Choose an AI consultant by testing them on your own problem, not their pitch. Send three candidates the same one-page brief, then ask each the same 20 questions about your processes, the tools they'd pick, who owns what they build, how they handle your data and what happens after handover. Hire the one whose answers are specific and written down.
Most poor hires happen because the owner judged the demo. A demo shows what AI can do in general; the questions below show whether this person understands your business, will tell you when AI isn't the answer, and will leave you able to run what they build. They work on anyone you talk to, independent or agency, and they work on me too. If you're not yet sure you need outside help at all, start with do I need an AI consultant.
Before the first call: a brief that makes answers comparable
Consultants can only give comparable answers to the same question, so write one page and send it to everyone before you talk. Here's an illustrative brief from a farm shop with a café and a weekly veg box scheme:
- The business: farm shop, café and veg box deliveries; 14 staff, about 220 box customers.
- Jobs we'd like help with: box changes (skips, swaps, address changes) arrive by email and are updated by hand in a spreadsheet, about 6 hours a week; the same customer questions about delivery days and box contents arrive by email and social messages; café supplier orders are placed by phone from a paper list.
- Tools: Google Workspace Business Standard, Shopify for the box shop, a card terminal at the till, a Google Sheets stock list.
- Constraints: nobody technical on staff; the shop manager can give two hours a week; summer is the busy season.
- What we want from the conversation: which job to tackle first, what it would involve, and a rough cost.
Keep a copy of the brief next to the scoring sheet at the end of this tutorial. Every answer gets judged against it.
Questions 1 to 4: have they understood how you work?
1. What would you need to see before recommending anything?
A consultant who recommends tools in the first ten minutes is selling, not diagnosing. You want someone who asks to see the work itself. A strong answer sounds like: "I'd like to watch someone process a normal week of box changes, see five real emails and the spreadsheet, and know how many changes you get in your busiest week." A weak answer: "Most farm shops need a chatbot and an email tool; let me show you what I built for another shop." The second consultant may be perfectly capable, but they're answering a question you didn't ask.
2. Which of our jobs would you look at first, and why that one?
This tests how they prioritise. Good consultants weigh volume, time per item, how rule-based the job is, what an error costs and whether the data is available. A strong answer: "Box changes first. Six hours a week, mostly rule-based, and mistakes mean a wrong box on someone's doorstep and a refund. The customer questions come second, because your answers change with the season and someone has to keep them current." A weak answer covers all three jobs at once, usually inside one new platform.
3. Can you walk me through something similar you've set up, including what went wrong?
You're looking for a job with the same shape as yours, such as changes arriving by email and being applied to a list, which matters more than the same industry. Ask for the steps they built, what broke in the first month and what they changed. A strong answer includes a failure and a fix. A weak one offers logos, a percentage with no context, or a slide about "AI transformation". Vetting a consultant's case studies and references covers how to check what you're told.
4. What would make you tell us not to use AI for this?
This is the honesty test, and it deserves double weight. Some jobs are fixed better by a form, a template or a change of process. A strong answer: "If most box changes could be made by customers themselves through a change form on your site, you'd remove the email step altogether and wouldn't need AI to read anything." A consultant who can't name a single situation where they'd advise against AI will find a reason to use it every time.
Questions 5 to 8: how will they choose the tools?
5. What can our current tools already do here?
Many business suites now include AI. Google Workspace Business Standard and above include Gemini across Gmail, Docs, Sheets and Drive; Microsoft 365 business plans include Copilot Chat at no extra cost; HubSpot's free and Starter tiers include its Breeze assistant for writing and summaries; Shopify Inbox includes a free AI agent that answers customers on the Basic plan or above, with new customer accounts switched on. A strong answer starts there: "Your Workspace plan already has Gemini in Gmail, so let's test whether it can draft the delivery-day replies before adding anything." A weak answer recommends a new subscription without asking what you run.
6. Do you earn commission or referral fees from any tool you might recommend?
Partner programmes and referral fees are normal, and a consultant who earns them can still give good advice. What matters is that they tell you, in writing, and that they'll name an alternative. A strong answer: "Yes, I'm a partner with one automation platform and receive a referral fee; if you'd rather avoid that, here's how the job would work on the other main platform." A weak answer is vague, or offended. Checking whether a consultant is independent goes further.
