What a small business can actually automate with AI (and what it can’t)
Over the past year, we’ve received dozens of inquiries that start the same way: “we want to bring AI into the company.” Almost none of them know yet what they want it for. This article tries to answer that question before you spend any money.
The rule we use to decide
Before we propose anything, we ask three questions about the process you want to automate:
- How many times a month does it repeat? Below 20 or 30 repetitions, it’s rarely worth it.
- Can the logic for resolving it be explained? If nobody on the team can explain how it’s decided, the machine isn’t going to get it right either.
- What happens if it gets it wrong? Misclassifying an email isn’t the same as sending an incorrect invoice.
If the process is frequent, the logic can be explained, and a mistake is tolerable or reviewable, automating usually pays off. If any of the three fails, it’s almost always better to wait or to solve it another way.
Where it does work today
Handling frequent questions
An assistant trained on your information handles business hours, prices, availability, order status and repeat questions. In the Kintaro Restaurant project, the chatbot answers questions and manages reservations synced with Google Calendar, without anyone having to keep an eye on the phone in the middle of service.
Classifying and routing what comes in
Emails, forms and messages are read, tagged and sent to the right person. It’s about as profitable as it gets: low risk, high volume and immediate savings in dead time.
Extracting data from documents
Invoices, delivery notes, contracts and PDF forms. The AI reads the document and loads the structured data into your system. It replaces a task nobody wants to do, and one where people make more mistakes the more tired they get.
Preparing drafts, not publishing them
Replies, product descriptions, meeting summaries. The key is that a human reviews them before they go out. AI spares you the blank page; it shouldn’t have the last word.
A quick calculation: if a task takes you 5 hours a month and your internal hourly cost is €20, you save €1,200 a year. An automation that costs €800 pays for itself in eight months, and after that it’s margin. If it takes you 30 minutes a month, it isn’t worth the effort.
Where it still isn’t worth it
- Decisions with direct legal or financial consequences and no human review: invoicing, dismissals, diagnoses, credit approval.
- Tasks that change every time. If the process hasn’t stabilized, automating it is freezing the chaos.
- Low-volume processes. Automating something that’s done twice a month costs more than it saves.
- High-value business relationships. A customer about to sign a big contract wants to talk to a person. A bot there only gets in the way.
- When the real problem is something else. Often you don’t need AI: you need to get the process in order, and that’s solved with a much cheaper and more reliable traditional automation.
Traditional automation vs. AI
This distinction saves a lot of money, and almost nobody explains it:
| Situation | Best fit |
|---|---|
| The rules are fixed and known (“if X arrives, do Y”) | Traditional automation with n8n or Make. Cheaper, predictable and with no cost per use. |
| You have to interpret free text, images or natural language | AI. That’s where it brings what no rule can cover. |
| A mix of both | The usual approach: a traditional workflow with an AI step where judgment is needed. |
Most of the projects people ask us for as “AI” are better solved with a regular automated workflow and a single intelligent step where it’s truly needed. It costs less and breaks less.
How to start without taking risks
- Pick a single process, the most repetitive one and the one the team hates most.
- Measure how much time it takes up today. Without that figure, you won’t know whether it worked.
- Automate with human review for the first few weeks.
- Compare. If the savings are real, expand. If not, you haven’t lost much.
Be wary of anyone who proposes “implementing AI in the company” without first asking which specific process is costing you hours. The tool isn’t the goal.
Do you have a process that eats up your hours?
Tell us which one it is and we’ll tell you frankly whether it’s worth automating or not. If we think it won’t pay off, we’ll tell you.
Tell us about your situation →