Do you need an AI consultant?
I am one, so discount accordingly. But the honest answer is a decision tree, and one of its branches is "no."
Somebody on your board or in your inbox has asked what you're doing about AI, and now you're evaluating help. The options on the table are usually a consultant, a contractor, a full-time hire, or doing nothing. The pitch you'll hear from each of them is that they're the answer. Here is the version with the incentives removed, as far as I can manage it.
The choice turns on two questions, and neither is about AI. Do you already know what to build? And will this need someone permanently?
(If you'd rather click than read, there's a quiz version. It reaches the same verdicts, including the unprofitable one.)
If you know what to build and it's bounded: contractor
You've already diagnosed the problem and can write a page describing the system you want. What you need is hands. A good contractor or dev shop is cheaper than a consultant for pure execution, and a consultant who takes that work is mostly reselling you a contractor at a markup. The risk in this branch isn't the builder, it's your page of requirements: if the diagnosis is wrong, you'll get exactly what you asked for, which is the expensive way to find out it wasn't what you needed.
If AI will be your product or a permanent capability: hire
If the thing touches your core product, or you'll be building and maintaining AI systems indefinitely, you need that knowledge inside the company, accumulating instead of walking out the door at the end of an engagement. Hire full-time, and treat consultants at most as a bridge while you do. Anyone who tries to talk you out of a justified hire and into a retainer is describing their revenue model, not your interest.
If you don't actually know what's wrong: that's the consultant case
Most teams considering AI are not in either branch above. They have a symptom, the team is underwater, the process is slow, costs grow with volume, and a suspicion that AI should help, but no diagnosis. The valuable purchase in that state isn't building at all. It's finding out where the hours actually go, what's automatable, what it's worth, and what should be left alone. That's diagnosis plus architecture, it's senior work, it's inherently temporary, and it's the one thing on this menu a contractor won't do and a new hire can't do yet. It's also small: a bounded audit, not a transformation program. If the diagnosis says "build," you can then use any of the three branches to build, including the cheap ones.
Buy diagnosis when you're uncertain. Buy hands when you're sure. Buy both from the same person only if they can show you both skills.
And sometimes the answer is nobody
Two honest cases. If your process pain is really a decision pain, nobody agrees on priorities, ownership is murky, the tooling fights are proxy wars, then AI will automate the confusion, and any consultant who takes the engagement is taking your money to laminate a problem. Fix the decision first, alone. And if your volume is low and your workflow basically works, the boring answer is that off-the-shelf tools plus a curious employee with a few protected hours a week will capture most of what's available. A lot of companies should simply do that and not call me.
The test for whoever you do bring in
One question filters most of the field: ask what they'd do if the right answer turned out to be "don't use AI here." The wrong hire has no answer, because their business model requires the answer to be yes. The right one can describe walking away, and can usually point to a time they did. Whatever their title says, that's the difference between advice and sales.