How to hire an AI consultant: a 12-question checklist
A step-by-step guide: what to ask, how to verify cases, which red flags you shouldn't ignore. From a consultant who turns down 30% of clients himself.
Andrew Maryasov, founder of Auspex and Grow2.ai, wrote this guide after 28 AI projects and dozens of discovery sessions with businesses of every size. This is the very checklist clients receive after the first call.
What an AI consultant is (and how it differs from an agency)
An AI consultant is a person (less often a team) who helps a business:
- Find the processes worth (and not worth) automating with AI.
- Design a solution for the specific business context.
- Launch a pilot, measure ROI, and decide on scaling.
This is not the same as:
- An AI agency — sells delivery (often without discovery). Takes on volume.
- A trainer — teaches the team but does not implement.
- An AI developer — writes code against a finished spec.
A good AI consultant often persuades the client not to implement AI where a decent CRM or a script is enough.
12 questions to ask before hiring
1. “Show me a case that failed”
If a consultant says “everything we do works” — that’s a red flag. The real share of successful AI projects in SMB is 60–70%. If a person can’t name their failure, they’re either a beginner or dishonest.
A good answer sounds like this:
“In 2024 we built an AI agent for a distributor’s B2B support. Accuracy in tests was 89%, in production it dropped to 71% — clients wrote in slang and abbreviations. We rolled back, rebuilt the RAG with domain-specific embeddings, and accuracy rose to 93%. But during discovery I underestimated the complexity of the domain language — that was my mistake.”
2. “Who owns the code and prompts after the project?”
The canonical answer: you, the client. Everything. If a consultant offers “we maintain the infrastructure on our side” without handover — that’s vendor lock-in. In 12 months you’ll be a hostage.
3. “How much discovery time before the first commit to code?”
If it’s less than 2 weeks, they most likely cut the analysis. In my practice, discovery takes 30–40% of the whole project.
4. “How do you measure accuracy?”
Red flag: “we test manually on 20 queries.”
The right way: regression tests on 200+ real dialogues that run on every prompt update. Without this, accuracy degrades silently.
5. “What do you do about hallucinations?”
The expected answer is three levels:
- A supervisor agent checks facts against the knowledge base.
- A system prompt: “if you don’t know, hand off to a human.”
- Regression tests after every change.
If you hear “we use the best model and there are almost no hallucinations” — this person has never run AI in production.
6. “What’s the minimum contract?”
If a consultant is ready to come in for 1 week — that’s a bad sign. A pilot for an AI agent is a minimum of 2–3 weeks. Standard is 6–10 weeks.
7. “What happens if the metrics don’t reach what was promised?”
The contract should have a concrete clause. For example: “If 30 days after launch the metric is < 50% of the target value, we refund 50% of the payment or do a free iteration.”
8. “Which CRMs/integrations have you done?”
If you have Bitrix24 or 1С, the question is critical. Most “AI consultants” from the US and EU have never touched these systems. Look for someone with regional experience.
9. “Who’s behind you?”
A solo consultant is fine if they have a transparent portfolio. But ask about backup: what happens if the person falls ill in the middle of the project?
10. “Show me an architecture diagram of one project”
A good consultant will draw a box-and-line in 2 minutes. If a person starts with “it depends” — dig deeper.
11. “Which stack do you use and why?”
Red flag: “we use the newest technology” (the person picks what’s trending, not what’s fit-for-purpose).
A good answer: “In most cases — Claude Sonnet 4.6 + pgvector + n8n. It’s a mature stack. If a project needs real-time or on-premise, we switch. I don’t like changing the stack for the sake of novelty.”
12. “How do you handle NDAs and confidentiality?”
Canonically: the NDA is signed before discovery, data is processed in EU regions, and there’s an on-premise option for critical cases.
How much an AI consultant costs (2026 benchmarks)
| Tier | Scenario | Budget |
|---|---|---|
| Pilot | 1 process, 2–3 weeks, MVP agent | $2,000 – $5,000 |
| Standard | 2–3 processes, 6–10 weeks, production | $8,000 – $15,000 |
| Enterprise | 5+ processes, 3–6 months, multi-agent | $25,000+ |
If someone offers “an AI agent for $500 turnkey” — that’s a templated chatbot, not consulting.
Red flags
- Can’t name concrete figures from past cases.
- Doesn’t show the architecture.
- Says “AI will solve everything” (a good consultant often turns you down).
- Doesn’t ask about metrics during discovery.
- Works without an NDA.
- Wants a prepayment of >50%.
- Refuses to put success criteria in the contract.
What an ideal discovery call looks like
30–60 minutes. The consultant asks you:
- “Which 3 processes take up the most of the team’s time?”
- “How much does an hour of work cost for the specialist who handles these processes?”
- “What have you already tried, and why didn’t it take off?”
- “What budget are you ready to allocate for a pilot?”
- “What result, in numbers, would you call a success after 90 days?”
If a consultant sells during discovery instead of listening — that’s not a consultant, that’s a salesperson.
Source authority
Andrew Maryasov is an AI practitioner, founder of Auspex (CRM) and Grow2.ai (AI agents). 28 AI projects across 14 industries and 7 countries, 25+ years in business automation. He often turns down prospective clients during discovery — recommending they set up their CRM properly first.
FAQ
How much does it cost to hire an AI consultant in Ukraine?
A discovery session is usually free (30 minutes). Market benchmarks among providers: a pilot project — from $2,000 to $5,000, a standard implementation — $8,000–$15,000, enterprise — from $25,000. Strategic consulting is separate — see the ranges on consultants’ pages (mine are at /en/advisory/).
Do I need an AI consultant if I don’t have a technical team?
Yes — that’s an even better scenario. The consultant designs the solution so your operations team can use it without developers. Knowledge transfer is included in the project.
How do I know an AI consultant isn’t a fraud?
Check 5 things: 1) Cases with figures and client contacts, 2) Architecture diagrams, 3) An NDA before discovery, 4) Success criteria locked into the contract, 5) Willingness to turn you down if AI won’t deliver ROI.
How long does a typical AI project take?
Pilot — 2–3 weeks. Standard — 6–10 weeks. Enterprise — 3–6 months.
Can I hire an AI consultant at an hourly rate?
You can — for small tasks (an audit, training the team). For implementing an agent it’s better to use fixed-price with clearly defined deliverables.
A letter on AI — 2–3 emails a month
Personal essays on how AI changes the way we think and work: what I tried myself, what worked and what didn't. At the end, links to the best of what I read.
No spam. Unsubscribe in one click.