The market filled up with AI suppliers in a short time, and most proposals look alike from the outside. These are the criteria that separate those who deliver from those who give a good demo.
1. Ask which process will be automated, not which technology will be used
A supplier who opens with models, architectures and platforms is selling technology. Whoever is going to deliver starts by asking how your process works, how much time it consumes and where it fails. The technical conversation comes after, as a consequence.
2. Insist on knowing what happens when the AI gets it wrong
No system never fails. What marks out a serious supplier is having a ready answer: where answers come from, how the source is verified, which actions require human approval and what gets logged. If the answer is that the model is very good and rarely fails, that is not an answer.
3. Confirm the code and infrastructure end up yours
Ask directly: at the end of the project, who owns the code? Where does the system run? Can I move maintenance to another team without renegotiating? A clear yes to those three questions removes half the risk of a project like this.
4. Get scope, timeline and price in writing before starting
AI projects have a reputation for overrunning precisely because many start without fixed scope. A serious proposal states what it delivers, in how long and for how much, and also states what it does not deliver. If the price depends on hours spent with no ceiling, the risk sits entirely on your side.
5. Check where the data lives
Hosting in the European Union, data not used to train public models, permissions by profile and usage logging. This belongs in the contract, not just in a conversation. It is the difference between having an answer when someone asks and not having one.
6. Prefer someone who turns work down
This is the least obvious criterion and probably the most useful. A supplier who says yes to everything, guarantees returns before understanding the operation and never suggests not proceeding is selling. Whoever works seriously says when a case does not add up, because a project with no return costs them their reputation.
7. Start small, but start properly
Be wary both of anyone proposing to transform the whole company at once and of anyone proposing a pilot with no connection to real systems. The first is unnecessary risk; the second is an expensive demo that goes nowhere. The useful path is one real process, connected to real systems, in production, with a metric defined upfront.
In short
If a supplier explains the process before the technology, tells you what happens when the AI is wrong, hands over the code, fixes scope and price, keeps the data in Europe and is capable of telling you no, you are talking to someone who has done this before.