After three meetings everything blurs: the most likeable provider rises, so does the one with the prettiest demo, and the substantive questions disappear. The only defence is to score cold, the next day, on the same criteria for everyone.

Here is the grid. Ten criteria, ten points each. It goes with the ten questions in this series.

The ten criteria

  • Understanding of your business processes. Did they ask about how you work, or talk about their method?
  • Real AI expertise. Can they say what a model does badly, and not only what it does well?
  • Automation and integration expertise. Do they know the tools you already use, and their limits?
  • Technical architecture. Can they describe an end to end path, with its breaking points?
  • Error handling and security. Do they give precise mechanisms, or general reassurance?
  • Return on investment. Can they put figures on it, and will they do so in front of you?
  • Ability to industrialise. Have they already taken a project from prototype to daily operation?
  • Maintenance. Do they announce an annual cost and a responsible person?
  • Documentation and skills transfer. Do they accept that you could take back control?
  • Quality of references. Can they describe a case end to end, incidents included?

How to read the total

  • 85 to 100. Excellent candidate. What remains is to validate the contract and the real availability of the announced team.
  • 70 to 84. Interesting. Book a second meeting on the two weakest criteria before deciding.
  • 55 to 69. Watch closely. Ask for a short, paid framing phase before any long commitment.
  • Below 55. The risk sits with you, not with them: you will carry the consequences of what they cannot yet do.

Two rules of use

First, score alone before discussing it with the team: a collective score forms around the most convinced person, not around the facts. Second, write "zero" when a question went unanswered, not "five". A non answer is information, and often the most useful of the meeting.

A provider who loses points by answering honestly will gain more of them in operation.

What the grid does not measure

It measures neither the real availability of the team that will work for you, nor their ability to say no during the project, nor what happens the day the budget tightens. Those three only reveal themselves by starting small: a short, paid framing phase with a precise deliverable tells you more than three meetings.

We publish this grid knowing it will be used against us, and that is exactly as it should be. If you would rather start with a costed picture of your own situation before comparing providers at all, that is precisely what our digital maturity audit is for.