The hiring decision is easier than the technology. You do not need to understand transformers to tell whether someone is going to waste your money.
Written for the person signing the invoice. Last reviewed 8 August 2026 — sources at the foot of the page.
Most advice on this subject is written by firms that sell AI consulting, which makes it advice about why you need AI consulting. This is not that. The questions below are the ones that reliably separate someone who will change how your business runs from someone who will deliver a slide deck and an invoice.
The context worth knowing before you start. An MIT report widely cited through 2025 and 2026 found that roughly 95% of enterprise generative-AI pilots produced no measurable P&L return. That number gets quoted carelessly, so here is the honest version: it measured whether a pilot produced rapid profit-and-loss impact within six months, and it looked mostly at sales and marketing deployments — the weakest category in the study. It is not proof that AI does not work. It is good evidence that most pilots are scoped badly, and scoping is exactly what you are hiring for.
This single question does more work than any other. A consultant who answers "nothing, it can all be automated" is selling. A good one will name specific things immediately — decisions with legal exposure, anything where a confident wrong answer costs more than a slow right one, workflows where your data is too messy to be worth automating yet. If they cannot name a bad use case in your business within a minute of hearing about it, they do not understand your business yet.
Not a case study written by their marketing department — a system that is running. Ask who built it, whether they built it or subcontracted, and what broke in the first month. The answer to "what broke" is more informative than the whole rest of the conversation. Everyone who has actually shipped an AI system has a list; people who have only sold them do not.
Most people ask for references and never call. Call them. Ask one question: would you hire them again for the same money? Then stay silent. The pause is where you learn something.
Never sign a large engagement first. Pay for a bounded piece of work — an audit, a mapped workflow, a working prototype of one process — with a fixed fee and a fixed date. You will learn more about how someone works from two weeks of paid work than from two months of proposals, and if it goes badly you have lost a small amount of money instead of a quarter.
Be wary of free discovery. A free audit is a sales process, and its conclusion is reliably that you need the thing being sold.
The failure mode is nearly always the same: a tool gets installed, nobody's job changes, and six months later the licence gets cancelled. Ask specifically whose daily work will be different when this is done, who will own it after the consultant leaves, and what happens to the hours that get freed. If the answer is vague, the project will be too.
Get it in writing before work starts: who owns the prompts, the fine-tuned models, the workflow configurations and the data you supplied. Ask what happens if you stop working together — can you export and keep running it, or does it live inside their account? A consultant who has thought about this will answer immediately. One who has not will improvise, which is your answer.
"AI consultant" covers at least four different jobs — see what an AI consultant actually does. Hiring a strategist when you needed an automation builder, or a builder when you needed someone to argue with your board, is the most common and most expensive mistake in this market. Decide which job you are hiring for before you take a single call.
Spend an hour writing down the one process that costs you the most time, with real numbers: how many people, how many hours, what it costs you a year. Bring that to the conversation. It changes the dynamic completely — you are no longer being told what your problem is, you are asking whether they can solve the one you already identified. The free ROI and readiness calculators will get you those numbers in about ten minutes.
Last reviewed 8 August 2026. Where a source is published by a firm that sells the service it is describing, we have said so, because that is a conflict of interest you should weigh.
Next: read the thirteen profiles to see who is actually publishing work in this field, or build your own numbers first with the free ROI and readiness calculators.