Where AI belongs, and where it doesn't
Putting AI everywhere is easy. Deciding where it actually earns its place is the hard part. Here's the simple test we use.
For the last two years, almost every project brings the same question: "Could we put AI here?"
Usually we could. The real question is whether we should.
## A simple test
The test we use: **if the task is repetitive and time-consuming, and the result can be verified**, AI is a good candidate.
Both conditions matter. For repetitive tasks whose results can't be verified, the model will produce fluent output that nobody can check. For verifiable tasks that already take two seconds, it saves nobody any time.
## Where it works well
**Understanding an incoming message.** A customer writes "could we look at the day after tomorrow, afternoon?". Turning that into a date is what models are good at — and the result is verifiable: the calendar is either free or it isn't.
**Structuring messy text.** Uploading a photo of a printed menu and splitting it into categories and items. The venue sees the result and can correct it, so verification stays with a human.
**Answering from documents.** Producing answers from the business's own price list, policies and FAQs. The model speaks from your documents, not from general knowledge.
## Where it doesn't
**Making the final call.** The system should open the appointment, not the model. The model says "this slot looks free", the system checks the calendar and decides.
**Producing exact numbers.** Price calculations, discounts, stock counts. That's arithmetic; handing it to a model gains nothing and risks errors.
**Answers with legal or medical consequences.** Here, who is accountable matters more than what the model said.
## And there's a hidden cost
Adding AI isn't a one-off job. Models change, pricing changes, and one day the output degrades in a way nobody expected. You need to monitor it, measure it, and be able to switch it off.
So in the systems we build, the AI side is always a layer that can be turned off. Switch it off and the product keeps working — it just becomes more manual.
## In short
AI is a tool, not a feature. Put it in the right place and it genuinely saves time. Put it in the wrong place and it only looks impressive in a demo.
If you have a use in mind, write to us — we'll tell you honestly whether it's worth it.