More efficient and consistent classification of collective agreements at Partena using AI
AI for Partena Professional
More time savings
Partena Professional is a Belgian HR service provider and social secretariat that supports companies and self-employed individuals with personnel administration, payroll processing, and HR policies. They handle large quantities of employment and social documents daily, such as collective agreements (CAO’s), and therefore have a strong need for efficient and reliable information processing.
Focus on content and advice
Considering the large number of collective labour agreements they see passing by daily and the repetitive, labour-intensive, and time-consuming work that employees have to perform, a solution was sought that would primarily deliver efficiency gains.
The goal was not to replace people, but to relieve employees from dull and time-consuming work, so they can focus on tasks with greater substantive or advisory value.
How it works
The staff at Partena have a central location where companies can upload their new Collective Labour Agreements (CLAs) into a general folder. From this environment, an overview was first created of all new documents that had not been processed before to avoid duplicates. Subsequently, Azure Document Intelligence was used to automatically extract the full text of the CLAs via public links.
Using an AI model and targeted prompts that guide the AI's behaviour, it can be determined which information should be extracted from the documents.
A fixed set of themes was established that Partena uses for their classification. What was asked?
- A brief summary of the contents
- To highlight the main changes
- To indicate which sectors the CLA applies to
- The date when the CLA comes into effect.
All this enriched information was gathered in an Excel file that is clear and easy to read. To make the results even more insightful, the assigned themes were visually supported with colour codes. Green, yellow, and red indicators show the level of certainty with which the AI assigned a particular theme. The threshold values for these certainty levels could be set by the users themselves via parameters.
In a further phase and close collaboration, several small optimisations were later implemented to increase ease of use, such as applying abbreviations, shortening descriptions, and other practical improvements. This way, the solution perfectly aligns with the daily operations of the users.