TLDR: The research paper introduces Agentic AI, a proactive, goal-driven system that supports human agents in contact centers. A case study of Minerva CQ, a real-time Agent Assist product, demonstrates significant improvements in customer experience (CX) and agent efficiency. Minerva CQ integrates real-time transcription, intent detection, dynamic profiling, and AI-generated queries with FAQ caching, leading to a 38% reduction in average handling time, a 33% uplift in lead-to-enquiry conversion, and a 4.8% uplift in booking conversion in live production pilots.
Customer experience (CX) is a critical differentiator for businesses today, yet contact centers often struggle with long call times, low first-call resolution, and frustrated customers. Traditional AI tools for agent assistance are often reactive, relying on static rules or simple prompts, which places a heavy cognitive load on human agents.
A new approach, called Agentic AI, is emerging to address these challenges. Agentic AI systems are goal-driven, autonomous, and use tools proactively to support agents in real time. Unlike older systems, Agentic AI can identify customer intent, trigger specific workflows, maintain an evolving understanding of the conversation, and adapt dynamically as the discussion progresses.
This paper presents a case study of Minerva CQ, a real-time Agent Assist product specifically designed for voice-based customer support. Minerva CQ acts as an AI co-pilot, integrating several advanced capabilities. It uses real-time transcription, detects customer intent and sentiment, recognizes important entities (like names or account numbers), retrieves relevant information, builds dynamic customer profiles, and generates partial conversational summaries. These features enable proactive workflows and continuous context building throughout a call.
Minerva CQ distinguishes itself from typical Retrieval-Augmented Generation (RAG) platforms by using RAG as just one component within a larger, agentic framework. The system automatically identifies customer intent from live conversations and generates AI-suggested queries, allowing agents to find answers without manual typing. It also maintains a validated cache of frequently asked questions (FAQs), providing instant responses and reducing reliance on more resource-intensive LLMs.
During a call, Minerva CQ produces compact, incremental “partial summaries” that evolve with the conversation, capturing key facts and maintaining context. It also suggests intent-aware workflows and “next-best actions” to guide agents. At the end of an interaction, it automatically generates a call summary, estimates customer satisfaction (CSAT) and Net Promoter Score (NPS), and tracks sentiment throughout the call.
The system operates through an “observe → understand → decide → act → assist → learn” loop, ensuring continuity and goal-oriented support. Key features include real-time entity extraction, dynamic workflow triggering, proactive AI-suggested query generation with FAQ caching, partial conversation summaries, and live sentiment/CSAT/NPS tracking.
A live production A/B test involving 100 agents (50 with Minerva CQ, 50 without) over approximately 40,000 voice interactions demonstrated significant improvements. Agents using Minerva CQ achieved a 38% reduction in Average Handling Time (AHT), decreasing from 4 minutes 43 seconds to 2 minutes 55 seconds. There was also a 33% uplift in Lead-to-Enquiry (L2E) conversion and a 4.8% uplift in booking conversion. The FAQ cache alone saved agents an average of 6 seconds per query, avoiding costly RAG calls and cumulatively saving about 11.7 hours of latency across 10,000 calls.
Minerva CQ also supports multilingual and mixed-language conversations, including English, Italian, Hindi, and Hinglish. Its speech-to-text pipeline accurately transcribes these languages in real time and converts transcriptions into English text for AI-suggested prompts, ensuring all agents can understand the guidance without delay.
Also Read:
- Streamlining Financial Crime Compliance with Agentic AI
- Guiding AI: A Human-Centered Approach to Web Browsing Agents
The success of Minerva CQ highlights the benefits of moving from reactive agent assistance to proactive, agentic AI co-piloting. This approach not only improves agent efficiency and customer satisfaction but also generates strategic intelligence by analyzing customer pain points and sentiment trends, offering insights for continuous business improvement. For more details, you can read the full research paper here.
