Extracting trends from technician feedback using Large Language Models

To analyse Proximus’ technician feedback, Möbius applied AI technology, transforming how the company derives insights from on-site service experiences.

CASE_proximus logo
Proximus Large Language Models

Proximus, a leading telecommunications company, faces the challenge of extracting valuable insights from the feedback provided by technicians after customer visits. These technicians typically report on actions taken or problems solved, such as repairing decoders, replacing modems, or identifying technical issues like oxidation on cables or short circuits.

Crucially, technician feedback contains valuable information about problems that are suddenly arising. Detecting these anomalies early using AI can ensure that service remains optimal. Spelling errors, technical jargon, abbreviations, and the fact that texts are written in both French and Dutch, lead to a primary obstacle. The sheer volume of feedback, exceeding 15,000 texts, further complicates manual analysis.


Strategic challenge

The strategic challenge was to decipher trends and detect emerging problems from these unstructured texts, enabling Proximus to proactively address issues. This task was beyond the scope of manual analysis due to the poor quality and high volume of the data.



The project involved key steps using large language models (LLMs):

1. Utilising AI for text analysis and translation

LLMs were employed to extract relevant information from the technicians' feedback, converting these unstructured texts into a structured format in English. This process involved identifying specific actions mentioned in the feedback.

2. Clustering analysis for structured insight

The structured texts were subjected to clustering analysis, grouping similar actions together. LLMs then identified the central theme in each cluster, categorising the technical feedback into specific topics.

3. Trend analysis and anomaly detection

With the feedback organised into clusters, it became possible to track changes over time. This analysis identified problems with increasing trends, highlighting areas needing further examination.



The integration of AI in analysing technician feedback led to several advancements:

  • Enhanced data structuring: The conversion of unstructured, bilingual texts into a structured, standardised format in English.

  • Clarity in trend analysis: The clustering analysis provided clear insights into recurring issues and emerging trends.

  • Proactive problem-solving: This system enables Proximus to anticipate and address problems more effectively, potentially reducing future customer service issues.



The application of AI and LLMs in processing technician feedback marks a significant leap forward for Proximus in understanding and improving its field services. Despite the small scale of the project, it demonstrates how AI can extract valuable insights from even the most unstructured data, setting a precedent for similar applications in various industries.