The market is suddenly full of AI consultants, and most of them were something else eighteen months ago. If you’re about to hire an AI consultant, for an enterprise or a small business, this guide is the filter I’d hand a friend.
TL;DR
- A serious AI consultant works from a named workflow, not a named technology. If they arrive selling a chatbot or an agent, they are looking for somewhere to put one.
- The job has four deliverables: an audit with prices on it, an architecture that fits your reality, a build that reaches production, and a result measured against a baseline recorded before the work began.
- Vet on evidence: ask for a deployment they can name, a number it returned, and a reference you can call. Anything else is portfolio theatre.
- A good statement of work names the workflow, the baseline, the acceptance test, the handover, and a kill clause. A good first call shrinks the project rather than growing it.
- Small businesses often see faster payback than enterprises, because the work AI does best is exactly what consumes a small team’s week.
Start with the workflow, not the technology
The single best preparation takes one sentence: name the workflow that is too slow, too manual, or too expensive. “Our quotes take three days.” “Support drowns every Monday.” “Reporting eats a full headcount.”
A serious AI consultant works from that sentence. A weak one works from a technology. They arrive selling chatbots, or agents, or whatever the current season favors, and go looking for somewhere to put one.
What an AI consultant should actually deliver
Strip the mystique and the job has four parts:
- An audit with prices on it. Where AI belongs in your business, where it doesn’t, and the expected return of each candidate, so the decision is financial, not fashionable.
- An architecture that fits your reality. Your data, your compliance constraints, your team’s ability to operate the thing after the consultant leaves.
- A build that reaches production. Demos are weekend work. Production, with error handling, monitoring, and a rollback plan, is the actual job.
- A measured result. Against a baseline recorded before the work began. No baseline, no claim.
If a proposal is missing the prices or the baseline, you are buying enthusiasm.
Questions that expose a weak consultant
- “What shouldn’t we automate?” A real AI automation consultant has a confident answer; a salesman hesitates, because every workflow looks automatable when you’re paid to automate.
- “What happens when the model is wrong?” Production AI is wrong on schedule. You’re listening for error budgets, human review loops, and fallback paths, not reassurance.
- “Who operates this in month six?” The correct answer involves your team and a handover plan, not a permanent retainer disguised as architecture.
- “What did your last deployment return?” A number, a timeframe, a name you could call. Anything else is portfolio theatre.
How to vet an AI consultant: run the references, not the deck
The deck is the sales artifact. The reference is the evidence. Before you sign anything, do three things in this order.
First, ask for one deployment they can name in full: the company (or its shape, if it is confidential), the workflow, the baseline before, and the number after. A consultant who has shipped real work answers in specifics and offers a name you can call. One who deflects to “we work with several enterprise clients” is protecting the fact that there is no number.
Second, make the call. A two-minute conversation with a past client answers what no proposal can: did the thing reach production, did it stay there after the consultant left, and would they hire them again. Ask the reference what broke, not what worked. The useful answer is in the failure mode.
Third, check who operates it now. If the past deployment only survives because the consultant is still on retainer, you are not buying a capability, you are renting a dependency. The work you want is the work that outlived the engagement.
What a good statement of work looks like
The contract is where good intentions become accountable, and most weak engagements are weak because the statement of work was vague on purpose. A statement of work worth signing names five things.
- The workflow. One process, specifically. Not “improve operations with AI.”
- The baseline. The current cost in time or money, recorded before work begins. This is the line everything is measured against, and a consultant who skips it is leaving themselves room to claim a win that cannot be checked.
- The acceptance test. The number and the date that decide whether the work succeeded. Written down, agreed by both sides, before the build.
- The handover. Who operates the system in month six, what documentation and training they get, and when the dependency on the consultant ends.
- The kill clause. What happens, and what it costs, if the acceptance test fails. A consultant confident in the work will agree to be judged by it.
If a proposal is missing the baseline or the acceptance test, you do not have a statement of work. You have an invoice with a story attached.
Red flags in an AI consulting proposal
Some warnings are visible before you ever sign. Treat any of these as a reason to slow down.
- No prices on the audit. A recommendation without an expected return is a preference, not an analysis.
- A company-wide transformation as step one. Anyone who insists on starting big is asking you to fund their learning curve on your budget. The right first project is small and falsifiable.
- A permanent retainer disguised as architecture. If “the architecture” only functions with the consultant permanently attached, the design is the lock-in.
- Silence on what happens when the model is wrong. Production AI is wrong on schedule. A proposal with no error budget, no human review loop, and no fallback path has not met production.
- Confidence with no baseline. Enthusiasm is cheap. A number you can check is not.
How to run the first call
You can learn most of what matters in thirty minutes if you drive the call instead of being pitched. Open with the workflow, not the technology: describe the process that is too slow, too manual, or too expensive, and watch whether they work from it or steer back to their product. Then ask the three questions that separate operators from salesmen, what you should not automate, what happens when the model is wrong, and who operates this in month six, and listen for confident specifics rather than reassurance. Close by asking them to scope the smallest version of the project that would prove or disprove the idea in ninety days. The consultant worth hiring will try to make the first project smaller. They are the one planning to be judged by the result.
AI consulting for small business: the honest math
Counterintuitively, small businesses often see faster AI payback than enterprises. The work AI automation does best (quoting, follow-up, support triage, document handling, reporting) is precisely the work that consumes a small team’s week. And a small business can deploy in weeks what an enterprise debates for quarters.
The math that matters: a workflow that consumes twenty hours a week, automated at even eighty percent reliability with human review, returns the cost of a fixed-scope engagement inside a quarter. Start with one workflow. Prove the return. Then widen.
The shape of a good first engagement
Small, fixed-scope, and falsifiable: one workflow, one baseline, one number that will prove or disprove the work within ninety days. Anyone who insists on starting with a company-wide AI transformation is asking you to fund their learning curve. (And if you haven’t settled which workflow deserves the bet, that’s a strategy question. The growth strategy consulting guide covers how to decide.)
Hire the consultant who tries to shrink the first project. They’re the one planning to be judged by the result.
Frequently asked questions
What should an AI consultant deliver? Four things: an audit with the expected return priced against each recommendation, an architecture that fits your data and team, a build that reaches production with error handling and monitoring, and a result measured against a baseline recorded before the work began. Missing the prices or the baseline means you are buying enthusiasm.
What questions should I ask an AI consultant before hiring? “What shouldn’t we automate?”, “What happens when the model is wrong?”, “Who operates this in month six?”, and “What did your last deployment return?” A real operator answers the first three with confident specifics and the fourth with a number, a timeframe, and a name you can call.
How much should a first AI engagement cost or cover? It should be small, fixed-scope, and falsifiable: one workflow, one baseline, one number that proves or disproves the work within ninety days. Favor a consultant who tries to shrink the first project over one who wants to start with a company-wide transformation.
Do small businesses need an AI consultant, or is it only for enterprises? Small businesses often see faster payback. The work AI does best, quoting, follow-up, support triage, document handling, reporting, is exactly what consumes a small team’s week, and a small business can deploy in weeks what an enterprise debates for quarters.