To hire an AI automation consultant, shortlist two or three people and ask each the same 12 questions on a first call. Look for proof of a live build, a named person doing the work, value-based pricing in writing, your accounts and code, and a monitoring plan. Hire the one with specific answers and almost no red flags.
Why the Questions Matter More Than the Vendor Type
Whether you search for an AI automation consultant, an AI automation expert, or an agency, you're buying the same thing: someone who can turn a manual process into a system that runs in production. The title doesn't predict the outcome. The answers do.
The stakes are real. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing cost, unclear business value, and weak risk controls. MIT's NANDA research, reported by Forbes, found 95% of enterprise generative AI pilots showed no measurable P&L impact, while projects done with an outside partner succeeded about 67% of the time against roughly 33% for internal builds. Who you hire, and how you vet them, moves those odds.
If you're still deciding whether you need outside help at all, start with what an AI automation consultant actually does. If you're choosing between vendor types, see agency vs freelancer vs in-house. This post assumes you've decided to hire and need to pick well.
The 12 Questions at a Glance
| # | Question | Good answer | Red flag |
|---|---|---|---|
| 1 | Can I see a live build? | A running system, plus a story about what broke | Slides, mockups, or "it's confidential" for everything |
| 2 | What won't you automate? | A specific list for your business | "We can automate anything" |
| 3 | What does this cost us today? | Asks for your numbers before naming a price | A fixed quote before mapping the process |
| 4 | Who does the work? | A named person you'll talk to directly | An account manager, builder unnamed |
| 5 | What happens when it breaks? | Monitoring, alerts, a human escalation path | "It won't break" |
| 6 | Who owns the code and accounts? | You do, in your accounts, in the contract | "It runs on our platform" |
| 7 | Where does our data go? | Named providers and their training terms | Vague, or "we handle it" |
| 8 | Agent or rules? | Rules where rules work, AI only where they break | AI for everything, or can't explain the split |
| 9 | How do you price? | Tied to the value of the outcome, in writing | Open-ended hourly, or "contact us" forever |
| 10 | What's the monthly run cost? | Itemized estimate with a volume assumption | Not mentioned until the first bill |
| 11 | How do milestones and payments work? | Paid at signing plus objective, testable milestones | 100% upfront, or "when it works as expected" |
| 12 | What does handoff include? | Docs, runbook, code, and a defined support model | Nothing written down |
Questions 1 to 3: Can They Actually Build It?
1. "Can I see something you built that's running in production right now?"
Why it matters: demos run on ten clean examples. Production runs on the 400 messy ones that arrive in a real month. Building something that survives that is the skill you're paying for.
Good answer: a live system they can walk through, or a client you can contact, plus a specific story about something that broke after launch and how they fixed it.
Red flag: only slides, a polished demo video, or a blanket "all our work is under NDA." Some work is confidential. All of it isn't.
2. "What would you refuse to automate in our business?"
Why it matters: a consultant who knows the limits of AI will name them. A salesperson won't.
Good answer: a specific list. For example: anything that releases money or signs off on compliance without a person approving it, processes that change every month, and tasks that happen a few times a year.
Red flag: "We can automate anything." Some work should stay manual, and some should get a better process before it gets any software.
3. "What does this problem cost us today, and how will we measure the result?"
Why it matters: if nobody measures a baseline, nobody can say whether the project worked, and it gets cut in the next budget review.
Good answer: they ask you for volume, minutes per item, error rate, and who does the work before they talk price. They agree on one or two numbers to measure at 30 and 60 days. The ROI calculator and how to measure AI automation ROI show the method.
Red flag: a fixed quote in the first ten minutes. A reference range is fine. A firm price before anyone has mapped the process is a guess.
Questions 4 to 6: Who Is Accountable?
4. "Who exactly does the work, and who do I talk to?"
Why it matters: in many shops, the senior person sells the project and someone else builds it. Every message relayed through an account manager loses detail.
Good answer: a named person who scopes it, builds it, and answers your messages.
Red flag: they can't tell you who the builder is, or the work is subcontracted without your knowledge.
5. "What happens when it breaks?"
Why it matters: automations break when an API changes, a form field moves, or volume spikes. The question is whether you find out from an alert or from a customer.
Good answer: monitoring and error handling built into the workflow, alerts to a named person, a human escalation path for anything the system isn't sure about, and a supervised rollout where a person approves output before it runs alone.
Red flag: "It won't break." Everything breaks eventually.
6. "Who owns the code, prompts, and accounts?"
Why it matters: if the system lives in the consultant's accounts, firing them means starting over.
Good answer: everything runs in your accounts (model provider, hosting, automation platform), and the contract assigns the custom work to you.
Red flag: "It runs on our platform" with a monthly license and no export path.
Questions 7 and 8: Data and Technical Judgment
7. "Where does our data go, and is it used for training?"
