15 Questions to Ask Before You Buy Any AI Tool

I sell AI tools for a living, so believe me when I say: most AI vendor demos are theater. A cherry-picked input, a confident narrator, a price that turns out to be per-seat-per-model-per-API-call. You don't need to be technical to see through it. You need a list of questions and the nerve to ask them in order. Here's the list.

Data: where does my information go?

1. Do you train your models on my data? The only good answers are "no" or "only if you opt in." "We may use data to improve our services" is a yes wearing a costume.
2. Where is my data stored, and who can access it? Country, provider, and whether subcontractors touch it.
3. How do I delete my data - all of it? Ask them to describe the process. Vagueness here is a dealbreaker.
4. What happens to my data if I cancel? Retention policies outlive subscriptions. Get the number of days.

Reality: what does it actually do?

5. Show me the product failing. Every tool fails somewhere. Vendors who won't show you the failure modes are selling the demo, not the product.
6. Which parts are your own model and which are a wrapper around someone else's? Wrappers aren't bad - unpriced honesty about them is.
7. What does output review look like? If the answer is "trust it," walk. AI output needs a human checkpoint, and serious vendors design for that.
8. Can I talk to a customer my size, in my industry? Reference calls with the wrong-size customer are how you end up as someone's enterprise experiment.

Money: what does it cost at my real usage?

9. What does the bill look like at 3x my expected usage? Usage-based pricing is a trap with good marketing. Model the spike, not the average.
10. What's included in the tier you're quoting me, and what's an add-on? Seats, API calls, support, the features from the demo - which of those were add-ons?
11. What's the price history? A vendor that raised prices 40% last year will do it again after you're integrated.
12. Is there a monthly cap or an overage cliff? Know which one you're signing.

Exit: how do I leave?

13. Can I export everything I put in, in a usable format? Your prompts, your data, your configurations. If leaving means starting over, you're locked in by design.
14. What breaks the day I cancel? Integrations, widgets, workflows - know the blast radius before it exists.
15. Is there a contract term, or is it month-to-month? Annual commitments for unproven tools are how small businesses fund vendor roadmaps.

What are the instant dealbreakers?

Three answers end the evaluation on the spot: they train on your data with no opt-out, they can't tell you where your data lives, or there's no deletion path. Everything else is a negotiation. Those three are a no.

How long should an evaluation take?

For a small business buying a tool under a few hundred a month: one week. Run the 15 questions by email (written answers beat sales-call charm), do one real-task trial with your own data - the boring kind, not the demo kind - and make the call. If a vendor can't survive a one-week evaluation, the tool won't survive a year of use.

FAQ