How to choose an AI consultant: the questions to ask before you sign
Choose an AI consultant for judgement, not AI knowledge. Knowledge dates in months; judgement compounds.
The seven questions below expose it on the first call: whether they diagnose before prescribing, whether they can name the payback, whether the scope is fixed, what they did before AI, whether the capability stays with your team, how specific they get, and what they would tell you not to do.
Most firms choose an AI consultant on the wrong signals: confidence, a slick deck, fluency with the latest tools, a logo you recognise. None of those tell you whether the work will pay back. The market is full of people 18 months into AI, selling certainty. Ask these questions on the first call. The answers will sort the field faster than any proposal.
Will you diagnose before you prescribe?
The first thing a good consultant does is measure what your AI is actually doing, not reach for a tool. Ask how they would start. A good answer describes a diagnosis: what they would look at, what they would measure, how they would decide what to fix first.
A weak answer is already pitching a build or a platform before they have seen your business. If someone prescribes before they diagnose, they are selling what they have, not what you need.
Knowledge dates in months. Judgement compounds.
What will this pay back, and how will we know?
AI is easy to buy and hard to measure. Ask what the work will return and how you will both know it worked. A good answer puts a number on it, in hours or cash, and names how it will be tracked.
A weak answer stays on the technology: what the tool does, not what it returns. If nobody can describe the payback before the work starts, nobody will be able to prove it after.
Is the scope fixed, or open-ended?
Open-ended engagements move the risk onto you. Ask exactly what you get, by when, and for how much. A good answer has edges: a defined deliverable, a defined price, a defined finish.
A weak answer is a retainer from day one, with “we will see where it goes.” There is a place for ongoing work, but it should come after a piece with a clear shape, not instead of one.
What did you do before AI?
This is the question that separates judgement from novelty. AI knowledge dates in months. The judgement to know what is worth doing, in what order, is built over years. Ask what someone did before they became an AI consultant.
A good answer has a track record of real business work: operations, product, conversion, delivery, whatever it is, with results. A weak answer is 18 months of AI and not much underneath it.
Will the capability stay with my team, or with you?
The best outcome is that your people can run the systems after the consultant leaves. Ask who owns the work at the end. A good answer builds capability into your team as it goes, so the firm benefits twice: the systems pay back now, and the people compound for years.
A weak answer keeps the knowledge with the consultant. Good for them, expensive for you.
Can you show me specifics?
Vague is cheap. Ask for named clients, named tools, real numbers, and the things that did not work. A good answer is specific and a little uncomfortable in its honesty:
This client, this result, and this is exactly what we got wrong.
A weak answer hides behind “leading”, “world-class”, and case studies with every number redacted. Specifics cost effort to earn. That is exactly why they are a signal.
What would you tell me not to do?
The most useful thing a consultant can do is talk you out of low-value work. Ask what they would tell you to skip. A good answer says no to something: this is not worth it yet, that will not pay back, start smaller.
A weak answer is that everything is a great opportunity, because everything is something they can sell you. Anyone who never says no is not advising you. They are quoting you.
What runs through all seven
You are not hiring AI knowledge. It dates in months and everyone claims it. You are hiring judgement: what to fix first, what to skip, how to prove it paid back. The questions that expose judgement are the ones about evidence, scope, payback, and what someone will tell you not to do.
If you want to see what diagnose-before-prescribe looks like in practice, the AI Visibility Scorecard is a fixed-scope, fixed-price example: a defined diagnosis with a number at the end.
Anyone can sound certain about AI. Fewer can tell you what to leave alone.
