A sales rep just sent you a demo. The pitch is slick. The dashboard glows. There's a customer logo you recognize. You feel a little behind — like everyone else is already doing this and you're not.
That feeling is the most expensive thing in small business AI right now. It is sold deliberately. And it has very little to do with whether the tool will actually help you.
Before you buy anything — a chatbot, an AI assistant, a "co-pilot," an automation platform, a generative whatever — ask these five questions. Out loud, in writing, in the second sales call. If the rep can't answer them clearly, that is the answer.
1. What specific work, by whom, does this replace?
Not "saves time." Not "increases productivity." Not "drives efficiency." Those are vendor words. They are designed to be unfalsifiable.
You need a sentence that sounds like this: Maria spends about four hours a week drafting follow-up emails. This tool would write the first draft of each one, and she'd review and send.
That sentence has names. It has hours. It has a specific task that already exists in your company. If you cannot write that sentence after the demo, you do not understand what the tool does — and neither does the vendor.
Vendors hate this question because most AI tools are sold as capability, not replacement. They want you to imagine the possibilities. You want to know exactly which 20 minutes of your Tuesday this is going to save.
2. What does it cost when it gets something wrong?
Every AI tool is wrong sometimes. That is not a flaw. That is how the technology works. The question is what happens when it's wrong, not if.
If your AI tool drafts a follow-up email and the tone is off, the cost is small — your employee notices and rewrites it. If your AI tool quotes a price to a customer and that price is wrong, the cost is a refund, an apology, and maybe a lost account. If your AI tool tells a patient something incorrect about their medication, the cost is a lawsuit.
Map the failure mode before you sign. Three questions inside the question:
- What's the worst thing this tool could output, and who sees it before a human?
- How often will it be wrong? (If they say "almost never," ask for the number.)
- What's our recovery cost when it is?
The right answer isn't zero. The right answer is small and recoverable. If the failure mode is "the AI sounds slightly weird in an internal Slack message," buy with confidence. If the failure mode is "the AI bills a customer the wrong amount," you need a human in the loop, no matter what the demo showed.
3. Who on our team actually has to use this — and have they tried it?
This is the question that quietly kills more deals than any other. Buying happens in the executive office. Using happens at the front desk, in the kitchen, on the floor, in the truck.
If the people who will actually use the tool every day haven't sat down and tried it for thirty minutes, you are buying a guess. Their guess. Their boss's guess. The vendor's guess.
We have watched companies spend twenty thousand dollars on AI tools their staff quietly refused to use after week two. The tool wasn't bad. It was wrong for the people, the workflow, or the moment.
Before signing:
- Pick the two or three employees who will use this most.
- Have them — not you — drive a real task in the tool for thirty minutes.
- Listen to their language afterward. "I could see using this for…" is a good sign. "It's interesting…" is not.
If your team is excited, you've found a tool. If your team is polite, you've found a future shelfware purchase.
4. What happens if you are not here in 18 months?
This is the question vendors find most uncomfortable, which is exactly why you should ask it.
The AI software market in 2026 is the dot-com era again, with similar economics. A lot of companies pitching you have raised venture money, have no profit, and are betting on growth that may not arrive. Some of them will not exist in 2027.
That is not necessarily a reason to not buy. It is a reason to know your exit before your entrance.
Three sub-questions:
- Can we export our data? In what format, how often, and at what cost? Are there limits on what we can take with us? "We're working on it" means no.
- Are we locked into a long contract? Auto-renewing annual contracts with 60-day cancellation windows are the small business AI industry's favorite hostage tactic.
- What's the migration path to a competitor? If the answer is "there isn't one because we're unique," translate that to: you can never leave us.
A good vendor answers these calmly because they aren't worried about you leaving. A nervous vendor changes the subject.
5. Show me one customer my size, in my industry, doing exactly this.
Not a Fortune 500 case study. Not "a leading enterprise client." Not a logo wall.
You want to talk to a real, small business owner — five to fifty employees, ideally in your industry, ideally within driving distance — who uses this tool the way you would use this tool. On the phone. For ten minutes. With permission to ask anything.
A vendor that can produce this reference quickly is selling a product. A vendor that cannot is selling a category.
What to ask the reference customer:
- How long did it actually take to roll out? (Not what the vendor promised. What happened.)
- What did your team complain about in month one?
- If you were buying this again tomorrow, what would you ask the vendor that you didn't ask the first time?
- What's it really costing you? (Including the time of the person who manages it.)
Ten minutes on the phone with the right reference will tell you more than ten hours of demos. If the vendor cannot find that reference for you in a week, the customer base is either too small, too unhappy, or too unlike you to matter.
The honest truth about this checklist
If you ask all five questions honestly, you will not buy most of the AI tools you are pitched in 2026. That is not a bug.
The current market is full of products that are good demos and bad purchases. The checklist isn't here to make you a harder sale. It's here to make you a smarter buyer — so when you do say yes, you say yes to something that will still be paying off in a year.
Print this. Tape it to your monitor. The next time someone forwards you a glossy AI pitch, run it through the five questions before you book the call.
If it can't survive these questions, it doesn't deserve thirty minutes of your time. If it can, you've already done more diligence than 90% of the small businesses that bought before you.