In SMEs, the problem is not always a lack of AI. It is that too many people act as human APIs: they take a piece of data from an email, copy it into Excel, then enter it into the business system.
Every step requires attention and can introduce an error. The person’s contribution gets absorbed by transferring information rather than doing the work that information is meant to support.
A measurable problem
Research presented in March 2026 by the Intelligent Business Process Automation Observatory at Politecnico di Milano reports that repetitive tasks account for 22% of working time in SMEs, while 15% of these businesses use process automation technologies.
Politecnico di Milano’s research on process automation
Questions to ask before choosing an agent
-
Why is the same data entered more than once?
-
Why does the status of a task exist only in one person’s head?
-
Why do two software systems not communicate?
-
At which step do errors actually arise?
The answers show where to intervene. If structured data needs to move between two systems, an integration may be enough. If a free-form request needs interpreting or information needs extracting from a document, an AI model may be part of the solution.
Make the workflow clear first
You still need to know which data is authoritative, who can change it, and what should happen when something does not add up. Adding AI to a confused process risks making its inconsistencies move faster too.
I would start where a person acts as a bridge between two systems. I would observe what they copy, what they check, and which decisions they make. These are different activities: automating them all in the same way would be a mistake.
Adapted from one of my LinkedIn posts and added to the blog on 10 September 2026. The date shown refers to publication on this website.