0
1.0.7
Zurich Patch 4
The primary aim of this application is to transform Conversation Evaluator’s evaluations (part of AI Control Tower) into long-term performance insights by analyzing conversation quality data over time. The approach focuses on identifying recurring patterns linked to low (and high) conversation quality and categorizing user requests into actionable themes using metadata-driven classification powered by large language models (LLMs).
1. Automated theme detection
- Uses a domain-specific LLM theme classification framework to tag KB articles, catalog items, VA topics, and agents into themes.
- Every evaluated conversation is classified under one of these themes.
2. Thematic insights and drill-down
- Provides breakdown of poor/good quality conversations (based on conversation evaluator scores) into top underlying themes.
- Users can drill down into a specific theme and review those conversations and also identify gaps in corresponding thematic content/catalog/agents.
3. Benchmarking
- Users are able to compare themes based on whether they lead to good or bad conversations.
4.Operational and strategic alignment
- Supports both daily operational tuning and long-term strategic planning.
- Bridges the gap between AI system performance data and business value realization.
New
- Initial release.
- Conversation Evaluator 1.1.10 (sn_na_conv_eval)
- Identity Agent Security (com.glide.identity.agent.security)
- ZP4 or YP8