Short answer
You probably do not need AI if the task has a fixed set of correct answers, if the real problem is that information is missing rather than hard to find, or if the volume is low enough that a person handles it in minutes a day.
1. The answers are fixed and countable
If a customer question has one of eight possible answers, such as opening hours, delivery zones, return policy or price list, that is not a language problem. That is a page on your website and a well-organised FAQ. A language model can answer those questions, but it introduces a failure mode you did not have before: occasionally it will answer confidently and wrongly.
The test: write down every question customers actually ask, in order of frequency. If the top twenty cover most of the volume and each has one correct answer, build the page, not the model.
2. The problem is missing information, not slow retrieval
A common request is a system that answers questions about stock, orders, or availability. When we dig in, the underlying issue is often that the information does not exist in any system. It lives in someone's head, or in a notebook, or it is written down inconsistently enough that no tool could read it.
AI cannot retrieve information that was never recorded. If your inventory is not tracked, the first project is tracking inventory. That is unglamorous and it is the thing that will actually change your week.
3. The volume does not justify the build
Automation earns its cost when the task is frequent. If a task happens four times a day and takes two minutes, that is eight minutes. A build that costs weeks of engineering to save eight minutes a day is a bad trade unless the task is also error-prone or blocks something else.
The honest calculation is not "could AI do this?", because it almost always could. It is "what does this cost me today, and what will the tool cost to build and keep running?"
Where AI genuinely does earn its place
The cases that consistently justify a build share one of these traits:
- The input is unstructured and varied. Free-text messages, documents in different formats, handwriting. No rule set covers it.
- The volume is high and repetitive. Hundreds of similar documents or messages, where minutes each add up to days.
- Expertise is the bottleneck. The work requires someone knowledgeable, and there are not enough of them. This is why Kateb exists: drafting is slow because it requires a notaire's judgment on every act.
Why we say this out loud
A build that does not fit the workflow gets abandoned within a month, and the client remembers who sold it. Turning down a project we cannot make useful costs us one invoice. Selling one that fails costs the relationship and the referral. That trade is not close.
Frequently asked questions
How do I know if my business needs AI?
Check three things: whether the inputs are unstructured and varied, whether the volume is high enough that minutes add up, and whether the bottleneck is scarce expertise. If none apply, a simpler tool will likely serve you better.
Is a chatbot worth it for a small shop?
Often not. If customer questions have a small set of fixed answers, a clear webpage or FAQ handles them without the risk of a confidently wrong reply. A chatbot earns its place when questions are varied and volume is high.
Will you tell us if AI is the wrong tool?
Yes. We would rather lose one project than deliver something that gets abandoned. If a form, spreadsheet or phone number solves it better, we say so during scoping.