Hiring an AI Consultant: 12 Questions to Ask Before You Sign

Most companies hiring an AI consultant for the first time are buying something they cannot fully judge yet. Every proposal promises efficiency, a roadmap and quick wins, and most of them use the same words. The difference between a useful engagement and an expensive slide deck usually shows up in the answers to a handful of plain questions asked before anything is signed.

Below are twelve questions grouped by what they test: whether the consultant understands your business, whether they will tell you where AI is not worth it, how they handle your data and risk, what you will actually receive, and what happens after they leave. For each one there is a note on what a good answer sounds like and what should worry you.

Before hiring an AI consultant: know your own problem

You will get better proposals, and be able to compare them, if you prepare three things first:

• A one-paragraph problem statement described as a business result, for example "quotes take three days to go out and we lose deals because of it", not "we want to use AI".
• An internal owner who will make decisions and give the consultant access to people and data.
• The basics: rough volumes, how the process runs today, and where the data lives.

If you cannot answer simple questions about the process from your own data, the first piece of work may be getting the data in order rather than AI. A good consultant will tell you so.

Questions about your business

1. What do you need to see of our operations before recommending anything? A good answer involves time with the people who do the work, watching the process as it actually runs, and looking at real data. A consultant who offers recommendations in the first meeting is guessing.

2. Which of our processes would you not automate, and why? A good consultant names candidates that are not worth doing yet, and explains why: too rare, too risky, or the data is not ready. If everything sounds like an AI opportunity, the analysis has not been done.

3. What have you built that is running in production today? Ask to see live systems, not demos, and if possible speak to someone who uses one. Strategy without implementation experience tends to underestimate integration, data and adoption work.

Questions about the recommendation

4. How will you rank the candidates? Look for clear criteria: business impact, how often the task happens, data readiness, the cost of a wrong output, and the effort to implement. Ranking by how impressive a use case sounds is a warning sign.

5. Will your recommendation be tied to a particular product? Many consultants also sell or resell software. That is not a problem in itself, but it must be disclosed, and the recommendation should say plainly when another option would fit better.

6. What will each recommendation cost to implement and to run? A good answer separates one-off implementation cost from running cost: model usage, integration upkeep, monitoring and the staff time to review outputs. A recommendation without a running cost is incomplete.

Questions about data and risk

7. What data will you need, and where will it be stored and processed? Under the PDPA you remain responsible for personal data you share. Good consultants ask for the minimum, work with masked samples where possible, and can tell you exactly where any data goes.

8. How do you handle wrong outputs from AI? Listen for measurement against real examples, human approval for high-risk actions, and a plan for what happens when the AI is unsure. "The model is very accurate" is not an answer.

9. Who will actually do the work? Ask for the names and roles of the people on your engagement, and how much of their time you get. The senior person in the sales meeting is not always the person who does the analysis.

Questions about the output and what comes after

10. What exactly do we receive at the end, in writing? Expect documents your team can read, question and decide on: a map of how operations run today, a ranked list of candidates including the ones not worth doing, a cost estimate per recommendation, and a recommended order. A presentation alone is not enough.

11. How is the engagement priced, and how long will it run? Both should be written in the quote before work starts. Open-ended hourly engagements without a cap tend to grow. A fixed scope with a fixed duration keeps both sides focused on the decision you need.

12. Can we implement with someone else afterwards? The output should be usable by any vendor you choose. If the consultant also implements, ask how the consulting fee is treated if you continue with them, and make sure you are not locked in if you do not.

Red flags when hiring an AI consultant

• Talking about models and tools before asking about your operations.
• Promising a percentage of savings before seeing any data.
• Every recommendation leads to the consultant's own product, without disclosure.
• A vague deliverable such as "an AI roadmap" with no description of its contents.
• No estimate of running costs.
• Nothing running in production that they can show you.

What a good written output contains

• How the process runs today, including where the hours go and the exceptions.
• Ranked candidates with the reasoning, and the candidates not worth doing yet named plainly.
• A cost estimate to implement and to run each recommendation.
• The recommended order, and what kind of implementation fits each item.
• Prerequisites, such as data that must be cleaned first.
• How success will be measured after implementation.

Our guide on what a fixed-scope AI readiness assessment should deliver goes deeper into each of these documents.

How to compare proposals fairly

Send every consultant the same problem statement and the same basic facts, then score each proposal on five points: understanding of your business, independence, data handling, clarity of the deliverables, and price structure. Ask each shortlisted firm the twelve questions above in the same order, and write the answers down. Differences that were invisible in the proposals usually become obvious in the answers.

At Etzal Group, AI Consulting is a fixed-duration package: we map how your operations run, rank the candidates, estimate the cost of each from our published implementation prices, and hand everything over in writing, including which implementation line fits, if any. We also build our own product, Polynusa Autopilot, and we say so; the package does not require you to use it. The duration and fee are written in the quote before any work starts, and the fee is deducted if you proceed to implementation. The first 30 minutes are free: book a consultation and we will tell you honestly whether the package is worth it for your size.