12 Questions to Ask Before Choosing an AI Consultant
Comparing quotes on price is misleading. Ask the right questions instead — and know what an answer reveals.
In short: comparing AI consulting quotes on price is misleading, because quotes usually describe different scopes. The twelve questions below make the scope visible. Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027; most of these questions exist to surface, at selection time, the gaps that lead to those cancellations.
Scope and start
1. What is the first step of the work? Be wary if the answer is something generic like "a needs analysis". What you want is a measurable first step: which processes get examined, who gets interviewed, how long it takes, and what you hold at the end.
2. What does the diagnosis phase produce, and do we keep it if we do not continue? Good consulting delivers the first phase's output independently of the decision to continue. If the report only arrives once the contract is extended, the diagnosis has been positioned as a sales instrument.
3. How do you decide which process to tackle first? What you want is a ranking criterion: impact, cost, and feasibility. "We start with whatever people complain about most" signals intuition in place of measurement.
Outcome and measurement
4. Which metric will we measure success by? The metric has to be written up front and has to be something you accept: a step's duration, its error rate, its handoff count. "Productivity improvement" is not a metric.
5. Who measures the baseline, and when? If the baseline is not measured before work starts, no improvement can be proven afterwards. The answer should be "we measure it together at the start".
6. What happens if there is no improvement? This is the question whose answer says the most. A consultant with a concrete answer is following the outcome; one who deflects it treats responsibility as ending at delivery.
7. What will you report at days 30, 60, and 90? Predefined measurement points are the sign that the outcome will actually be tracked.
Data, security, and governance
8. Which of our data goes where? The answer should be in writing: which data goes to which service, on servers in which country, retained for how long. The word "secure" is not an answer on its own.
9. What do you do if the data has to stay inside our organization? A consultant who cannot give a clear technical answer here is not a fit if you operate in a regulated sector.
10. Who decides where AI decides and where a human stays? The right answer is "together" and "in writing". An automation built without that boundary defined will stop at the first exception, or quietly do the wrong thing.
Handover and dependency
11. What exactly do we hold when the work ends? The list you want: written process definitions, an ownership matrix, a handover document, and a trained team. If all you get is a working system, you will come back to the consultant for every change to it.
12. Will our team be able to change the system without you? This is the only real completion criterion for consulting. "We provide training" is not enough; "we provide training, hand over documentation with it, and check back after four weeks" is.
How to read the answers
You do not need a provider who answers all twelve well. But if no concrete answer comes back from at least one question in each of the three groups, that gap will show up during the project.
The most telling signal when reading answers is this: does the answer describe a method or a product? A consultant describing a method has thought about the process. A provider describing a product is assuming your process will fit that product.
A scope comparison is also worth doing. The price gap between two quotes is usually a scope gap rather than a quality gap: are diagnosis, training, and measurement included or not? A comparison made without adding those three compares two different jobs.
Sources
- Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (June 2025) — escalating costs, unclear business value, and inadequate risk controls sit behind the cancellations.
- McKinsey & Company, The State of AI: Agents, innovation, and transformation (2025) — 88% of organizations use AI, but only 39% can show an effect on earnings.
Closing
The right consultant is revealed by how concrete the answers are, not by a price list. Take these twelve questions into the meeting in writing, note the answers, and compare quotes against those notes.
See our enterprise AI consulting service or look at the AI transformation data.
Let's find out where your organization stands
In a free 30-minute discovery call, we'll talk through your current state and the right first move.
Book a Call