There is a sticker being applied to almost every piece of business software right now, and the sticker reads "AI-powered." Sometimes it is true and meaningful. Often it is the same software that existed last year, repriced upward, with the sticker as the only new feature. As a buyer earning in shillings, you cannot afford to pay an AI premium for software that is not actually doing anything new.
So this essay is a filter. Five questions you can ask any vendor — none of them technical — that separate a real AI capability from a marketing label. You do not need to understand how the technology works. You need to be able to tell whether it is there at all.
Why the label is so often empty
Understand the incentive first. "AI" is the word that loosens budgets in 2026. A vendor who adds it to a product can often charge more and close faster, whether or not anything underneath changed. The pressure to apply the sticker is enormous, and the cost of applying it falsely is, for now, almost nothing. So the market is flooded with it.
This does not mean AI is fake. It means the label is unreliable, and you have to look past it to the capability. A rule of thumb: the more prominent the word "AI" is in the marketing, relative to any description of what it actually does for you, the more suspicious you should be. Real capability tends to describe itself in terms of outcomes. Empty labels lean on the buzzword because there is nothing else to lean on.
The more loudly a product shouts "AI," relative to describing what it actually does for you, the more likely the word is doing the work the product cannot.
The five questions
1. "What can it do this year that last year's version couldn't?"
If the answer is vague — "it's smarter," "it's AI-enhanced" — be wary. A genuine new capability is specific and nameable: "it now drafts the reply for you," "it now reads the document and pulls out the figures," "it now answers customers in Kiswahili on its own." If the vendor cannot point to a concrete new thing it does, the AI is probably on the sticker, not in the product.
2. "Does it generate or understand, or does it just follow rules I set?"
This is the cleanest technical line, asked in plain language. Old software follows rules you configure: if this, then that. Modern AI generates or interprets — it writes new text, understands a messy request, reads an unstructured document. Ask the vendor to show you it doing something that no fixed set of rules could have anticipated. If everything it does could be a rule someone wrote in advance, you are looking at ordinary automation with a new name. That may still be useful — but you should not pay an AI premium for it.
3. "Can I see it work on my own example, right now?"
The fastest test of all. Hand the vendor a real input from your business — a real customer message, a real document — and ask to watch it respond, live, unrehearsed. Real capability survives your example. A marketing label tends to come with reasons why a live demo on your data is "not possible today." The polished demo on their carefully chosen example proves nothing; the messy test on yours proves everything.
4. "What happens when it gets something wrong?"
Real AI is probabilistic — it is sometimes wrong, and an honest vendor will tell you so and explain how their product handles it (a human review step, a confidence flag, an easy correction). A vendor who claims their AI is never wrong is either misunderstanding their own product or lying to you. Paradoxically, the willingness to discuss failure is a sign the thing is real. Software that "never makes mistakes" is usually software that is not doing anything hard enough to.
5. "What exactly am I paying extra for?"
Finally, follow the money. If the AI version costs more than the non-AI version, ask precisely what the extra buys, in terms of work done for you. A clear answer — "it saves your team the hour a day they spend on X" — is a real value you can weigh. A circular answer — "you're paying for the AI" — means you are paying for the sticker.
When the label is honest
To be fair to the real products: there are many, and they earn the word. You will know them because they answer the five questions easily and concretely. They tell you exactly what is new, they let you test on your own messy example, they are honest about when they are wrong, and they can name the work they save you in hours or shillings. The honest vendors are not hurt by these questions. They are the ones who built something worth asking about.
The questions are not a way to dismiss AI. They are a way to find the real ones in a market full of stickers — so that your money goes to capability you can point to, not to the most fashionable two letters in business software.
The takeaway
Do not buy "AI." Buy a capability you can name, test on your own data, and tie to hours or shillings saved. The label is free to apply and therefore worthless as a signal. The five questions cost a vendor nothing to answer if the product is real, and everything if it is not — which is exactly why they work. Ask them before you pay the premium, every time.