How to interview an AI consultant

A pitch meeting tells you how good a firm is at pitch meetings. In a field where marketing skill and delivery skill are barely related, that is close to useless.

So do not run a pitch. Run a working session and make them think in front of you.

Structure the meeting so they cannot present

Say in advance that you would like to spend the time on a specific problem rather than on their credentials, and that you will send the problem beforehand.

Two effects. It removes the rehearsed material, and it filters: firms whose value is a template will be visibly uncomfortable, and some will decline.

Give them roughly an hour. Bring the person who actually does the work in the process you are discussing, not only the decision-maker, because the consultant's questions to that person are the most informative part of the whole meeting.

Hand them a real problem, including one that is not automatable

Pick two processes. One that you think is a genuine candidate, and one that is not, without telling them which is which.

The second is the test. A good consultant will identify the bad candidate and explain why: the frequency is too low, the exception rate too high, the process is disputed between departments, the human contact is the value. Somebody who enthusiastically maps a solution onto both has a hammer and has just shown it to you.

This single exercise separates the field faster than any question you can ask.

Watch what they ask, not what they answer

The strongest signal in the room, and it is available in ten minutes.

Do they ask how many times a week it happens? What the exceptions are and how often? What happens when somebody is on vacation? Where the data lives and whether the systems agree? Whose job changes if this works?

A conversation dominated by the technology's capabilities is a sales meeting. A conversation dominated by how work moves through your company is a diagnosis. The full list of what a competent one asks.

Ask them to be wrong on purpose

Two prompts that produce more information than anything else.

"Tell me about something you built that did not work." Real practitioners have several and tell them readily, with specifics. People who have not built anything produce a challenge that was overcome.

"What would you recommend if you did not build anything yourself?" A good answer includes buying something off the shelf, fixing a process without technology, or waiting. An answer that lands exactly on what they sell is the answer to a different question.

Make them show you something live, and drive it yourself

Ask to see a working thing, then take the keyboard.

Give it an odd input. Ask something outside its scope. Feed it the awkward case from your own business. You are not testing whether it handles everything. You are watching what happens when it cannot: clean stop and handover, or confident improvisation.

Reluctance to let you touch it is itself the result.

Interview the person who will do the work

The commonest substitution in consulting is a senior person selling and a junior person delivering.

Ask to meet the individual who will be on your project, and interview them, not their manager. Then put the names in the agreement with a substitution clause.

What to notice in how they handle disagreement

Push back on something they say. Not aggressively, just properly: tell them you think they are wrong about one point.

What you want is somebody who either defends the position with reasoning or concedes cleanly. What you do not want is somebody who immediately agrees with you, because you are hiring them for judgment, and a consultant who folds in a sales meeting will fold in month three when a senior person in your company wants the wrong thing built.

Close with logistics, because the answers are diagnostic

What do you need from us and how much of our time. Whose accounts will this run under. What does it cost to run monthly. What happens when it does not know something. Who owns the configuration.

Specific answers to all five, offered without discomfort, is what a competent supplier sounds like. The full question list with good and bad answers.

Then buy something small

No amount of interviewing beats two weeks of working together. Buy a diagnosis, or a review of a plan you already have, or one contained automation, and treat it as the real evaluation.

A supplier pushing for a long commitment instead of a small first piece is managing their revenue rather than your risk.

Point all of this at us if you like. Our first session is deliberately a working session rather than a pitch, and it is set up so the roadmap is yours whether we build it or not: how it runs.