You’re about to pay someone. Ask first.
Hiring a sub is second nature to you. You wouldn’t hand a crew keys to a job without knowing the scope, the schedule, who’s showing up, and what happens if something goes wrong. But when it comes to AI consultants, most contractors skip the vetting and go on vibes — the consultant sounded smart, the demo looked slick, the price felt about right.
That’s a problem because the AI services market has no licensing, no standard contracts, and no inspector checking the work. A bad hire doesn’t just waste money — it can leave your customer data scattered across tools you don’t control and your office trained on nothing. The fix is cheap: seven questions, asked before you sign anything.
Question 1: What exactly do I get — deliverables in writing?
Why it matters: In construction, scope is everything — you’d never sign a contract that just said “plumbing.” AI proposals get away with the equivalent constantly: “AI implementation,” “automation setup,” “optimization.” Those words describe a category, not a thing you’ll own.
A good answer sounds like:
- “You’ll get an automated review-request system: a text goes to every customer 24 hours after the job closes, and your office manager gets a weekly summary. Live by the 15th.”
- Specific things, specific dates, named in the proposal — not in a sales call you’ll forget.
A worrying answer sounds like:
- “We’ll assess your workflows and implement AI solutions tailored to your business.”
- Anything you’d have to explain to your foreman three times and still couldn’t.
Question 2: Who does the work — you, or subcontractors I’ve never met?
Why it matters: You sub work out yourself, so there’s nothing wrong with a consultant using help. The problem is when the person who impressed you in the pitch disappears and strangers touch your customer data and your systems. You wouldn’t accept a GC who won’t tell you who’s on site.
A good answer sounds like: “I do the setup myself. My assistant handles documentation. Here’s who’s who.” Names or roles, named upfront, with one accountable person.
A worrying answer sounds like: “We have a team.” Which team? Doing what? With access to what? Deflection is not an answer to a normal question.
Question 3: What happens to my data — where does it go, who sees it, can I get it back?
Why it matters: AI tools run on data — your customer list, job history, pricing, employee info. A build might move any of that into systems you’ve never heard of. If you part ways with the consultant, you need to take your data with you and revoke everyone else’s access. This is the question contractors ask least and regret most.
A good answer sounds like: “Your data lives in your accounts — your Google, your CRM — not mine. I’ll need temporary access to set things up, and I’ll document exactly what I touched. When we’re done, you revoke my access and keep everything.”
A worrying answer sounds like: “Don’t worry, it’s all secure in the cloud.” No passwords in email, no shared logins you can’t change — a consultant who shrugs at access control shouldn’t get the keys.
Question 4: Is training included — will my team actually be able to run this after you leave?
Why it matters: A system your people can’t operate is a very expensive demo. The most common failure in small-business AI projects isn’t the technology — it’s that nobody on the team learned to use it, so it dies the week the consultant leaves. Training is what makes the thing you bought actually yours.
A good answer sounds like: “Two hands-on sessions with your dispatcher and office manager, using your real data, plus a one-page cheat sheet for each workflow.” Named people, named format, written down.
A worrying answer sounds like: “It’s very intuitive, your team will pick it up.” If training isn’t in the proposal, it isn’t included.
Question 5: What support comes after handoff — and for how long, at what cost?
Why it matters: Software changes — tools update, your business adds a service line, a key employee leaves. A build with no support plan has a quiet expiration date. You need to know how long the consultant is on the hook after go-live, what help costs after that, and what counts as a fix versus new work.
A good answer sounds like: “Thirty days of fixes included — if something I built breaks, I fix it free. After that, support is $X/month, and new features are scoped separately.” A clear line between warranty and new work.
A worrying answer sounds like: “Call me anytime, I’ll take care of you.” Verbal promises with no timeframe and no price. “Anytime” lasts until the first invoice.
Question 6: How is pricing structured — flat, hourly, and what triggers extra charges?
Why it matters: Surprises on the invoice are how good projects turn into bad relationships. Flat vs. hourly matters less than clarity: you need the total you’re committing to, what counts as “in scope,” and exactly what triggers an extra charge — more users, more workflows, data cleanup, extra training.
A good answer sounds like: “$6,500 flat for the two workflows we scoped. If you add a third mid-project, I’ll quote it separately before starting.” Triggers named before they trigger.