7. What happens if the product you build on is retired or reprices?
This isn't hypothetical. OpenAI's custom GPTs stop running on 11 December 2026, OpenAI closed the Sora app in April 2026 and its API in September, and the calendar tool Clockwise shut down in March 2026 and deleted user data rather than transferring it. A strong answer explains how they keep your business logic portable: prompts and rules written in a document you hold, data in your own spreadsheet or database, mainstream platforms with export options. A weak answer is "that won't happen".
8. What will this cost to run each month at our volumes?
Automation platforms bill by use. Zapier counts every successful action step as a task, and on Make each scheduled check uses a credit even when nothing has arrived. A strong answer arrives with sums: "About 300 box changes a month through five action steps is 1,500 tasks, so you'd need a tier above Zapier Professional's 750-task entry level; I'll price both options." A weak answer is "pennies", which is sometimes true of the AI usage and rarely true of the platform.
Questions 9 to 12: your accounts, your data and the rules
9. Whose accounts will everything be built in?
Give this one double weight. Automations built in a consultant's own accounts stop working, or become hostage, when the relationship ends. A strong answer: "Yours. You create the accounts on your company card, add me as a user, and I hand back admin rights at the end." A weak answer: "It's simpler if I run it from my agency account." Simpler for whom?
10. What access do you need, and how will it be removed afterwards?
Look for a short, specific list and a plan for taking it away. A strong answer names the systems ("editor access to the box spreadsheet, a user on the Zapier account, read access to the orders inbox"), asks for their own login rather than a shared password, expects two-factor authentication to stay on, and puts access removal on the handover checklist. A weak answer asks for your main admin password on the first call.
11. Where will our customer data go, and could any of it train a model?
Double weight again. The answer should name the tools, the plan and the setting. Business plans such as ChatGPT Business and Enterprise, Claude Team and Enterprise, Microsoft 365 Copilot and Gemini in Workspace don't train on business content by default, OpenAI's API doesn't either, and personal plans need the model-training switch turned off in privacy settings. A strong answer adds that they'll sign written terms on how they handle personal data and will use made-up or anonymised examples until those are signed. A weak answer: "It's all secure."
12. Which rules apply to what you'd build for us?
A good consultant knows the obvious rules for the thing they're building and knows when to send you to a lawyer. For example, if a customer-facing chatbot will talk to customers in the EU, the EU AI Act's duty to tell people they're dealing with an AI has applied since 2 August 2026. A strong answer names the rules that plausibly apply, builds in the practical steps (a clear "you're chatting with an AI assistant" line, a route to a person) and says where legal advice is needed. A weak answer: "Compliance is your side."
Questions 13 to 16: scope, price and who does the work
13. What exactly will be delivered, and how will we both know it works?
You want deliverables described as results you can test. A strong answer: "Box changes sent by email get applied to the spreadsheet; we'll test on 40 past emails and it has to get every skip and swap right; anything it can't read goes to the shop manager with the original email attached." A weak answer: "An AI-powered box management solution." Scoping an AI project with deliverables and acceptance criteria shows how to write these yourself if the consultant won't.
14. Is the price fixed, and what counts as a change?
A fixed price for a described scope protects your budget, but only if both sides agree what falls outside it. A strong answer: "Fixed for the scope in the proposal. Adding the café supplier orders would be a change; I'd price it in writing before starting." A weak answer is an hourly estimate with no cap, or a fixed price attached to a scope so vague that everything later becomes a change.
15. Can we speak to a client whose system has been running for six months?
Plenty of automations work on launch day and fail by month three, when a form changes or volumes grow. A strong answer offers a reference who has lived with the work for a while, or, where clients are confidential, an anonymised walkthrough of a live system and what maintenance it has needed. A weak answer offers only launch-day testimonials.
16. Who will actually do the work?
The person on the sales call isn't always the person who builds. That's fine if you know. A strong answer names who does each part and says what, if anything, is subcontracted and whether subcontractors will see your data. A weak answer: "Our team." Ask the follow-up: who will I email when something breaks?
Questions 17 to 20: the day they hand it over
17. What will we own when you finish?
Double weight. Everything built for you, in your accounts, should be yours to keep, change and hand to someone else. Some consultants reuse their own templates across clients, which is reasonable, as long as you get a permanent right to use what's in your system. A strong answer is plain: "You own the workflows, prompts and documentation; my standard templates are licensed to you for good." Who owns the AI workflows a consultant builds explains the traps.
18. What will the handover include?
A strong answer lists it: a plain-language note of how each automation works, a list of every account and connection, the prompts in a document you hold, a recorded walkthrough, and a named person on your staff shown how to make small edits such as changing the delivery days in a reply. A weak answer: "It's all self-explanatory."
