spexa GmbH implements sustainable health and performance management in companies with the goal of promoting both operational efficiency and employee well-being. This is based on data collected from employees within the company. This data is used to calculate the proprietary, scientifically based Business Health Index (BHI®), which provides an aggregated and anonymized snapshot of the company’s status.

The AI-FIT network satellite project uses the HUMAINE Toolbox to analyze whether the questionnaire—which has thus far been administered as an online form—can be effectively supported and improved in terms of user acceptance through AI-based methods, and whether AI-based text analysis of free-text entries can be performed.

Data is collected through online surveys with predefined questions about health and the work environment. A comprehensive survey consisting of multiple-choice questions is specifically supplemented by opportunities for free-text responses.

Data quality and completeness play a crucial role in BHI calculations. This is where AI-based solutions come into play. Using the “Potential Analysis” and “Capturing User-Centered Interfaces” tools, three use cases were identified:

1. Chatbot to assist with comprehension issues – Goal: Provides explanations for questions or technical terms, improves understanding, reduces the dropout rate, and improves data quality.
2. Automated anonymization of free-text entries—Goal: removes personally identifiable information, preserves the meaning of the response, builds trust in data protection, and improves the quality of the data.
3. Avatar to Boost Motivation to Participate – Goal: To demonstrate the impact of participation on individual users and the company as a whole, and to provide information about other possible concrete steps the company can take in the area of workplace health management.

Focus on Use Case 2 – Pipeline Consisting of a PII Detector and an LLM

1. Identification and Masking of Sensitive Information
   (PII Detector + LLM)
2. Converting the masked text into continuous text (LLM)

The older woman in my department is often very unfriendly to me.

↓

[WORK_RELATION_ROLE] from [WORK_RELATION_ROLE] is very often very unfriendly to me.

↓

I very often encounter extreme unfriendliness toward me in my work environment.

Spexa GmbH, Essen

Bochumer Institut für Technologie, Bochum

Lehrstuhl Industrial Sales and Service Engineering, Ruhr-Universität Bochum

Institut für Neuroinformatik, Ruhr-Universität Bochum