In short
Automation isn't worth it when the repetitive manual work adds up to only a few hours a week, when the data exists only on paper or in colleagues' heads, and when a chatbot is wanted only because everyone has one. In those cases, sort out the process and the data first. Automation pays off where it replaces a lot of regular, written-down work.
Key points
- Building, testing and maintaining an automation is work too, and saving a few hours a week rarely pays that back quickly.
- Artificial intelligence can only work with data that is written down: whatever is on paper or in someone's head has to go into a system first.
- A chatbot is useful when it answers specific, frequently repeated questions and a person is reachable behind it.
- In our view, a good assessment answers two questions: what is worth automating, and what isn't. The second answer is worth just as much.
When is a process not worth automating?
A process isn't worth automating if building and maintaining the system costs more time and money than it saves within a reasonable period. We see this most often in three situations: there is little repetitive work, there is no written-down data, or there is no real task behind the system.
Sometimes we talk people out of automating for exactly this reason. We don't say it at the end of a long assessment, but in the first conversation, if it is already clear.
This is not a rare conclusion. According to Eurostat data published in December 2025, 20.7% of EU companies that considered artificial intelligence but don't use it gave up on it because they saw no use for it in their business. The most common reason was lack of expertise, at 70.9%.
How little manual work is too little for automation to pay off?
There is no single threshold, but if a task takes only a few hours a week, automation rarely pays off quickly. Alongside the time saved, the calculation has to include building, testing, maintenance and keeping the rules up to date.
For small volumes a simpler tool is often enough: a good template, a shared spreadsheet, a formula, or a setting in the software you already use.
What if the data exists only on paper or in colleagues' heads?
If the data exists only on paper or in someone's head, it has to be recorded first, and automated only after that. Artificial intelligence can only work with what is written down. If only the warehouse manager knows the rules for the exceptions, the system won't know them either.
According to a Gartner forecast from February 2025, by the end of 2026 organisations will abandon 60% of the AI projects that are not backed by data artificial intelligence can use.
In that case, we suggest this order:
- 01
Write down who does what, in what order and with what exceptions.
- 02
Put the data in one place: a shared spreadsheet, a form or the system you already have.
- 03
Work like that for a few weeks and watch where the process gets stuck.
- 04
If there is still a lot of repetitive, rule-based work after that, it is worth automating.
This has a side effect too: a written-down process works even when the colleague is on holiday. No automation will write that knowledge down for you.
When does a company not need a chatbot?
If you can't say which questions it would answer regularly, you don't need one. A chatbot is a chat program that answers customers on the website or in a messaging app. It is useful because it resolves a frequent, repetitive question faster than a person, not because a competitor has one.
Customers don't necessarily expect one either. According to a Gartner survey of 5,728 customers carried out in December 2023, 64% of customers would prefer that companies didn't use artificial intelligence in customer service, and 53% would consider going elsewhere because of it. Their biggest concern was that it would be harder to reach a person.
Before you order a chatbot, go through a month of incoming e-mails and calls. If the same few questions keep coming, an automatic reply, a better FAQ page or a chatbot with a person reachable behind it may be the answer. If every enquiry is different, you need a person.
When is automation worth it after all?
It is worth it when there is a lot of repetitive work that can be described with rules, and the data is already in a system. This table shows the difference:
| Signal | Better wait | Worth starting |
|---|---|---|
| Repetitive manual work | a few hours a week | many hours a week, across several colleagues |
| Data | on paper, in people's heads, in many formats | in a system, in one format |
| Rules | every case is different | can be written down, few exceptions |
| Goal | “because everyone has one” | measurable: time, errors, turnaround |
| The process | still changes every month | stable, everyone does it the same way |
- Repetitive manual work
- Better waita few hours a week
- Worth startingmany hours a week, across several colleagues
- Data
- Better waiton paper, in people's heads, in many formats
- Worth startingin a system, in one format
- Rules
- Better waitevery case is different
- Worth startingcan be written down, few exceptions
- Goal
- Better wait“because everyone has one”
- Worth startingmeasurable: time, errors, turnaround
- The process
- Better waitstill changes every month
- Worth startingstable, everyone does it the same way
What does this mean for a company of 10 to 250 people?
A company of 10 to 250 people has little spare money and attention, so the aim is not to automate as many processes as possible, but the one that brings the most. That is why we start with an assessment.
The AI Potential Audit takes two weeks. At the end you know what is worth automating, and also what isn't. The second list is worth at least as much as the first, because it shows where money and attention should not go.
Where it pays off, we start with one process, from start to finish: we call this the Sprint. Where money or customers are at stake, validation rules, a log and human approval go into the process. We quote a fixed price after the first conversation.