hiding something.
3. Tell me about the last AI project you failed
The answer "we have never had one" is a red flag: either they are lying, or they have worked too little to have statistically encountered failures.
4. What percentage of your clients renew after the first project?
If the renewal rate is under 50%, something is broken in their delivery model. If it is above 80%, verify it is not lock-in disguised as loyalty.
5. Who, physically, will work on my project? Can we meet them before signing?
Slide salesmen organize the first meeting with the commercial director and the founder. Then they disappear, and the project ends up with unsupervised external freelancers or juniors.
Want to apply this in your business?
At IL DOGE DI VENEZIA we support Italian SMEs through every phase of AI transformation. The first conversation is free.
Tell us about your projectQuestions about method and deliverables
6. What is your working method? Can you show it to me in writing?
Serious consultants have a documented method with phases, timelines, deliverables, and KPIs. Weak ones talk about a "custom approach" to mask the absence of a process.
7. What are the concrete deliverables for each phase of the project?
They should list them: document X by week Y, working prototype by week Z, deployed model by week W. If the answer is vague, you are buying air.
8. How do we define success KPIs together before the project starts?
A serious consultant insists on fixing baselines and KPIs before kickoff.
9. How much of my team's time will be required during the project?
They should answer with estimated hours per role. "Very little, we handle everything" means they are building a system your team will never know how to manage.
10. What happens if the pilot does not reach its objectives?
There must be written go/no-go criteria, possible pivots, exit conditions, and partial refunds.
Questions about ownership, responsibility, and exit
11. Who owns the models, prompts, datasets, and code you will build?
Acceptable answer: everything belongs to the client company. Unacceptable: everything belongs to the consultant and you only get a usage license.
12. Where will company data live during and after the project?
They must be able to answer precisely: cloud infrastructure, region, encryption, deletion policies.
13. Who will be responsible for the system after go-live?
The right answer always includes an internal contact on your side, with a detailed training plan.
14. How can we exit the contract if we decide to change providers?
There must be a clear exit clause: notice period, data and artifact return, complete technical documentation, migration assistance.
15. What is your price range for a project like mine?
Do not ask for the exact price at the first meeting, but ask for a range: 5-10k, 20-40k, 60-120k. A serious consultant gives it immediately because they have done enough projects to know their market's ranges.
Answers that should make you walk away
- "We have never failed a project" — they are either lying or inexperienced. Both are dangerous.
- "The models stay with us, you just get a license" — you are buying permanent lock-in, not a business asset.
- "Our method is completely custom, nothing is written down" — without a standardized method, every project is an improvisation.
- "Your team does not need to be involved, we handle everything" — the system will never take root and will die the day the consultant leaves.
- "We cannot give references because our clients are confidential" — satisfied clients always authorize reference calls. Those who do not have them cannot give them.
The next step
These fifteen questions are the final step in a broader selection process. If you want the complete 4-week method, read our guide on choosing an AI consulting firm. If you prefer to skip the scouting phase and speak directly with a consultant who accepts all these questions without flinching, discover our AI consulting service for SMEs or talk to us. The first call is free.