How to Choose the Right AI Consultant for Your Small Business

If you are a small-business owner looking for an AI consultant, you are walking into a market that has tripled in size since 2023 and where most of the new entrants have less experience than your average TikTok marketer. This is not a criticism of anyone in particular. It is just the state of the field. LLMs made a lot of surface-level work feasible for people without a traditional engineering background, and the consulting market has filled up accordingly.

That is not automatically bad for you. Some of those consultants are excellent. Some of the traditional engineers are out of touch. The question is not what credentials someone has. The question is whether they can solve your problem without wasting your time or money.

Here is the framework I use when a business owner asks me how to pick between AI consultants, including how to decide whether I am the right fit or not.

Signal 1: Can they describe what they will do in plain English?

If a consultant cannot, in a single conversation, tell you:

• What specific process of yours they intend to change
• What the inputs and outputs are
• What a successful outcome would look like
• What they will hand off when the engagement ends

then they have not thought about your problem yet. They are selling the category, not the solution. Ask the same three questions to three different consultants and see whose answers sound the most specific to your business, not the most impressive in a vacuum.

Signal 2: Do they ask for your data before pitching a solution?

AI projects fail more often because of bad inputs than bad models. If a consultant proposes a custom GPT, an AI agent, or a workflow automation before they have seen your actual data, your actual team, or your actual customer interactions, they are working from assumptions.

A good first conversation looks less like a pitch and more like a discovery session:

• "Can I see an example of the spreadsheet you are trying to replace?"
• "How often does this process run, and what breaks when it fails?"
• "Who currently owns this work, and what is their opinion on automating it?"

If a consultant skips this and jumps straight to architecture or tooling, they are more likely to deliver a solution in search of a problem.

Signal 3: Are they transparent about what AI cannot do?

This is the cleanest filter. AI is not magic. It is a toolkit with well-known limitations: it hallucinates on rare cases, it is expensive per token at scale, it can leak data if integrated badly, it struggles with truly novel situations. A consultant who acknowledges these limits up front is almost always more useful than one who treats every problem as solvable.

The most expensive AI consultants I see are the ones who promise what the technology cannot actually deliver. By the time you find out, the retainer has already billed for three months.

A useful gut check: ask "what kinds of problems would you turn down?" If the answer is "none, we can build anything," you are talking to someone whose business model depends on saying yes. If the answer lists three or four categories they would not touch, you are talking to someone with experience.

Signal 4: Can they show real work, not just logos?

A portfolio page full of client logos is almost meaningless. Logos are easy to get from pilot engagements that never shipped. What matters is whether a consultant can show you:

1. Something that is actually running - a live URL, a live product, a live workflow. Not a deck, not a demo video, the actual thing in production.
2. A measurable outcome - time saved, revenue lifted, cost reduced, with numbers they can defend.
3. A clear account of the failure modes - what broke, how they fixed it, what they would do differently. Anyone who pretends their past work was seamless is lying.

Case studies without numbers are marketing. Numbers without a live reference are also marketing. Ask for both, in the first meeting, and see how comfortable they are sharing them.

Signal 5: Do their pricing and scope align with a real business model?

Price is informative in both directions. Very cheap AI consulting (under $50-$75/hour) often means the consultant is offshoring the work or using pre-built templates you could buy yourself. Very expensive ($500+/hour) for a small-business project usually means you are paying for agency overhead you do not need.

A reasonable range for a single-operator consultant working on small-business AI projects in North America sits between $150 and $350 per hour, or equivalent fixed-price projects scoped from that. If the price is far outside that band, ask why.

Beyond the hourly rate, look at how the engagement is structured. Good engagements have:

• A clear, bounded first deliverable - not an open-ended "we will explore opportunities"
• A fixed-price option for the build phase, even if the consultation is hourly
• An explicit handoff plan - what you get to keep, what documentation exists, what happens if you stop working with them
• A subscription or retainer component only if there is an ongoing product to maintain, not just "we will be available"

Red flags worth paying attention to

A few things that should make you walk away, even if the other signals look fine:

• Vague AI jargon where plain words would work. "Leverage proprietary agentic architectures to optimize your stakeholder workflows" means nothing. "Automate the emails you send after a customer books" means something.
• No code samples when you ask for them. A consultant who cannot share even anonymized code snippets is either not writing code or is hiding something.
• Unable to explain their tooling choices. If they can not tell you why they chose GPT-4 over Claude, or Airtable over Postgres, they are not making choices - they are defaulting to what everyone else uses.
• Contracts that bind you into monthly fees forever. Ongoing support is fine. Lock-in is not.
• They will not discuss competitors. If asked "who else might be a good fit for this work?" they should be able to name real alternatives. If they deflect, they are selling themselves, not solving your problem.

A short process that almost always works

Here is a three-week process I recommend to owners who are serious about finding the right consultant:

Week 1: Write a one-page problem statement

Not a specification. A description. What is the work that is painful right now. Who does it. How often. What breaks when it breaks. What a "fixed" version would look like to you. One page, in your own words.

Week 2: Talk to three consultants

Give each of them the one-pager. Ask them the same questions. Notice who asks for more data, who jumps to a pitch, and who tries to sell you something different than what you asked for. Do not decide anything yet.

Week 3: Ask for scoped proposals from the two who listened best

A proposal should include: the specific problem they heard you describe, the approach they would take, the timeline, a price, and what they would hand you at the end. Compare them side by side, not on price but on who seems to understand your situation most clearly.

The short version

Choosing an AI consultant is mostly about finding someone who has done the specific kind of work you need before, can describe it plainly, and is honest about what they can not do. The price is informative but not decisive. The portfolio matters, but only when it includes live references with real numbers.

And if the person across the table cannot answer "what would you turn down?" without squirming, keep looking.