AWS re:Invent 2025-How enterprises are deploying PolyAI's voice agents at scale to reduce handling..

Summary of AWS re:Invent 2025 Presentation: How Enterprises are Deploying PolyAI's Voice Agents at Scale

Introduction to PolyAI

• PolyAI was founded in 2017, working with large language models and transformer-based architectures for enterprise customer service
• The co-founders were researchers at the University of Cambridge studying dialogue and speech recognition systems
• PolyAI has deployed over 2,000 individual AI agents for more than 100 enterprise customers
• The company has raised over $120 million in venture capital and is recognized in Gartner's Magic Quadrant for Conversational AI platforms
• PolyAI is built on AWS and available on the AWS Marketplace, with multiple live deployments integrated with Amazon Connect and other contact center platforms

Case Study: Deploying PolyAI's Voice Agent at Pacific Gas & Electric (PG&E)

• As a utility company, PG&E had critical customer interactions around outages and billing that were previously handled by a legacy IVR system
• Customers were experiencing over an hour of wait time during unpredictable and sensitive periods, and the legacy system was difficult to adapt
• PolyAI helped PG&E migrate 7,000 intents from the legacy IVR to a Generative AI (GenAI) solution, resulting in:
  • 67% call resolution
  • 25% reduction in overall customer effort
  • 22% increase in customer satisfaction (CSAT) during outages by scaling the AI agent 50x within 5 minutes

Key Capabilities for Deploying Enterprise-Grade Voice AI Agents

1. AI Agent Experience:

   • Optimized speech-to-text models, including ensemble models for different accents, languages, and use cases
   • Adaptive voice activity detection to improve conversational flow
   • Ability to seamlessly integrate multiple modalities (voice, forms, photos, videos, etc.) while maintaining voice as the primary interface
   • Custom, domain-specific large language models (LLMs) trained on contact center conversations for more natural, engaging responses
   • Architectural flexibility to address data processing and hosting requirements of enterprise customers

2. Platform Capabilities:

   • Comprehensive observability, including custom metrics, conversation state tracking, and detailed transcripts and recordings
   • Collaboration features like multi-user access, environment management, audit trails, and merge capabilities for concurrent development
   • Natural language-based analytics and insights, leveraging technologies like Amazon Quicksight and Anthropic's Bedrock

3. Operational Model:

   • Forward-deployed engineering teams to provide expertise in architecture, solution design, and dialogue optimization
   • Dedicated product and solution managers to guide the continuous improvement of the AI agents
   • In-house dialogue designers to help enterprises implement their brand experience into the AI agents

Key Takeaways

• PolyAI has demonstrated the ability to deploy voice AI agents at true enterprise scale, taking on the work of over 1,000 full-time employees (FTEs)
• Achieving this level of scale requires a comprehensive approach, addressing the AI agent experience, the platform capabilities, and the operational model
• Specific technical capabilities like optimized speech recognition, multi-modal interactions, and custom LLMs are critical for delivering an engaging customer experience
• Robust platform features, such as observability, collaboration tools, and natural language-based analytics, enable enterprises to effectively manage and improve the AI agents
• The operational model, with forward-deployed experts and a focus on continuous improvement, is essential for ensuring the long-term success of enterprise-grade voice AI deployments