What to ask an AI consultant before you hire one

An owner watches a demo or takes a call, and comes away thinking two things at once. First, that AI could probably take some real work off the team's plate. Second, that they cannot tell whether the person pitching it knows the business or just knows the script. That moment is where most bad decisions happen, because the demo looks impressive and the questions never get asked. A short checklist, used before any money moves, does most of the sorting for you.

The checklist covers four areas. Who does the actual work after the pitch. What literally happens in a working session. How you will both know the engagement worked. And what the shape and terms of a fair engagement look like. Ask the same questions of any option you are considering, including doing it yourself.

Question one: who does the work

The most common trap in this market is the person who sells the work and then hands you to someone you never met. Ask directly: after I sign, who is on my calls, and can I talk to that person before I commit? A good answer names the person and the arrangement plainly. A bad answer is a description of a team, a pod, or a bench, with no name attached. Also ask whether the helper runs their own operations on the tools they are pitching, because a person who uses AI in their own daily work answers practical questions faster than a person who only installs it for others.

Question two: what happens in a session

Ask for a walkthrough of one working session, minute by minute. A good answer describes real tasks: drafting a reply to a quote request in your voice, building a follow-up routine, cleaning up a process that eats hours each week. A bad answer describes concepts, trends and slides. You are buying time and hands-on help, so the session should involve your work product, in your accounts, with you watching or doing. If the session happens in the vendor's own tool that disappears when you leave, ask what you keep at the end.

Question three: how success is measured

Be suspicious of anyone who promises a specific dollar figure or a growth rate before they have seen your tasks. A good answer starts with picking one or two repetitive tasks, doing the work, and then measuring what changed: hours saved on a given chore, response time on quotes, whether the routine stuck after a month. Ask how they report that back to you, and what happens when something does not work. An honest helper will tell you plainly when a task is not worth automating.

Question four: what fair looks like

Ask what the engagement costs, in writing, before anything starts, and what is included versus quoted separately. Bigger builds, like API connections, CRM integrations, hosted agents or dashboards, are legitimately separate projects, but a fair helper scopes and quotes them before work begins rather than sliding them in later. Ask who owns the accounts and the work product. The right answer is that you do, in your own accounts, whether you stay or go. Ask what the exit looks like. A fair engagement does not hold your work hostage.

Your four real options

• Do it yourself. Costs the least in fees and the most in hours. You will need a subscription to a general AI assistant such as ChatGPT or Claude, and the discipline to work through free tutorials and articles until something sticks. The trade-off: nobody checks your work, and abandoned experiments are common.
• A peer or operator you know. Another Arizona owner who is further along with AI will give you straight answers for the price of coffee. The trade-off: their advice is general, they have no time commitment to you, and their situation may not match yours.
• A consultant or concierge. You pay for a person who learns your business and works on real tasks with you on a regular cadence. The trade-off: it costs money, and the quality gap between helpers is wide, which is the reason for the checklist above.
• Team training. Hands-on sessions built on the tasks your people actually do, with take-home routines and ground rules. The trade-off: it builds a shared skill rather than one person to lean on, and it needs the owner's backing to stick.

What good answers sound like, one example

One Arizona provider, AZ AI Lab, runs a model you can use as a measuring stick. Its entry point is a free 25-minute fit call on Zoom that looks at two tasks that repeat in the owner's week and picks a first result worth having, or says honestly if the fit is not there. From there, a named person becomes the owner's AI Concierge: two 1-hour working calls a month, voice messages and email between calls, a personal AI context file kept in the owner's own account, and a clear line where bigger builds get scoped and quoted separately. Everything built lives in the client's own accounts and stays there. That structure answers all four checklist questions before you have to ask.

The checklist, in short