Every software vendor's pitch deck has an AI slide now. Chatbots, "smart" recommendations, predictive this, generative that — it's become a default feature label rather than a meaningful description of what a product does. That's not necessarily a problem; some of these tools genuinely use AI well and deliver real value. But "AI-powered" on its own tells you almost nothing about whether a tool fits your business, what it will actually cost to run, or what happens when it gets something wrong.
If you're evaluating a vendor that leans heavily on the AI label, a short list of pointed questions will tell you more than any demo will.
What model is actually doing the work?
You don't need proprietary architecture details, but you should get a straight answer: is this built on a general-purpose model from a major provider, a fine-tuned version of one, or something built in-house? Each answer has different implications. A thin wrapper around a general model might mean fast updates but less control and dependency on someone else's pricing and uptime. A custom or fine-tuned model might mean better performance on your specific use case, but slower iteration. Neither answer is automatically wrong — but a vendor who dodges the question, or answers with marketing language instead of specifics, is worth a second look.
Where does our data go, and who can see it?
This is the question with the most legal and reputational weight, and it's the one vendors most often answer vaguely. Ask directly: is our data used to train or improve the underlying model, either for us or for other customers? Is it retained after we stop using the product, and for how long? Is it processed by a third-party model provider, and if so, under what terms? You want this in writing, not just verbally confirmed on a sales call. If the vendor can't produce a clear data handling policy, that's itself useful information.
What happens when it's wrong?
AI tools make mistakes — confidently, and sometimes in ways that are hard to spot. A tool that summarizes documents can misrepresent a key detail. A tool that drafts customer replies can say something inaccurate in a tone that sounds perfectly authoritative. Ask the vendor how errors are typically caught, whether there's a human-in-the-loop option, and what recourse exists if the tool's output causes a real problem — a bad customer interaction, an incorrect number in a report, a compliance issue. A vendor with a mature product will have thought about this already and can describe it clearly. One that hasn't will talk around the question.
How is pricing actually structured?
AI features often introduce usage-based costs that traditional software pricing didn't have — per-query fees, token consumption, rate limits that either throttle you or bill you more. Ask for real numbers: what does a typical month look like at your expected volume, and what happens if usage grows faster than planned? Vendors are generally happy to quote a low starting price; fewer are eager to walk through what happens at scale. Get that answer before you're locked in, not after the first surprising invoice.
Can we turn it off, or dial it back?
Some AI features are additive and easy to disable if they don't work out. Others are woven into the core workflow in a way that makes the tool hard to use without them. That distinction matters, especially if your team is cautious about AI-generated content reaching customers directly. Ask what the fallback looks like — a non-AI mode, a review step you can require, a way to scope the feature to internal use only while you build confidence in it.
What does "AI-powered" actually mean here?
Finally, it's fair to just ask this plainly. Have the vendor walk you through, in specific terms, what the AI component does, what it doesn't do, and what a normal user actually experiences day to day. A good vendor will be glad to get concrete — they know their product and can explain it without hiding behind the buzzword. If the explanation stays vague no matter how directly you ask, treat that as a signal, not an accident.
None of these questions require a technical background to ask, and a legitimate vendor should be able to answer all of them without hesitation. The goal isn't to be adversarial — it's to buy based on what a tool actually does for your business, rather than a label on a feature list. That distinction is worth more than most of the discount codes vendors offer to close a deal faster.
If you're in the middle of evaluating AI tools for your business and want a second opinion on whether a vendor's claims hold up, we're happy to talk it through — no pressure, just a practical sanity check.