Short answer: anyone can say they will "set up AI" for you. Ask them four things. What data will it touch? Who approves what it does? What happens when it is wrong? How do you leave? If the answers are vague, the setup will be too.
If you are still deciding whether to start at all, read where to start with AI first. If a tool you already pay for just added AI, see what to do when your software vendor adds AI. This post is for the moment a consultant, a nephew, or a vendor offers to do it for you.
1. What data will it touch?
Ask them to name the systems. Not "your data." The actual names: your practice management system, your accounting file, your email, your shared drive.
Then ask:
- Does it read only, or can it change records?
- Where does the data go while the AI works on it? Whose servers?
- Is any of it kept after the job is done? For how long?
- Is it used to train anyone's model?
A good partner answers these in writing. If client files, patient records, or tax returns are involved, ask whether the work can run on hardware you control. On-premises is not right for everyone, but you should know whether it is an option.
2. Who approves what it does?
This is the most important question. Ask for the exact point where a person in your business says yes.
- Does anything go to a customer, client, or patient without a person reading it first?
- Can it book, charge, refund, or promise anything on its own?
- Who in your office is the named approver for each kind of output?
Our view is simple: the AI prepares lists and drafts. A named person approves and sends. Decisions about price, eligibility, legal advice, or clinical care stay with your people. If the partner's plan has no approval step, ask why.
3. What happens when it is wrong?
It will be wrong sometimes. Every system is. Ask how you will find out.
- Does each draft show where its facts came from, so a reviewer can check?
- What does the tool do when the answer is not in your records? (You want "I could not find it," not a guess.)
- Who do you call when something goes wrong, and how fast do they answer?
- Is there a log of what it did, and can you read it?
A partner who says it will not make mistakes has not used it enough.
4. How do you leave?
Ask this before you sign, not after.
- If you stop paying, what happens to your data and to anything the setup built?
- Can you get your prompts, templates, and settings back in a form you can use?
- Is anything locked to their account that should be in yours?
- What does the contract say about deletion when you leave?
The accounts, the logins, and the data should be in your business's name. If a key person on their side leaves, your setup should still work.
Smaller questions worth asking
- Who on their side can see your data? By name or role.
- What will it cost after the first year? Usage-based pricing can grow.
- What does your staff need to learn? If the answer is "nothing," be careful. Someone has to review.
- Have they done this in a business like yours? Ask what went wrong last time. Their answer tells you more than a case study.
A realistic first step
Write the four questions on one page. Give the same page to every person who offers to help. Compare the answers side by side. The partner who answers in plain words, and who tells you where AI should not be used in your business, is usually the one to trust.
If you want to ask us the same four questions — talk with us. If your answer is that you do not need a partner yet, we will say so.
Related: where to start with AI · ChatGPT is not AI for your business · on-premises AI.