How to buy AI without being sold to

Two hosts go through four claims that come up constantly in AI sales pitches, and for each one, the specific question that exposes it. The episode ends with the listener better armed, not warmer to any particular vendor.

A: We build AI systems for a living, which means the pitches aimed at our own industry land on us too. Four claims come up often enough to be worth naming plainly.

B: And for each one, the useful thing isn't the objection. It's the specific question that exposes it.

A: First — "it learns your business." Retrieval over your own documents is real and useful, an assistant that answers from your policies, your pricing. A model that quietly absorbs how your company works and improves on its own — that's not usually what's actually being sold.

B: Ask instead: what exactly gets better, and how would I know if it didn't? If the answer stays vague after being asked twice, the phrase was marketing, not architecture.

A: Second — fully autonomous, and customer-facing. The technology can genuinely do it. Whether it should turns on one thing: do you want to hear about mistakes from a customer, or from a review step you actually control?

B: For most service businesses, a draft prepared automatically and a person sending it captures nearly all the saving with none of the exposure.

A: Ask instead: what happens the first time it's expensively wrong, before it happens? If the answer only exists in hindsight, the review step was discovered, not designed.

B: Third — anything that can't show its source. If an assistant answers a question about your own policies or prices, you need to see the document and the page. Without that, a wrong answer looks exactly like a right one.

A: Ask instead: show me a wrong answer, and show me how I would have caught it. Not describe it — show it. A vendor who can't produce that hasn't built citation in, whatever the deck says.

B: And fourth — "it replaces a person." In a team of twelve, nobody's doing exactly one job. Removing a task rarely removes a headcount — it usually gives someone their afternoons back.

A: Ask instead: price the task, not the salary. What does automating this specific task actually save, and what does it cost to build and run? If a pitch leans on headcount reduction for a twelve-person business, that's worth questioning directly.

B: None of this means don't buy, worth saying. It means buy the boring, checkable version — the one where you can see the source, see the failure mode, see the arithmetic.

A: And every one of these is a question you should ask when we quote you a project, specifically. If we can't show a source citation, name a failure mode, or price a task instead of a headcount, that's a reason to say no to us.

B: One honest limit — good answers to these questions are necessary, not sufficient. A vendor can pass all four and still build the wrong thing for your actual workflow.

A: One thing to check this week.

B: Next time anyone pitches you AI — including us — ask the exposing question, not the objection. "How would I know if it improved" gets you further than "I don't believe it learns."

A: Written version's at atinnovators.in, under insights.