Partou gains insight into complex HR environment through AI

In recent months, childcare organization Partou, together with Tacstone Technology, has conducted a proof-of-concept with UiPath Communications Mining within the HR absence team. This gives Partou better insight into the large flow of emails surrounding absenteeism, and provides a good basis for further automation opportunities that structurally reduce workloads.

Absenteeism is a structural issue in child care. While absenteeism in this sector fell more than a percentage point in the second quarter of 2025 compared to the first quarter, it is still high at 7.1 percent nationwide. This creates a continuous flow of communication between employees, managers, occupational health and safety service and HR absence directors.

“It’s not a local problem ,” says Dion Morskieft, Head Center of Excellence Intelligent Automation. “Throughout the childcare industry, absenteeism is unfortunately high. As a result, the administrative pressure on teams like HR absenteeism is also structurally high.” Yet for a long time, there was little visibility into the exact composition of that workload. The absence mailbox functioned as a central hub, but figures on numbers and types of emails were lacking.

“Our manager knew we were very busy, but not exactly with what,” says absentee worker Robin Blokker . “Colleagues work per region and thus fairly independently. You don’t have an overall picture of how much mail is coming in on which subject. It remained mainly a feeling.” Exactly that issue made the absence mailbox suitable as a starting point for a proof-of-concept with communications mining.

From abstract idea to concrete case

Within Partou, there had long been interest in communications mining, but it initially remained an abstract concept.

“We’ve talked to Tacstone more often than not about how we could apply UiPath Communications Mining and what the added value would be.”

Dion Morskieft | Head Center of Excellence Intelligent Automation at Partou

“But until you choose a concrete case, it remains theoretical.”

So it was important to choose a topic where automation adds value right away. “Absenteeism is always a hot topic,” says Morskieft. “That’s where a lot of different types of emails come in. In consultation with the managers, it was then determined that the absenteeism team was the most appropriate to start with.”

That required a structured approach. Blokker. “We first started thinking together about what categories we would probably need, such as company doctor, travel expenses, administrative questions. Then we started working on labeling.”

Model training

Communications Mining is not a ready-made AI product; it also needs manual work. Morskieft: “You have to create your own decision tree with categories and subcategories, and then the real work begins: training the model.” That training meant: loading large numbers of historical e-mails, and manually labeling a substantial portion of them. In the pilot, about 10,000 emails were put into the system, of which about 2,000 were manually categorized.

Once enough data was labeled, the Communications Mining model was able to show a reliable dashboard for the first time. This provided a concrete and in some ways surprising picture.

“We found out that about 33 percent of all incoming emails were about the company doctor.”

Robin Blokker | absentee worker at Partou

“We hadn’t estimated that so highly beforehand. We knew it was a lot, but not that it was such a big part of the total work flow.”

In total, the absence mailbox involves an order of magnitude of 10,000 to 15,000 emails per year. That means there are thousands of emails specifically related to appointments, changes and questions around the company doctor.

Two processes emerged as logical first steps. First, automatically processing cancellations of appointments with the company doctor. Second, automatically labeling and assigning emails to the appropriate region, operational area or an absence director, who supervises the entire process. These automations are now running in the absence mailbox.

‘Living product’

An important conclusion of the pilot is that communications mining is not a static solution. Whereas a classic RPA solution often remains relatively stable after implementation, a communications stream is constantly changing. “Communications mining is a living product,” says Morskieft. “You really have to stay with it. New topics, collective bargaining changes, seasonal effects – they cause your categories and labels to have to move with you.”

According to Blokker, it is therefore important for organizations not to think that the work ends after the initial round of labeling.

“If you deploy it well and you address your biggest problems, then over time they partially resolve themselves or shift. Then new categories emerge.”

Robin Blokker | absentee worker at Partou

“Then when you stop labeling, you actually do it for nothing. You may have solved one problem, but you lose sight of the rest.”

HR Services as the next step

The experience at absenteeism has led to further interest from other HR teams within Partou. The HR service desk has now signed up as the next candidate for communications mining. “There the focus initially is not on automation, but on insight,” says Morskieft. Only once that insight is there will we look at whether and where automation makes sense.
Tacstone Technology is Partou’s discussion partner in communications mining. Morskieft: “They help us in the choice of licenses, they support initiatives such as communications mining and now also in migrating robots to the cloud. For us, it’s nice to have one party that understands the whole spectrum of UiPath.”

Ebert Knol, Managing Partner at Tacstone Technology, says: “The success of this POC is really due to Partou. The collaboration went very smoothly, and all the expertise was present in the project team. Both on the content of e-mails and for developing the automations. The fact that Partou is now rolling out the UiPath Communications Mining product to other departments is proof of a successful POC!”

Lessons

The communications mining project yielded several concrete lessons. First, properly map existing mailbox practices. “With us, the absentee team already had its own system of flags and colors,” says Morskieft. “If you analyze that first, you can tailor your AI categories accordingly.”

Second, don’t underestimate the labeling effort. “You’re not done with a few hundred mails,” Blokker said. “We loaded ten thousand mails and labeled two thousand, and actually you would have to do even more for a really mature model.”

Third, choose a few clear quick wins in advance. “You can’t do everything at once,” says Morskieft. “By first looking at where the biggest volume flows are, you can target your automation capacity.”
And finally: accept that it’s never “finished. “Communications mining should be seen as an ongoing process,” Blokker summarizes. “As questions change, your categories have to change with them. That’s not a fault of the system, but exactly why you deploy it: you want to be able to see how your world is shifting.”

Overview of the model’s own rating, and recommended training to improve the least performing labels.