SAS Tech Support handles a relentless stream of emails—customer issues, misroutes, and spam—making manual triage slow and inconsistent. See how we built a transformer-based text analytics pipeline on SAS Viya that automates classification with minimal data preparation, all while preserving privacy. We’ll walk through the end-to-end modeling workflow, including training, evaluation, and deployment of BERT-based models that achieved ~3% misclassification on hold-out data and uncovered labeling inconsistencies in human-tagged content. This modeling approach also provides a foundation for future integration into SAS’s broader agentic AI framework for automated CSM ticket triage.
A flood of emails hits support teams every day—but before solving problems, someone has to figure out what each message actually is. In this session, Ann Kuo shows how SAS turned that bottleneck into an AI-driven advantage with a streamlined, LLM-powered classification pipeline on Viya. You’ll see how transformer models quickly separate real customer issues from spam and misrouted requests, delivering faster routing with impressive accuracy and minimal preprocessing. More than just a modeling exercise, it’s a practical look at how AI can plug directly into a decisioning workflow to reduce delays and get customers the help they need sooner.
Watch the recording
