Everyone is selling AI right now. How do we tell who is any good?

The market is crowded and most of it is repackaged. Here are the questions that sort it out, and none of them are technical.

Ask these

1. What have you talked a client out of? Anyone who has delivered has done this. Anyone who cannot name an example has either never delivered or never said no.

2. Tell me about one that went wrong. Listen for whether they take any responsibility. "The client changed their mind" as a complete answer is a warning.

3. Where will our data be, and whose account holds it? Hesitation here is disqualifying.

4. What does it cost to run and maintain? If they have not thought about it, you will find out later and it will not be pleasant.

5. What happens when it gets something wrong? Anyone who says it will not is either naive or selling.

6. Can I speak to a client who is nine months in? Not one who just launched — everyone is happy at launch. Nine months tells you about support, change requests, and whether they still answer the phone.

The one that reveals most

Ask what they would need from your team. A supplier who says "very little" is either not doing discovery properly or is telling you what you want to hear. Both end the same way.