Key takeaways
- The best single question: what was live and being used at the end of your last three engagements?
- Insist on fixed deliverables and an end date for any work whose scope is knowable.
- A consultant who does not raise spend caps and data handling unprompted is missing the parts that cause incidents.
- Everything built should belong to you, with admin access in your hands.
AI consulting has a low barrier to entry. Plenty of people now sell it, and from the outside the good and the weak look similar: confident proposals, familiar logos, the right vocabulary. These questions are designed to get past that. They work on anyone, including us.
Scope and outcome
1. What will be live and in use when you finish?
You want a specific answer: "your team on Claude Team with SSO, spend caps at the organisation and user level, connectors limited to Drive and Slack, a usage policy and a trained staff." Worry if the answer is a strategy, a roadmap or a maturity assessment and nothing working.
2. What did your last three clients have running at the end?
This tells you more than any case study page. Ask for the size of the business, what was implemented and how long it took.
3. How will we know it worked?
Good answers mention weekly active users at 30 and 90 days, time saved on named tasks, or a reduction in personal account use. Weak answers mention satisfaction surveys.
Price and risk
4. What is the total fee, and what makes it change?
For a standard rollout, the fee should be fixed. If it is hourly, ask for a cap. See fixed-price or hourly AI consulting.
5. What is the end date, and what happens if you miss it?
Whoever carries the schedule risk should be whoever controls the work.
6. What will we pay providers, separate from your fee?
Licence and usage costs should be estimated up front and paid directly to the provider, not marked up silently.
Neutrality
7. Do you receive commission or partner incentives from any AI provider?
It is not disqualifying. It is something you should know before you take their recommendation.
8. When would you recommend a different provider from the one you usually use?
If they cannot describe a case, they probably only use one.
9. Would you try an off-the-shelf tool before building something custom?
For most small and mid-sized businesses, business-tier assistants with good configuration cover most of the need. A consultant who jumps to a custom build is either solving a genuinely unusual problem or selling the bigger project. See build or buy AI.
Governance
10. How will you cap our spend?
The right answer is a hard limit at the provider, set at organisation level and, where available, per user. Alerts alone are not a cap. See hard AI spending limits.
11. What will our staff be told about client data?
There should be a written usage policy, in plain language, that covers what can and cannot go into the tool. See handling client data in AI tools.
12. Which connectors and integrations will be switched on, and who decides?
As assistants increasingly act on your systems, this is a security question. You want a deliberate list, not every connector on by default. See AI agent permissions and connectors.
Ownership and afterwards
13. Who will have admin access when you leave?
You. The consultant's access should be removed or reduced at handover.
14. What do we keep?
Configurations, policies, guides and training recordings should all be yours to keep and change.
15. What should we do in the first 90 days after you leave?
A good consultant will hand you a short list: review seats at 30 days, check usage and caps monthly, refresh training at 90 days. Our version is in the first 90 days after AI go-live.
Walk away if: the price is only available after three calls, the deliverable is a document, nobody mentions spend caps or data handling, or the consultant keeps the admin keys.
Frequently asked questions
How do I know if an AI consultant is any good?
Ask what was live and in use at the end of their last three engagements, and how they measured it. Good consultants talk about working systems and adoption numbers. Weak ones talk about strategies, frameworks and roadmaps.
What are red flags when hiring an AI consultant?
No published or fixed price for standard work, vague deliverables, recommending a custom build before trying off-the-shelf tools, no mention of spend caps or data handling, and an engagement that ends with a document rather than a working setup.
Should an AI consultant be vendor neutral?
For choosing a provider, yes. A consultant tied to one vendor can still do excellent work inside that platform, but get the provider decision made by someone without a stake in the answer.
Who should own the AI setup after the consultant leaves?
You should. Admin access, configurations, policies and training materials should all be yours, with a named internal owner. Be wary of any arrangement where only the consultant can make changes.
Get a straight answer for your business
The Clarity Package gives you a written business case and a detailed implementation plan for 500 USD. The Implementation Package does the whole build, live in 30 days, for a fixed 5,000 USD.