Why it matters: your customer records, invoices, and email pass through whatever model the workflow calls.
Good answer: named model providers and their terms. Both Anthropic and OpenAI state that API data isn't used to train their models by default. Ask for the consultant's own data-handling terms in writing too, and check how they apply to US and Canadian privacy rules if you serve both markets.
Red flag: "Don't worry, we handle security," with nothing specific behind it.
8. "Which parts of this are an AI agent, and which are plain rules?"
Why it matters: Gartner estimates only about 130 of the thousands of vendors selling agentic AI offer genuine agentic capability, and calls the rest "agent washing" (Gartner, 2025). Rules are cheaper and more predictable. AI earns its cost only where the input is messy.
Good answer: a clear split. For example, "routing by customer ID is a rule, reading a free-text supplier reply is the AI step." My view of the difference is in specialized agents vs. chatbots.
Red flag: AI on every step, or they can't explain what the model decides.
Questions 9 to 11: Money
9. "How do you price this?"
Why it matters: the pricing model tells you what the consultant is optimizing for. Hourly rewards slow work. Value-based pricing ties the fee to the result.
Good answer: a price set after discovery, tied to the value the workflow returns, often with a few scope options to choose from. If you say the price is too high, they reduce scope rather than discount the same work.
Red flag: an open-ended hourly engagement with no cap, or "contact us for pricing" that never turns into a number.
10. "What will it cost to run each month after launch?"
Why it matters: the build fee is one number. Model usage, hosting, and tool subscriptions are another, and they recur.
Good answer: an itemized monthly estimate with the volume assumption written next to it, billed to your own accounts. How much AI automation costs shows the split between build and run cost.
Red flag: run costs aren't mentioned, or they're bundled into a markup you can't see.
11. "How are milestones and payments structured?"
Why it matters: vague milestones lead to arguments about whether the work is done.
Good answer: a payment at signing, then payments tied to objective milestones you can test. For example: "When the owner submits a request, the system pulls from the agreed database and responds within one minute."
Red flag: 100% upfront, or a milestone that reads "the agent works as expected."
Question 12: After Launch
12. "What does handoff include, and what does support cost after?"
Why it matters: the system has to keep working after the consultant moves on to their next client.
Good answer: the workflows and prompts in your repository, an architecture document, a runbook your team can follow, and a clear split between maintenance (keeping the agreed system working) and new work (new workflows, which are priced separately).
Red flag: no documentation, or "unlimited support" that's undefined.
What Should an AI Automation Consultant Cost? (USD, 2026)
Price follows value, not hours. A good consultant first works out with you what the problem costs the business each year, then prices the build as a fraction of that number. Here's where those value-based numbers typically land for small and mid-sized businesses in the US and Canada:
| Engagement | Typical range (USD) |
|---|---|
| Automation audit and scoping | Often free |
| Single production workflow | $5,000–$12,000 |
| Multi-workflow system (3 to 5 connected) | $15,000–$35,000 |
| Ongoing retainer | $1,500–$6,000 per month |
These are reference ranges, not quotes. A workflow that saves a few hours a month won't justify the low end. A workflow that removes a hire may justify more. Model usage and hosting are billed separately to your own accounts. The pricing page explains how I set a number, and how much AI automation costs breaks down what moves a quote up or down.
Want to test these questions on me?
Book a free audit and ask all 12. I'll answer each one in writing, and tell you which workflow has the clearest payback.
How to Score the Answers
After two or three calls, count the red flags per candidate.
- Zero or one: a strong candidate. Ask for a written scope with options.
- Two: ask follow-up questions in writing. Sometimes a red flag is a communication gap.
- Three or more: walk away, however good the demo looked.
Also walk away if they refuse any discovery but want a fixed quote, if nobody on your side owns the decision, or if they promise revenue they can't control.
How Smart AI Workspace Answers These 12 Questions
I'm Tariq Osmani, and I run Smart AI Workspace as a founder-led practice. The person answering these questions is the person who builds your system.
Every engagement starts with a free audit where I map the process and work out what it costs you today. You get a written proposal with scope options, objective milestones, and an estimated monthly run cost before anything gets built. The system runs in your accounts, you own the workflows and prompts, and you get a runbook at handoff. I'll also tell you what I won't automate, and if a different kind of vendor fits better, I'll say so. See what I build and how I price.
Get Your First Workflow Scoped
If you're about to hire an AI automation consultant, put these 12 questions to me on a call. Contact me for a free audit, and I'll tell you which workflow to start with and what it will cost in USD. You can also see what I build, how I price, or check verified work history on my Upwork profile.
Sources: Gartner: Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 · Forbes: MIT NANDA, Why 95% of AI Pilots Fail · Anthropic Privacy Center: Is my data used for model training? · OpenAI: Data controls in the OpenAI platform
