Build an Agentic AI IVR That Shapes Itself Per Client using Amazon Connect Customer

This article demonstrates how to build an agentic AI IVR using Amazon Connect that dynamically adapts its behavior, tools, and business logic per client from a single deployment, eliminating the need for hundreds of separate contact flows.

• Single Amazon Connect deployment serves multiple clients with distinct configurations stored in DynamoDB
• Amazon Bedrock AgentCore orchestrates tools dynamically based on client-specific system prompts and available capabilities
• Amazon Connect Customer AI Agent with Nova Sonic provides natural, context-aware conversations with full session continuity
• Payment handoff isolates card data collection in a separate deterministic flow while preserving AI conversation context
• Per-client knowledge base segmentation prevents cross-client information leakage through document-level metadata filtering
• Tool Lambdas accept clientId parameter for multi-tenant routing without code duplication
• Onboarding new clients requires only a DynamoDB row addition—no code changes or new infrastructure
• Reference implementation includes parking/toll violations use case with 7 tool Lambdas and full CloudFormation deployment

This configuration-driven approach eliminates IVR sprawl, accelerates client onboarding, and improves caller experience through natural conversation while maintaining strict data isolation and security guardrails.