Before you sign
Choosing an AI consultant: the questions to ask, and the answers to listen for
Use this list on everyone you talk to, including us. A consultant worth hiring will be glad you asked.
Part one: about the work
- 1. “What would you need to see before recommending anything?”
- Listen for: real examples of the work — sample emails, forms, documents, screens — and time with the people who do it. Worry if: they can recommend a solution from a sales call alone.
- 2. “Which part of this would you not use AI for?”
- Listen for: specific exclusions, such as final figures, advice to clients, or anything a regulator may ask about. Someone who says AI can do all of it has either not looked closely or is not being straight.
- 3. “How will we know if it worked?”
- Listen for: a before measurement taken now, a target, and a date to check it. Worry if: the answer is about adoption, enthusiasm or “transformation” rather than minutes, turnaround or errors.
- 4. “Where does a person check the output?”
- Listen for: a named approval step before anything goes to a client, supplier or system of record, and a log of what was approved.
- 5. “What’s the smallest version of this we could try?”
- Listen for: one workflow, a few weeks, a fixed or capped price. A consultant who can only offer a large programme is telling you something about their business model.
Part two: about your data
- 6. “Where will our information be processed and stored?”
- You want a specific answer naming the services involved and their regions, not “the cloud”. If client or health information is involved, ask how that fits with your obligations under the Privacy Act and the Australian Privacy Principles.
- 7. “Will any of our data be used to train someone else’s model?”
- The answer should be no, with the setting or contract term that makes it so.
- 8. “What access will your tools have, and to what?”
- Access should be scoped to the job: the mailbox or folder the workflow needs, not every mailbox in the business. Ask how access is removed if you end the engagement.
- 9. “Who on your side can see our information?”
- A clear, short answer. If subcontractors are involved, you should know.
Part three: about the relationship
- 10. “Who owns what you build?”
- Prompts, configurations, documentation and any code written for you should be handed over or clearly licensed to you. Find out now, not when you want to leave.
- 11. “What happens if we stop paying you?”
- Does the automation keep running? Can your IT provider maintain it? Is anything locked to an account in the consultant’s name?
- 12. “Do you earn commission on anything you recommend?”
- It isn’t automatically wrong, but you should know. For the record: the team behind this site, SG1 Consulting, also makes its own AI product, The Everything. If it fits your situation you’ll be told it’s ours, and it won’t be the answer to every problem.
- 13. “Who will actually do the work?”
- The person you meet should be able to tell you who builds, who supports and how to reach them.
- 14. “What does support look like after go-live?”
- Models and the services around them change. Ask how changes are monitored, how you report a problem and what it costs.
- 15. “Can we start with an audit only?”
- A good consultant is comfortable selling you advice and letting you decide what to do with it.
Red flags worth walking away from
- Guaranteed savings quoted before anyone has seen your work.
- A single answer to every problem, especially if it’s a product they sell.
- No human approval step on anything that leaves the business.
- Vague data answers, or reluctance to put data handling in writing.
- Lock-in: automations or data held in accounts you don’t control.
- Pressure to decide quickly because of a deadline that isn’t yours.
- “Top consultant” lists published by consultants. Treat any ranking written by someone who appears in it with the scepticism it deserves.
Good signs
- They ask more than they tell in the first conversation.
- They recommend fixing something that isn’t AI when that’s the real problem.
- They talk to your IT provider early.
- They give you a written scope with what’s included and what isn’t.
- They are happy for you to take their audit to someone else to build.
Comparing proposals side by side
When you have two or three proposals, line them up against the same headings: the problem being solved, how success is measured, the approval steps, data handling, what you own at the end, ongoing costs, and how you exit. Proposals that are hard to fit under these headings are usually hard to manage later as well.
It is also worth asking each consultant to describe a project that didn’t work and what they learned. Everyone in this field has had one. An honest answer is a good sign; a claim of a perfect record is not.
Ready to ask us these questions? Book 15 minutes — you can use the whole call for them.