19. Who fixes it when it breaks in month four, and at what cost?
Double weight, because automations fail quietly. Zapier pauses a Zap automatically once 95% of its runs have errored over seven days, and Power Automate switches off a flow after 14 days of continuous failure. If nobody is watching, the first sign is an angry customer. A strong answer sets up failure alerts to your inbox, explains what your staff can fix themselves, and prices support now: an hourly rate, an allowance or a small retainer. A weak answer: "Just message me."
20. How will we measure whether this was worth it?
Ask for a baseline before any work starts: hours spent on box changes each week, how many boxes went out wrong last month, how long customer questions wait for a reply. A strong answer suggests measuring the same numbers at 30 and 90 days, and is willing to be judged on them without promising a figure in advance. A weak answer promises a percentage saving before seeing your work.
The farm shop's baseline, agreed on the first call, fitted in four lines:
| Measure | Today | How it's counted |
|---|---|---|
| Staff time on box changes | 6 hours a week | Manager's timesheet, four-week average |
| Wrong boxes delivered | 7 last month | Refunds and complaints logged in the box spreadsheet |
| Changes missed before the packing cut-off | 4 last month | Emails answered after Thursday noon |
| Box changes arriving by email | About 300 a month | Inbox label count |
Scoring three consultants for a farm shop
Score each answer 0 (weak or missing), 1 (acceptable) or 2 (strong). Double the score for questions 4, 9, 11, 17 and 19, the five where a bad answer does lasting damage. The maximum is 50. One rule overrides the total: a 0 on any double-weighted question ends the conversation.
Consultant: __________________ Date: ________ Quote: $________
Group Questions Max Score
Understanding your work 1, 2, 3, 4(x2) 10
Choosing tools 5, 6, 7, 8 8
Accounts, data, rules 9(x2), 10, 11(x2), 12 12
Scope, price, people 13, 14, 15, 16 8
Handover and support 17(x2), 18, 19(x2), 20 12
Total 50 ____
Any 0 on a double-weighted question? Yes / No
Answers still owed in writing: _______________________Here's how the illustrative farm shop's three conversations scored:
| Group (max) | Small agency | Independent consultant | Freelancer |
|---|---|---|---|
| Understanding your work (10) | 5 | 9 | 4 |
| Choosing tools (8) | 4 | 7 | 5 |
| Accounts, data, rules (12) | 10 | 10 | 4 |
| Scope, price, people (8) | 7 | 6 | 5 |
| Handover and support (12) | 7 | 10 | 3 |
| Total (50) | 33 | 42 | 21 |
| Illustrative quote | $9,500 | $4,200 | $1,600 |
The freelancer was cheapest and pleasant to deal with, but planned to build everything in their own automation account (question 9 scored 0), which ends the conversation under the shop's rule. The agency scored well on data and scope but pitched a customer chatbot alongside the box changes, a job the shop hadn't asked for, and couldn't name a time it had advised against AI. The independent consultant scored lowest of the three on scope, because the first proposal gave an hourly estimate. The shop went back with one request: a fixed price for the box-change job with change requests priced in writing. With that settled, the independent consultant got the work at $4,200 against 6 hours a week of manual updating.
Notice what decided it. Not the demo, not the price alone, but a handful of answers about ownership, honesty and what happens after launch. Those are also the answers people most often forget to ask for.
Answers that should end the conversation
A few answers are serious enough to stop at, whatever else the consultant gets right:
- "I'll run it from my account." Your automations, your accounts, or no deal.
- Silence or annoyance about commissions. Disclosure costs them nothing if they have nothing to hide.
- No case where they'd advise against AI. Every job will look like an AI job.
- Guarantees. "100% accurate" or "we'll cut your admin by half" before they've seen your work.
- Asking for customer data before agreeing how it's handled. Real data can wait until the terms are signed.
For the longer list of warning signs, including the subtler ones that show up after you've signed, see AI consultant red flags.
Shifting the weight for an agency, a freelancer or an IT firm
The same 20 questions work on every kind of supplier, but the ones that expose trouble differ by type, so it's worth adding extra weight in your own sheet.
- A small agency: lean on 14 and 16. Agencies price by scope and staff their work across a team, so what counts as a change, and who you'll actually deal with after the sales call, matter more than usual. Ask whether the person who scoped the job will be on it.
- A freelancer from a marketplace: lean on 9, 18 and 19. Freelancers are often excellent builders working alone, which makes their accounts, their documentation and their availability in month four the weak points. Ask what happens if they're ill or booked up when something breaks.
