Key takeaways
- Ask for proof they build, not just slides about what AI could do.
- Your business should own the accounts, workflows and data from day one.
- A named human should review AI output before it reaches you or your customers.
- Get an all-in monthly number and a plain list of what's not included.
- Know exactly what you keep and how the handoff works if you cancel.
On this page
Proof they actually build
Anyone can call themselves an AI consultant. The fastest way to separate builders from talkers is to ask them to show you something real. Plan on a 30-minute call and bring these three questions, plus a short list of the tasks you'd actually hand off. The 50 Tasks checklist is a quick way to build that list.
| Question | A good answer sounds like | Red flag |
|---|---|---|
| 1. Can you show me something you built that's running right now? | They share their screen and walk through a live workflow, what it does and what broke along the way. | Only slides, mockups or "I can't show client work," with nothing of their own to show either. |
| 2. Tell me about a project that didn't work. What did you learn? | A specific story with a specific lesson, told without blaming the client. | "Honestly, everything we build works." |
| 3. What tools do you build with, and when wouldn't you use them? | They name real tools and explain the trade-offs, including when a simple fix beats AI. | One proprietary platform for every problem, or a cloud of buzzwords. |
Question 3 matters more than it looks. A consultant who only sells one platform will fit every problem to that platform. You want someone who will say "you don't need AI for that, a spreadsheet formula will do."
Ownership and access
This is where small businesses get burned. If the consultant builds everything inside their own accounts, you're renting your own processes, and leaving gets expensive.
| Question | A good answer sounds like | Red flag |
|---|---|---|
| 4. Whose accounts will the work live in? | Yours, or built in theirs and transferred to you on a clear schedule. | Theirs, with no transfer plan. |
| 5. What access do you need, and how will you get it? | The minimum needed, through user invites or separate logins they can be removed from. | "Just send me your password." |
| 6. Where does my data go? | They name every tool your data touches and explain the data settings on each. | A shrug, or "the AI is secure." |
For more on what safe access looks like, see my answer on whether it's safe to give AI access to your business data.

Get the checklist
How quality is checked
AI makes mistakes. That's not a reason to avoid it, but it is a reason to ask how mistakes get caught. The answer should involve a person and a process, not faith in the software.
| Question | A good answer sounds like | Red flag |
|---|---|---|
| 7. Who checks AI output before it reaches me or my customers? | A named person, a review step and clear rules for what never goes out unreviewed. | "The AI is very accurate." |
| 8. What happens when the AI gets something wrong? | They describe a real catch: how it was spotted, fixed and prevented next time. | "It doesn't really make mistakes." |
| 9. How will I know what got done each week? | A written report on a set schedule, and they'll show you a sample. | "We'll check in as needed." |
If you want to understand the kinds of errors to watch for, my answer on what happens when AI makes a mistake goes deeper.
Pricing and lock-in
AI pricing can hide costs in usage fees, software subscriptions and setup charges. Ask for one all-in number and a list of exclusions, in writing.
| Question | A good answer sounds like | Red flag |
|---|---|---|
| 10. What's the total monthly cost, including tools and AI usage? | One number, plus a clear list of anything you pay separately. | A low base fee, then usage charges you can't predict. |
| 11. Is there a setup fee or a minimum term? | A plain answer. Month-to-month or a short initial term. | A 12-month commitment before you've seen any results. |
| 12. What's not included? | A specific list, like your own software subscriptions, ad spend or physical work. | "Everything's included." Nothing is. |
For the record, here's how I answer these for my own fractional AI manager service: it starts at $499 a month, there's no setup fee, it's month-to-month, and the exclusions are your own software subscriptions, ad spend, physical work and anything needing a licensed professional's sign-off. Hold me to the same standard you hold anyone else.
What happens if you leave
Ask this before you sign, while everyone is still friendly. The answers tell you how confident the consultant is that you'll want to stay.
| Question | A good answer sounds like | Red flag |
|---|---|---|
| 13. If I cancel, what do I keep? | Everything: workflows, documents, data, logins and instructions. | "The automations stop when the subscription ends." |
| 14. Is there documentation so someone else could take over? | Yes, plain-language notes on what each workflow does and how to fix it. | "It's all in my head." |
| 15. How much notice do you need, and how does the handoff work? | Thirty days or less, plus a handoff call and a transfer checklist. | Cancellation penalties, or vague answers about "winding down." |
What this looks like in a real small business
Example: picture a 15-person landscaping company interviewing three consultants to automate estimate follow-ups, crew scheduling notices and a weekly revenue report.
The first consultant has a great slide deck and talks about "AI transformation." When the owner asks to see something running, he gets a demo video from a software vendor. That's a pass.
The second consultant shows a real workflow on screen and clearly knows her stuff. But everything gets built in her agency's account, the contract is 12 months and leaving means starting over. The owner asks if the workflows can live in his accounts. She says no. Also a pass, though a closer call.
The third shows a live workflow, explains a time an automation sent a duplicate email and how she fixed it, and says the work lives in the company's accounts from day one. She sends a sample weekly report and a one-page list of what's not included. Month-to-month.
The owner hires the third and starts with one task: the estimate follow-ups. After a month of good weekly reports, he adds the crew notices. Starting small gave him proof before he expanded.
The honest part
No list of questions guarantees a good fit. Someone can answer all fifteen well and still not click with how you work. That's why starting with one or two tasks, rather than a big project, is the safest move with any new consultant.
These questions also favor people with a track record. A newer consultant may have fewer client stories but still be excellent. In that case, lean harder on question 1: what have they built for themselves, and does it actually run?
And be wary of anyone who promises specific results before they've seen your business. Honest consultants say "let's find out" more often than "guaranteed."
One more thing: I wrote this list, and I'd like to think I pass it. That's a bias. Use it on me too.
Quick questions
How much should an AI consultant cost?
It depends on the model. Some charge hourly, some per project, some a monthly retainer or flat fee. Compare all-in monthly cost, including software and AI usage, against the hours of work it takes off your team. My answer on AI automation costs walks through the math.
Do they need experience in my industry?
It helps, but it isn't required. Most small business work that AI handles well, like follow-ups, reports, data entry and content, looks similar across industries. What matters more is whether they ask good questions about how your work actually flows. Industry experience matters most where regulations apply, like healthcare or finance.
Should I ask for references?
Yes, and ask the references specific questions: how fast problems got fixed, whether they got weekly updates, and whether the work lived in their own accounts. A reference who says "they were great" tells you little. A reference who describes how a mistake was handled tells you a lot.

