Sooner or later somebody is going to pitch you AI. Some of those pitches are good. The trick is telling which, and you do not need to understand the technology to do it.
Here is the pattern in most bad AI deals: it is almost never that the technology did not work. It is that the incentives were pointed the wrong way and the ownership was vague, and nobody noticed until the invoice. Both of those you can check up front with five questions.
1. What happens when the AI company changes the rules?
Every AI tool is built on top of a handful of big providers, and those providers change prices, retire old models, and have outages. This is normal, it is nobody's fault, and it will happen during the life of anything you buy.
So ask: when that happens, who handles it, how fast, and what does it cost me? A good vendor has a plain answer because they have been through it. A vendor who says "that will not happen" has not been doing this long enough to be selling to you.
2. Who owns what I am paying for?
The expensive surprise in a lot of AI deals is finding out you rented. The tool runs in the vendor's accounts, on the vendor's keys, and the day you stop paying, it all goes with them.
Ask it as one clean question: if we part ways tomorrow, what do I keep? The answer should be in writing, and it should include the accounts, the data, and anything built specifically for you. While you are at it, ask whether your business data gets used to train anything. The right answer is no, also in writing.
3. What does it cost to keep running?
The build price is the number on the proposal. The real number includes what it costs every month after: subscriptions, AI usage, and upkeep.
Two things to know. First, for the text-heavy work most shops automate, the AI usage itself is cheap, pennies per task, so a fat monthly "AI costs" line is usually markup wearing a costume. Second, the honest setup is running costs in your accounts, on your card, at provider prices. If a vendor insists everything bills through them, ask why.
4. Who gets paid if the answer is no?
This one decides how much weight to give every answer you have collected so far.
If the person across the table only makes money when you buy, then "you do not need this" is not on their menu, no matter how nice they are. That does not make them crooked. It makes them a salesman, and you should weigh their advice the way you weigh any salesman's.
Ask them straight: what would you tell me if AI was not worth it for my shop? Watch whether the answer names a real alternative or circles back to the pitch.
5. What is the non-AI alternative?
This is the question that sorts advisors from sellers fastest. Every real problem has a ladder of fixes, and AI is the last rung, not the first: fix the process, use a spreadsheet, buy something boring off the shelf, then build.
So ask anyone pitching you: what would solve this without AI, and why is your thing better than that? If they cannot name the boring alternative, they did not look for it. And if they will not compare against it, that tells you how the comparison comes out.
What to do with the answers
You do not need all five to come back perfect. You need them answered plainly, in writing where it matters, from somebody who did not flinch. One dodge is a yellow flag. Two is a pattern. A pitch that survives all five questions is worth taking seriously, and they do exist.
If you want a second set of eyes on a pitch you have already received, bring it to the free call. Fifteen minutes, and I will tell you what I see, including when the deal in front of you is a good one.