Most AI products sell access to a capability. You still do the operating, the checking and the fixing. Some deliver a finished result and answer for it. Both can be worth paying for. Four questions tell you which one you’re buying.
Do this
- Name the exact problem it solves. One recurring problem you already have, in a sentence: “Inquiries from the website go unanswered for two days.” If the vendor answers with a feature list, ask again.
- Ask what finished output it delivers. A draft you still have to edit, check and publish is a different product from a page that’s live and correct. Know which one the price buys.
- Find where a person checks the work. Who reviews it before a customer sees it? If the answer is you, count that time as part of the price.
- Agree how you’ll know it worked. Pick a number you already track: calls, booked jobs, hours spent, replies sent. Write down where it is today and check it again after 30 and 90 days.
- Set the budget from the problem, not the price tag. Spend less than the conservative value of what solving that one problem would recover. If the tool’s price doesn’t clear that bar, it isn’t worth buying yet.
Red flags
- Guaranteed rankings or mentions. Nobody can guarantee a spot in Google results or in ChatGPT’s answers. The “pay to get listed in ChatGPT” offers are the clearest example: there is no listing to buy.
- No one answerable. A custom agent that’s set up, left running and has no one to call when it breaks something. Ask who you contact when it gets something wrong, and how fast they respond.
- A tool for every job. Each one solves a slice, and you end up as the person moving work between them. How to use AI for small business walks through that trap.
- Vague data terms. Ask what happens to the customer information you put in, whether it’s used to train the vendor’s models, and how you get it back out if you leave. Get the answers in writing.
- Numbers without a source. “Businesses using our AI see 3x growth” means nothing without who, measured how, against what.
Questions to ask before you sign
- What does a finished job look like, and who signs off on it?
- What happens when it’s wrong? Who fixes it, and does that cost extra?
- Can I cancel monthly, and what do I keep if I do?
- Which of my accounts or data does it need access to, and why?
Where small businesses start
Among U.S. firms that use AI, most keep it to a few areas: 57% use it in three or fewer parts of the business, most often sales and marketing, then strategy and business development, then IT, according to a Census Bureau working paper. Starting narrow is normal. One problem solved well beats five tools half set up.
If you want the long version, NIST’s AI Risk Management Framework covers how larger organizations think about AI risk. It’s voluntary, and you don’t need it to buy one tool well.
What you get back
A short list of AI tools you can justify, each tied to one problem, one finished output, one person who checks it and one number that tells you whether it worked. Anything that can’t fill in those four blanks waits.
Common questions
Should I buy one AI tool or several?
As few as solve the problem. Every extra tool is another login, another bill and another handoff you have to manage. If you find yourself copying output from one tool into another, you've become the integration layer.
Are free AI tools good enough for a small business?
For drafting, summarizing and first translations, often yes. Pay when you need something finished and checked, not just generated, or when a free plan's limits start costing you time.
How long should I try an AI tool before deciding?
Long enough to see the number you picked move, or not. For most tools that's 30 to 90 days. Decide the number before you start, not after.
Who is responsible if an AI tool gets something wrong?
In practice, whoever published it, which is usually you. That's why the third question matters. Find out who checks the work before a customer sees it.