Hiring an AI consultant starts with deciding what help you need and asking for evidence that matches it. A good product demonstration can be useful, but it does not tell you who will prepare your sources, teach your staff, or support the workflow when something changes.
For a Minnesota owner comparing providers, I would use the same questions whether you are considering me or someone else. The aim is a clear agreement and a team that can understand the work it receives.
Describe the next decision before choosing a provider
Say where you are today. You might need to identify a first task, establish shared team practices, test a workflow, or improve something people already use. These are different engagements.
Bring a general description of the task, who does it, and what makes the result acceptable. You should be able to discuss fit without sharing confidential documents or handing over system access in an introductory call.
Ask the provider to explain what they would learn first and what you would receive. If you need a roadmap, a demonstration of a custom assistant does not answer whether the proposed roadmap will help you choose.
Inspect evidence and ask what it establishes
Ask to see relevant work and understand the provider's role. Did they build it, co-build it, design it, or contribute one part? Is the example a live product, a prototype, a prepared walkthrough, or a proposal?
Ask what was tested, with which kinds of inputs, and who reviewed the results. A prototype can demonstrate useful technical judgment. It should be described as a prototype, with its remaining integration and adoption work visible.
On this site, the agency operations assistant is designed and prototyped; its prepared walkthrough and mock-meeting validation do not establish production adoption or business impact. That distinction is a question worth asking about any example you inspect.
A provider may be unable to show confidential client work. Ask whether a permitted demonstration, public artifact, or clearly labeled fictional example can explain their approach without exposing another business's information.
Ask them to explain the whole task
Have the provider describe the inputs, output, review step, and next action in plain language. Ask where the proposal depends on access, an integration, source cleanup, or a decision from your team.
Find out what happens when information is missing, contradictory, or out of date. Ask who can stop the workflow and who resolves an exception. Preparing a client reply and sending it should have separately understood responsibilities.
NIST's voluntary AI Risk Management Framework and companion Playbook offer further reading on governing, mapping, measuring, and managing AI risk. They can inform your questions; naming a framework is not evidence that a particular project has been evaluated well.
Request a test you can review
Agree on representative examples and acceptance conditions. Include a difficult input as well as an ordinary one. Ask to see the source material and understand why the reviewer accepts each important statement.
Record preparation, checking, corrections, and whether the result supports the next step. Ask how the provider will address failures and which limits remain. The evidence should relate to the task you intend to buy.
For training, ask what participants will practice and take away. A worked example, reusable instructions, and a review guide give you something concrete to assess. Discuss how staff will get help after the session.
Clarify access, accounts, and ownership
Identify the accounts and information the work requires. Confirm who authorizes access, who administers accounts, and how temporary access is removed. Ask the provider to verify the chosen service's relevant settings and terms rather than promising blanket protection.
Discuss which materials, code, configuration, and instructions the team receives, and under what ownership or license terms. Ask about dependencies and continuing subscriptions. Confirm these details in the agreement; a verbal promise does not describe the handoff sufficiently.
Find out who actually does the work and which parts involve other people or suppliers. Your team needs a clear contact for decisions and access requests.
Read the proposal through the handoff
Check deliverables, exclusions, dependencies, fees, payment arrangements, and how changes are approved. Ask what happens if a source or integration is unavailable. Make your own responsibilities visible too.
Discuss maintenance, support boundaries, and the process for reporting a problem. Ask how you could continue with an internal maintainer or another provider, and what information they would need. Keep a workable fallback for the underlying business task.
Use the hiring checklist below to record evidence and unanswered questions instead of awarding a generic score. Choose based on the fit you can explain and the agreement you can review. If you want to discuss your task with me, the free 30-minute AI consultation is a place to start; booking it does not commit you to paid work.