Before signing a contract with an AI consultant, ask these 12 questions: who will actually work on the project, what experience they have in your industry, how they measure ROI, who owns the data and the code, and what happens if the project doesn't work. The answers will tell you whether you're talking to an expert or a salesperson.
Why the right questions are worth more than any demo
The AI consulting market has exploded. In 2026, dozens of firms — from specialized boutiques to the Big4 — offer "AI transformation" services to businesses. The problem: quality varies enormously, and an entrepreneur without specific technical skills struggles to tell who can actually implement AI from who can only sell it.
The questions you ask before signing are your most effective filter. You don't need technical skills to ask them — you just need to know what to ask and how to interpret the answers.
This checklist was born from hands-on experience with dozens of SMEs who chose (and in some cases, got burned by) their AI partner. Every question has a specific reason behind it and a "warning sign" to help you evaluate the answer.
The 12 essential questions
1. "Can you show me a real case similar to mine, with verifiable numbers?"
Why it matters: The difference between an experienced consultant and a beginner is real cases. Not demos, not internal proof of concepts — actual implementations, with real clients, that you can verify.
Warning sign: If the consultant only shows generic demos, slides with vague percentages ("up to 90% efficiency"), or refuses to provide verifiable references, it's a serious red flag. A serious professional has at least 3-5 documented cases with real metrics.
Ideal answer: "We implemented a similar system for [verifiable company] in the [your] sector. The 6-month results were [specific numbers]. I can connect you with their project lead."
2. "What is your methodology? How do you structure the project?"
Why it matters: A structured approach (audit, POC, validation, scaling) is the mark of professional maturity. Someone who improvises the path while selling it doesn't know where they're going.
Warning sign: Vague answers like "it depends on the project" without a reference framework. Or, conversely, a rigid approach that doesn't adapt to your specific reality.
Ideal answer: A clear 3-5 phase method with defined deliverables for each, indicative timelines, and decision checkpoints. Exactly like the method we use at IL DOGE DI VENEZIA.
3. "What happens if the POC doesn't work?"
Why it matters: Not all AI projects succeed. A consultant's maturity is measured by their transparency on this point and the backup plan they have in place.
Warning sign: "It's never happened" or "our projects always work." Anyone with real experience knows that some POCs fail, and that's normal.
Ideal answer: "If the POC doesn't hit the agreed KPIs, we analyze the causes and propose a pivot to an alternative use case. The POC cost is kept low precisely to manage this risk."
4. "Who will do the actual work? Will I see the same people from pre-sales?"
Why it matters: In the consulting world, the sales team and the delivery team are often different people. You want to know who will actually work on your business and what experience they have.
Warning sign: Evasive answers about who will be on the project team, or the certainty that the senior person at the meeting won't be involved in implementation.
5. "What technologies and AI models will you use?"
Why it matters: Not to evaluate the technical choice (that's not your role), but to understand whether the consultant has specific expertise or is selling something they don't yet know how to do.
Warning sign: Anyone who only mentions "ChatGPT" probably has superficial skills. The AI ecosystem is broad: Claude, GPT-4, open-source models, computer vision, NLP — an experienced consultant knows when to use what.
6. "How will we measure success? What KPIs?"
Why it matters: If you don't define KPIs before starting, you'll never know if the project worked. KPIs must be specific, measurable, and agreed upon in advance.
Warning sign: Vague KPIs like "improved efficiency" without numbers. Or technical KPIs (model accuracy) without connection to business metrics (hours saved, errors reduced, incremental revenue).
7. "What access to company data do you need? How will you protect it?"
Why it matters: AI needs data to work. You need to understand what data will be required, who will have access, where it will be processed, and how it will be protected.
Warning sign: Blanket requests for access to all data, no data handling policy, use of APIs without zero-data-retention, no mention of GDPR.
8. "Who owns the intellectual property of the code and models?"
Why it matters: If the consultant leaves, the code and models must stay yours. This point needs to be defined contractually before starting — not after. For contractual details, also read our guide to essential clauses in an AI consulting contract.
Warning sign: Clauses that tie IP to the consultant, proprietary licenses that create dependency, code that isn't accessible or documented.
9. "How will you handle change management with our team?"
Why it matters: Technology is never the main problem. Team adoption is. A consultant who ignores change management will deliver a system that nobody uses.
Warning sign: No training plan, no mention of involving end users, exclusive focus on technology.
10. "What does post-implementation support include?"
Why it matters: An AI system isn't "deploy and forget." Performance degrades over time (data drift), business processes change, unforeseen edge cases emerge. You need a maintenance plan.
Warning sign: "The project ends at go-live" or post-launch support that's expensive and not defined upfront. A serious partner includes at least 1-3 months of support in the price.
11. "What is the total cost, including recurring costs?"
Why it matters: The project cost is only part of the picture. Recurring costs — AI APIs, cloud infrastructure, maintenance, licenses — can be significant and must be clear from day 1.
Warning sign: No estimate of monthly operational costs, or unrealistically low estimates. A production AI system has real recurring costs that need to be planned for.
12. "Can you provide 2-3 references from clients with similar projects?"
Why it matters: References are the final test. A consultant with a real track record has no problem providing contacts of satisfied clients.
Warning sign: Refusal to provide references, only generic references (no company names), or references only from projects different from yours.
Want to apply this in your business?
At IL DOGE DI VENEZIA we support Italian SMEs through every phase of AI transformation. The first conversation is free.
Tell us about your projectHow to use this checklist in your next meeting
You don't have to ask all 12 questions in a single meeting. Here's a practical approach:
- First meeting (getting to know each other): Questions 1, 2, 5, 6 — to understand if it's worth digging deeper.
- Second meeting (deep dive): Questions 3, 4, 7, 9 — to evaluate the approach and seriousness.
- Pre-signing: Questions 8, 10, 11, 12 — the contractual and operational details.
Patterns that distinguish the best AI consultants
After seeing dozens of AI engagements in SMEs, here are the recurring patterns of consultants who deliver real results:
- They always start with an audit, not a demo. First they understand your processes, then they propose solutions.
- They propose a contained POC before a big project. No 100K+ commitment without having validated the approach with 15-30K.
- They're transparent about AI's limitations. They'll also tell you when AI isn't the right solution for a given process.
- They invest in change management. Training, workshops with the team, and adoption support are part of the project, not optional extras.
- They transfer knowledge. The goal is to make you self-sufficient, not to create perpetual dependency.
For an in-depth comparison of different types of AI partners (boutique, Big4, freelancer), check out our guide to choosing an AI consulting firm.
When things go wrong: early warning signs
Sometimes, despite the right questions, the project starts and shows signs of trouble. Here are the early signals that indicate problems:
- The assigned team changes after kickoff without notice.
- Milestones slip without clear explanation.
- Progress reports are vague and don't include agreed-upon metrics.
- The consultant proposes significant scope changes with unplanned additional costs.
- Your team isn't involved in technical decisions that affect them.
If you recognize 2 or more of these signals, it's time for a frank conversation with the consultant. To understand the most common causes of failure, also read why AI projects fail in SMEs.
The next step
Now you have the tools to evaluate any AI consultant. Use this checklist in your next meeting and watch how they react to the questions — the answers will tell you a lot, but so will the way they answer.
If you want to talk to a team that has no problem answering any of these questions, book a conversation with IL DOGE DI VENEZIA. The first call is free.