Explained
AI in business: where it works and where it does not
Some promise everything is about to change. Others say it is a bubble. A narrower question is more useful: what properties does a job need for software to do it reliably.
AI in business gets written about in two tones. One promises that everything is about to change. The other says it is a bubble. Both are useless to someone who has to decide whether it is worth spending a week on.
A narrower question is more useful: what properties does a job need for software to do it reliably.
Three properties that decide everything
Can the result be checked in a few seconds
An answer about available dates can be verified immediately. A forecast of how many clients will come next quarter cannot be verified at all. The first job fits, the second does not.
Are all the necessary facts written down somewhere
If the answer is in the price list, in the terms or in an earlier exchange, it can be found. If the answer exists only in your head, nobody will find it.
Is a mistake cheap
A misfiled email costs a second. A wrongly sent invoice costs a client. Start where a mistake costs little.
Where it works
| Job | Why |
|---|---|
| Sorting and labelling incoming mail | Verified instantly, a mistake is cheap |
| Pulling data out of free text | The facts are in the text itself |
| Drafting a reply from the price list | All facts are written down, a person reviews before sending |
| Finding what has stalled | The rule is clear and the result is checkable |
| Rewriting the same text in a different tone | There is no correct answer to get wrong |
Where it does not work
| Job | Why |
|---|---|
| Pricing a one-off project | There are no facts, and a mistake is expensive |
| Settling a dispute with a client | It needs responsibility, not text |
| Advising on legal or medical questions | A mistake costs too much and cannot be checked quickly |
| Forecasting demand from a small amount of data | The result cannot be verified until time passes |
The most common question: will it make things up
It will, if you let it. A language model is built to always produce a plausible answer, not to admit it does not know. Asked about a price it has never seen, it will offer a number that looks right.
That is not solved by a better model but by two rules. First: facts come only from the base you wrote down, not from the model's memory. Second: if the answer is not in the base, the answer is "I will check and come back", not a guess.
How to test this before trusting it. Ask the software ten questions whose answers are deliberately not in your price list. If it invents something even once, it is not ready to go anywhere near clients.
What the law says
Since August 2026 the EU AI Act requires telling a person clearly that they are dealing with artificial intelligence. In practice that is one line at the end of the reply.
The other side is data. If client emails travel to a model, that is processing of personal data, which has to be described in the privacy policy and needs a lawful basis.
An honest conclusion
AI in business is neither a revolution nor a bubble. It is a tool that does a narrow range of jobs very well: the ones where the facts are written down, the result is quick to check, and a mistake is cheap. There are more such jobs in a small business than it seems, and considerably fewer than the advertising promises.
If you want to see this with your own numbers
Give it thirty minutes and we will walk through one real enquiry of yours from start to finish. After the call you get a map of where data is retyped by hand at your company. If there is nothing worth automating, I will say so.
Let us look at your workflow