Before hiring an AI consultancy, find the part that actually needs AI
Before hiring an AI consultancy, work out which part of the problem actually requires AI. It is usually smaller than the proposal.
Say a logistics company wants "AI for dispatch." Six weeks into scoping, the problem breaks into four pieces. Drivers are not updating status, which is a phone app and a manager. The rate table lives in a spreadsheet with three owners, which is a database and one owner. Exceptions are handled by whoever is in the office, which is a rule and a rota. And there is one genuine judgment call: which of two late loads to prioritise when a customer relationship is at stake.
One of four needs a model. The consultancy's proposal covers all four, because the consultancy sells models.
This is not dishonesty. If you ask a plumber what your house needs, the answer will involve plumbing.
So before the consultancy arrives, do the split yourself. Which decisions are rules that nobody wrote down? Which are data that nobody cleaned? Which are handoffs that nobody owns? What is left after those three is the AI problem, and it is the only part the consultancy should be quoting.
The proposal that comes back will be smaller. It will also be the first one you have received that describes your business instead of their product.