What challenges did U.S. Bank face before AI implementation?

Agents faced high operational burden from manually searching multiple knowledge bases during real-time voice calls, impacting efficiency and customer satisfaction. Call transfers increased handling times and operational costs when customer queries fell outside an agent's skill set. Post-call case documentation consumed significant agent time, reducing availability for customer service.

What was the impact of the AI solution?

Improved agent efficiency with real-time AI assistance and accurate intent detection. Reduced manual search time and call transfer rates, enhancing customer experience. Automated post-call documentation processes reducing administrative burden. Scalable pilot foundation with ongoing monitoring, evaluation, and potential for expanded AI use cases including multi-agent frameworks and customer self-service voice AI.

Source

ProductionEvidence: Low40/100

U.S. Bank Enhances Contact Center Operations with Generative AI Using Amazon Q and Amazon Bedrock

Use case typeAI platformUpdated Jun 13, 2026

U. S. Bank implemented a generative AI solution using Amazon Q in Connect and Amazon Bedrock with Anthropic's Claude model to improve real-time voice-based contact center operations. The AI system provides real-time call transcription, intent detection, and tailored knowledge base recommendations to agents, reducing manual searches, improving call handling, minimizing transfers, and automating post-call documentation. The pilot leverages Amazon Contact Lens for transcription and speech analytics, Amazon Q as the AI orchestrator, and Amazon Bedrock for AI response generation with multi-KB management and guardrails ensuring compliance with financial regulations. U. S. Bank maintains multiple specialized knowledge bases and restricts AI searches based on agent skill sets and call routing to ensure accurate and context-appropriate recommendations. The implementation includes automated data cleansing, extensive prompt engineering, and operational guardrails to ensure security, accuracy, and domain-specific appropriateness. The production pilot is limited in scope for controlled rollout with continuous monitoring and iterative improvement, aiming for enterprise scaling and advanced multi-agent AI capabilities.

Organization: U.S. Bank
Industry: Finance
Location: United States

Reported outcomes

Strategic outcomes

Speed & agilityImproved real-time agent assistanceCustomer experience & trustReduced call transfers and search effortSpeed & agilityAutomated post-call documentationScale & capacityCreated foundation for expanded AI use

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: U.S. Bank
Provider: AWS
Maturity: Production
Linked source: ZenML AI Use Case Database

The implementation includes automated data cleansing, extensive prompt engineering, and operational guardrails to ensure security, accuracy, and domain-specific appropriateness

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 3 of 3

1Real-time AI Assistance
2Generative AI
3Contact Center Automation

Agents faced high operational burden from manually searching multiple knowledge bases during real-time voice calls, impacting efficiency and customer satisfaction.
Call transfers increased handling times and operational costs when customer queries fell outside an agent's skill set.
Post-call case documentation consumed significant agent time, reducing availability for customer service.

Used Amazon Contact Lens for real-time call transcription and speech analytics to identify conversation topics and issues.
Amazon Q in Connect orchestrates generative AI functions, performing selective intent detection to trigger AI assistance when relevant.
Multi-knowledge base search with tagging restricts AI queries to appropriate domains based on agent skills and call context.
Amazon Bedrock with Anthropic's Claude model generates AI-driven recommendations with RAG architecture for low latency.
Implemented extensive prompt engineering and guardrails for regulatory compliance and security.
Adopted a human-in-the-loop design enabling agents to request AI suggestions selectively during live calls.
Maintained data quality through automated cleansing pipelines to ensure reliable AI responses.

Technologies

Amazon Q Amazon Bedrock Amazon Connect Amazon Contact Lens AWS Lambda Amazon S3

Improved agent efficiency with real-time AI assistance and accurate intent detection.
Reduced manual search time and call transfer rates, enhancing customer experience.
Automated post-call documentation processes reducing administrative burden.
Scalable pilot foundation with ongoing monitoring, evaluation, and potential for expanded AI use cases including multi-agent frameworks and customer self-service voice AI.

Architecture

Multi-component AWS architecture integrating Amazon Contact Lens for transcription and speech analytics, Amazon Q in Connect as generative AI orchestrator with intent filtering, and Amazon Bedrock with Anthropic's Claude model for real-time AI response generation. Uses tagging for knowledge base selection, RAG vector search, and extensive prompt engineering and security guardrails tailored for financial regulations.

Sources & evidence1

Evidence: Low40/100Evidence strength