- Your IT support company offering AI work: lean on 1, 2 and 4. IT firms know your accounts and security well, and are usually strong on 9 to 11. Where they can be thinner is process work: watching how a job is done and deciding whether AI should do it at all.
- A software vendor's own implementation team: lean on 5 and 6. Their answer to "which tool?" is predictable, so the useful questions are what their product can't do for you and what it costs at your volumes.
Whichever type you're talking to, keep the override rule: a 0 on any double-weighted question ends the conversation, however good the rest sounded.
Running the questions on a 30-minute call
Twenty questions won't fit comfortably into a first call, and they don't need to. Split them by how they're best answered:
- Ask live, because the way they answer tells you something: 1, 2, 3, 4, 7 and 19. Watch whether they ask you questions back.
- Ask for in writing, because you'll want to hold them to it: 6, 11, 13, 14, 17 and 18.
- Ask in either, whichever comes up: 5, 8, 9, 10, 12, 15, 16 and 20.
Send the brief and the written questions a few days before the call, fill in the scoring sheet during it, and send a short follow-up afterwards: "Thanks for the call. Could you answer questions 6, 11, 13, 14, 17 and 18 in writing with your proposal?" Compare all three within 48 hours, while the conversations are fresh. A consultant who takes the written questions seriously is showing you how they'll handle the documentation later.
One last check before you sign: re-read your brief. The winning consultant should be solving the problem you wrote down, not a bigger one they talked you into. If the proposal has grown, ask why, and ask what the smallest useful first step would cost on its own.
Further reads
- Where to Find a Good AI Consultant for a Small Business — Where to find candidates worth putting these questions to.
- AI Consultant Handover Checklist: What You Need Before They Leave — The full list of what to collect before a consultant leaves.
- Should You Sign an NDA Before Sharing Data With an AI Consultant? — Whether to sign an NDA before you share business data.
- How to Judge Whether Your AI Consultant Delivered Value — How to judge afterwards whether the work was worth it.
- How to Prepare for an AI Consultation and Leave With a Plan — Get ready for the first call so it produces a plan.
- AI Consulting Contracts: 9 Clauses to Check Before You Sign — Turn the answers you liked into contract wording.
- What Does an AI Implementation Consultant Actually Do? — What the work itself involves, stage by stage.
- DIY or Get Help? Setting Up AI Bookings for a Small Business — Six questions that tell you whether to build AI bookings yourself, with three worked cases and a weekend plan if you go it alone.
- What an AI Consultant Does for a Law Firm, and What It Costs — The five phases of a law-firm AI engagement, what each should hand over, how fees are built from consultant days, and proposal red flags.
- AI Consultant or DIY: Which Suits a Small Professional Firm? — When a small professional firm can set up AI itself, when outside help earns its fee, and the split approach many firms settle on.
- Do Small Medical Practices Need an AI Consultant? — When a small medical practice can handle AI itself, the signs outside help will pay for itself, what a consultant shouldn't do, and a scoped brief to adapt.
- When Should a Trades Business Bring In an AI Consultant? — The signals that a trades firm needs outside AI help, a ten-minute scoring sheet, what to gather first and what a good engagement leaves behind.
- Questions to Ask Before You Pay Anyone to Set Up AI Marketing — 30 questions in six groups, with what a good answer sounds like, plus a filled-in comparison of two real-looking quotes for a food truck.
- DIY or Done-With-You: Who Should Set Up Your AI Marketing? — A scoring table for choosing DIY, done-with-you or done-for-you AI marketing, the setup jobs ranked by difficulty, and what a proper handover includes.
- AI Consultant vs Agency vs Freelancer: Which Should You Hire? — Who diagnoses, who builds and who supports: a decision table for consultants, agencies and freelancers, with a coffee roaster's three jobs sorted.
- What Happens in a 1:1 AI Implementation Consultation? — What to send beforehand, how a consultant maps your work and scores AI jobs, the tool check, and the one-page plan you should leave with.
- What Is a Fractional Chief AI Officer and Do You Need One? — What a fractional chief AI officer does each month, what providers publish as prices, and cheaper set-ups that suit most small businesses.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor help and pricing pages for Zapier (task counting, auto-pause rules), Make (credit use), Microsoft Power Automate, Google Workspace with Gemini, HubSpot Breeze, Shopify Inbox; OpenAI help pages on custom GPT retirement and the Sora discontinuation; EU AI Act Article 50 transparency duties (checked September 2026).