Questions about independence
1. Which vendors pay you, and how? Referral fees, partner tiers and resale margins are all legitimate and all shape advice. You want to know before you hear the recommendation. A good answer names them. A bad answer is “we are vendor-agnostic” with no detail.
2. When did you last recommend that a client not do an AI project? A consultancy that has never said no has never done a readiness assessment honestly. Ask for the story, without the client’s name.
3. If our existing platform can do this, will you tell us? Much of what businesses are sold as a build is configuration of something they already licence. The right answer is yes, and it costs you less.
Questions about method
4. What do you produce before you build anything? You are listening for a design: what the AI may know, what it may do, where a person approves, how it connects, what is recorded. An answer that goes straight from “discovery” to “build” means the design happens in the code, where you cannot see it.
5. How do you decide what the AI may do without a person? This is the question that separates a designed agent from a demo. A good consultant has a way of drawing the line, draws it conservatively first, and widens it on evidence from the record.
6. How will we know whether it worked? They should ask you to define done in a sentence before starting, and measure something true and small: hours, errors, days. If they offer you a percentage of anything, ask where the number will come from.
Questions about risk and governance
7. What happens to our data, and where does the model run? The answer should cover account types, training opt-outs, storage location and retention, in plain words. If the consultant does not know for the tools they are recommending, they have not assessed them.
8. Are you certified against ISO/IEC 42001, or aligned to it? Almost every consultancy is aligned, not certified, and the honest ones say so. This site says so. An answer that implies a certification the firm does not hold tells you how the rest of the engagement will go.
9. What have you seen go wrong, and what did you change? Every consultancy that has shipped anything has a story. The useful ones involve permissions, approval limits and missing records. A consultant with no story has no experience or no candour.
Questions about what you keep
10. Who owns the architecture and the documentation at the end? You should. In a form another consultant could pick up. If the answer involves their proprietary platform or a retainer to keep access, the price of leaving is part of the price.
11. What does running it cost after you leave, and who do we call? Model usage, monitoring, changes when the business changes. A consultancy that has thought about year two will answer immediately; one that has only thought about the build will change the subject.
Answers that should end the conversation
- A fixed price for a build before any design exists.
- A promise about outcomes expressed as a number the consultant cannot source.
- “Our AI” where the AI is somebody else’s model with a logo on it.
- No mention of approval rules, records or permissions when you asked about risk.
- A recommendation made before anyone has asked what you are trying to do.
Ask us the same questions
Our answers, briefly: vendor relationships are disclosed in writing before any recommendation; we have told clients not to proceed and we will tell you; we design before we build and you own the design; the line between what the AI does alone and what a person approves is drawn in that design; we are aligned to ISO/IEC 42001 and the NIST AI Risk Management Framework and not certified; and year two is a monthly service you can see the shape of on the Packages page.
Consultancy, agency or reseller: telling them apart in five minutes
Ask what they would do in the first two weeks. A consultancy describes understanding your business, looking at your data and permissions, and producing a design with the option of not proceeding. An agency describes starting the build, because the build is what it sells. A reseller describes the product, because the product is what it sells, and the consulting is the wrapper. None of the three is dishonest; they are different businesses, and only one of them is paid to tell you what not to buy.
Then ask who on their side has designed and run something like your use case, and whether that person will be in the room. A firm that sends a salesperson to the first meeting and an engineer to the last has two different views of your project, and you will hear the optimistic one first.
References, without names
You will not get client names from a firm that takes confidentiality seriously, and you should be wary of one that hands them over freely. Ask instead for a description of an engagement in your sector or with your kind of use case: what the problem was, what was built, what went wrong, what it costs to run now. A firm with real experience can do that in two minutes without breaching anything. A firm without it will offer you a logo wall.
Where to go from here
If you are choosing a consultant to help you decide what to do, that is the AI Strategy service on this site: an opportunity map, a ranked list and a business case for the first item. If you would rather test us before a meeting, the Contact page gets you thirty minutes with an AI architect and a written note afterwards, and the AI Opportunity Score below is the five-minute version.
Published 12 September 2026 · Be AI