How to hire an AI consultant

Decide what problem you are hiring for

Most owners begin with a vague brief: we should be doing something with AI. A vague brief attracts vague proposals, and you end up comparing slide decks instead of plans. Before you speak to a single consultant, pick one recurring workflow that costs real money. Good candidates are jobs your team does by hand every week: assembling a proposal after a sales call, chasing missing documents from new clients, retyping intake forms into the practice management system, or following up on quotes nobody has answered.

Next to that workflow, write three rough numbers: how often it happens, how long it takes each time, and what tends to go wrong. For example, an office manager might spend four hours every Monday pulling client reports together from QuickBooks Online and a shared spreadsheet, and one report in five goes out with a figure that has to be corrected. That short paragraph is the best filter you have. Strong consultants will ask about it in detail. Weaker ones will skip straight to the tool they always sell.

Know the kinds of help on the market

AI help is sold in a few distinct shapes, and the prices differ by an order of magnitude. Knowing which shape fits your problem keeps you from paying senior rates for a simple build, or bargain rates for a problem that really needs judgment.

• Freelancers: best when you can write the specification yourself. Upwork's rate guide says specialized development, AI and consulting work often reaches $75 to $150 or more per hour.
• Automation agencies: build one system at a time, such as missed-call text-back or lead routing, often for a setup fee plus a modest monthly charge.
• Boutique AI consultancies: one 2026 rate guide puts them at $150 to $350 per hour. A Denver firm's own published figures put local AI consulting at $100 to $250 per hour and defined projects at $3,000 to $15,000.
• Fractional technology leads: a senior person who owns priorities, vendor choices and adoption over several months, usually on a monthly retainer.

Match the type of help to the job

No type is better in general. A freelancer is the right call when you already know exactly what should be built, for example a Zapier flow that copies new Jotform submissions into HubSpot. An agency fits when you want one proven system installed quickly. A boutique consultancy is useful for a defined project with some design work in it, such as drafting proposals from call notes with a person reviewing every draft.

A fractional lead makes sense when the hard part is not the build. If you are unsure which workflow to fix first, have three overlapping subscriptions, or have bought tools before that nobody uses, the gap is leadership and follow-through. Paying a builder to solve a prioritization problem usually produces a working automation that does not move the numbers you care about.

A useful way to decide is to ask who will make the hard calls. If you or your operations manager can say exactly what should happen at each step, you are buying hands, and a freelancer or agency is efficient. If the honest answer is that nobody knows yet, you are buying judgment first, and the cheapest hourly rate will not be the cheapest outcome.

Put the scope in writing before work starts

The best single predictor of a good engagement is a written scope you can understand without a translator. It should name the workflow, the systems involved, what will be delivered, what is out of scope, and how everyone will know it worked. If a consultant resists writing this down, that tells you something.

• A baseline: today's time per task, turnaround or error rate, measured before anything changes.
• Acceptance criteria: specific tests the work must pass using your own real examples.
• Ownership: your company owns the accounts, logins, prompts and automations, not the consultant.
• Running costs: subscriptions, usage fees, and who maintains the system after launch.
• Review points: where a named person approves output before it reaches a client.

Run the interview around a real example

Skip the capabilities presentation. Bring one anonymized sample of the workflow, such as a redacted intake form or last month's proposal, and ask the consultant how they would approach it. Listen for questions about your process before answers about software. Useful questions: What would you measure first? What would you leave manual here, and why? Which tools we already pay for could handle part of this? What happens when the AI gets something wrong? Who on my team would have to change how they work?

Pay attention when a candidate says the fix is a setting in software you already own. That answer costs them revenue, which is exactly why it is worth hearing. Ask for references from businesses of similar size and ask each reference one question: is the work still in use six months later?

Check how they handle your data

An AI consultant will touch client records, email and often financial data. Ask which AI services they plan to use, whether those services train on your data, where the data is stored and who can see it. Ask them to work inside accounts your company owns, with multi-factor authentication turned on, rather than their own logins.

Regulated firms have further questions. A law firm should ask how the approach fits its confidentiality duties and ABA Formal Opinion 512 on generative AI. A practice handling patient information should ask whether a HIPAA business associate agreement is needed with each vendor. Tax and accounting firms should ask how the work fits their written security plan under the FTC Safeguards Rule and IRS Publication 4557. A consultant should not give you legal advice, but they should know these questions and expect your counsel to weigh in.

Start small, then measure

Structure the first engagement around one workflow over roughly 60 to 90 days with a fixed end date: a short diagnostic, a defined build tested with real examples, then a measurement period. At the end you should be able to put the baseline next to the new numbers and decide to expand, adjust or stop. InTheory runs its own first 90 days this way, with a written scope signed before kickoff. Whoever you hire, ask for the same discipline. Someone confident in their work will welcome a scorecard.

Agree in advance what happens at the end. If the numbers improved, the next phase might extend the same approach to a second workflow. If they did not, the consultant should explain why in writing, whether the cause was the tool, the process or adoption, so the money still bought you knowledge.