How to Choose an AI Consultant: 8 Questions to Ask

Why the choice matters

AI projects often fail, and rarely because of the tool. The research organization RAND interviewed 65 experienced practitioners and notes that, by some estimates, more than 80 percent of AI projects fail, twice the rate of IT projects without AI. The leading root causes RAND identified:

1. The problem AI is supposed to solve is misunderstood or poorly communicated.
2. There isn't enough data, or the data isn't good enough.
3. The project chases the latest technology instead of a real problem.
4. The infrastructure to manage data and put results into production is missing.
5. The problem is too hard for AI to solve.

Good AI consulting addresses exactly these points. The questions below follow from them.

How to choose an AI consultant: eight questions to ask

1. Do they ask about your work before naming a tool?

A good AI consultant first wants to know which tasks take time, who does them, and where things get stuck. If someone already knows on the first call that you need "a chatbot" or "Copilot," they're more likely selling a tool than solving a problem.

2. Do they suggest a small first step?

Instead of a big AI strategy, start with one workflow that can be finished in a reasonable time and measured, for example in hours saved per week. Then you decide whether to continue.

3. Can they show what they've built?

Slides and case studies say little. Ask for a demo: a real workflow, ideally with sample data, and an explanation of what was hard about it.

4. Are they open about tools and their own interests?

Ask whether the consultant is a vendor partner, earns referral fees, or sells their own platform. None of that is wrong, but you should know. A good consultant builds with what you already have when it fits, for example Microsoft 365 or Google Workspace, and explains why if something new is needed.

5. Do they sort out data and privacy before building?

Which data does the workflow process, where does it live, who can access it, and which AI service sees it? If customer or employee data is involved, you need clear terms with every provider that touches it, and possibly your legal or compliance team. Leaving that until after the build is a risk.

6. Do they define where a person decides?

AI makes mistakes. A good workflow defines up front what runs automatically and where a person reviews and approves, for example before a reply goes to a customer or an invoice is posted.

7. Do you own what gets built?

Clarify who owns the workflows, accounts, and documentation, and whether everything runs in your own accounts. A good test: does the workflow keep running if you end the engagement?

8. Do they think about the people who use it?

A workflow nobody uses is worth nothing. Ask how rollout works, who gets trained, and who your contact is when something changes.

Red flags

• Guaranteed savings or ROI. Results depend on your data and your team. Serious firms state assumptions, not guarantees.
• An AI strategy without a single concrete workflow.
• Only one tool on offer, whatever you need.
• Unclear ongoing costs for licenses, usage, and support.
• No answer on privacy before data starts flowing.

Questions for the first call

1. Which workflows would you look at first, and why?
2. What's the smallest useful first step, and how do we measure it?
3. Which tools would you build with, and which of those do we already have?
4. What data do you need, and where is it processed?
5. Where does a person review in this workflow?
6. What does the first step cost, and what ongoing costs come with it?
7. Who owns what you build?
8. How will you roll it out to our team?

For how costs break down, see How much does AI consulting cost?.

Where aenuma fits

aenuma starts with your idea or your workflows, not with a tool. We build in your environment, for example with Microsoft 365, n8n, or Make, test with real examples from your daily work, and define up front where a person makes the call. Then we roll the workflow out with the people who use it. More about AI consulting at aenuma.

FAQ

Do AI consultants need a certification?

No. There is no required license or certification to call yourself an AI consultant. That makes real work samples and clear answers to the questions above all the more important.

Do we need a consultant from our industry?

Not necessarily. Many workflows, such as answering inquiries or processing invoices, look similar across industries. What matters more is that the consultant takes the time to understand your work.

Does the consultant need to be local?

For most AI projects, no. Assessment, build, and training work well over video calls. An on-site visit can help when many people are involved in one workflow.

How many proposals should we get?

Two or three is usually enough. Compare them using the questions above and the cost blocks, not just the price.

Sources

The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed (RAND Corporation, 2024)

By Joshua Schütt, founder of aenuma. Updated November 2026.