A worrying answer sounds like: “We’ll figure out the details as we go.” Open-ended hourly with no cap and no milestones is how a $3,000 project becomes $9,000.
Question 7: Can you show me proof — past work, references, or a small paid pilot?
Why it matters: Anyone can sound competent for an hour. Proof is what separates a consultant from a salesperson: past projects you can look at, references you can call, or — when the field is too new for a track record — a small paid pilot where they prove it on your actual problem before you commit.
A good answer sounds like: “Here’s a similar build I did for a shop like yours — talk to the owner. Or we can do a two-week pilot on your review requests for $X, and if you don’t like it, you keep what we built.” References you can contact, or skin in the game.
A worrying answer sounds like: “All my clients are under NDA.” All of them? No references, no pilot option, full payment upfront — that’s you carrying all the risk.
A picture of how this plays out
Picture a typical 6-person HVAC company. The owner is choosing between two consultants. The first quotes $4,000 for “AI-powered customer engagement” — no deliverables listed, training “available,” support “ongoing.” The second quotes $5,500: named system (after-hours text-back plus automated review requests), live date, two training sessions with the dispatcher, 30 days of fixes, data stays in the company’s accounts, and two references to call. The owner calls the references, asks all seven questions, and picks the second one. Not because it was cheaper — it wasn’t — but because it was the only one he could actually understand. That’s what the questions buy you: not the lowest price, but the one that survives contact with your actual business.
Turn these questions into your own interview script
Copy this into ChatGPT, Claude, or any similar tool. Fill in your details, and it will customize the seven questions into a short script you can use on a call — or paste a proposal and have it scored against all seven.
You are helping a small contractor interview an AI consultant. Be direct — plain language, no jargon, no hype.
My trade: [YOUR TRADE, e.g., plumbing, roofing, HVAC]
Size of my business: [NUMBER OF EMPLOYEES]
The problem I want AI help with: [ONE OR TWO SENTENCES]
My budget range, if I have one: [OPTIONAL]
Do ONE of the following, based on what I paste below:
OPTION A — If I describe my situation only (no proposal pasted):
Turn the 7 questions below into a short interview script for a sales call. For each question, give me the exact wording to ask (one or two sentences, contractor-plain), plus one follow-up question in case the answer is vague. Short enough to get through in 20 minutes.
OPTION B — If I paste a consultant's proposal below:
Score it against each of the 7 questions: PASS (clearly answered), WEAK (partially — say what's missing), or FAIL (not addressed). Then list the exact questions to send back, in plain language.
The 7 questions:
1. What exactly do I get — deliverables in writing?
2. Who does the work — the consultant, or subcontractors I've never met?
3. What happens to my data — where does it go, who sees it, can I get it back?
4. Is training included — will my team run this after the consultant leaves?
5. What support comes after handoff — for how long, at what cost?
6. How is pricing structured — flat, hourly, and what triggers extra charges?
7. Can you show me proof — past work, references, or a small paid pilot?
[PASTE PROPOSAL HERE FOR OPTION B — or leave blank for OPTION A]Before you sign
- Can I name, in one sentence each, what I’m getting and when it’s live?
- Do I know who does the work — and who is accountable if it goes wrong?
- Do I know where my data will live, who can see it, and how I take it back?
- Is training for my actual team named in the proposal — who, how, and when?
- Do I know the support terms after handoff: how long, what it costs, fix vs. new work?
- Is the pricing structure clear — flat or hourly, total commitment, and what triggers extra charges?
- Have I seen proof — references I can call, past work I can look at, or a small paid pilot?
- Did they answer in plain English, without pressure, without guaranteed results, without jargon walls?
- Would I hire this person to run a crew on my jobsite — because that’s the level of trust I’m actually giving them?
That last one is the gut check the other eight support. You’re not buying software; you’re hiring someone to rewire how your business handles information. Vet them like it.
One offer, no pitch
If you want a second opinion before you sign with anyone — us or someone else — we offer a free AI Workflow Review: a 20–30 minute call where we look at how your shop runs and give you a straight read, including whether we’d recommend hiring anyone at all. If it makes sense to go deeper, our AI Audit is $2,500 flat — a written implementation roadmap plus quick wins, fully credited toward a build if you start one within 90 days.