Customer support · Production

Swisscom builds enterprise agentic AI for customer support and sales using Amazon Bedrock AgentCore

The problem

Swisscom faced the challenge of scaling AI agents enterprise-wide while managing siloed agentic solutions, ensuring cross-departmental coordination, and maintaining compliance with Switzerland's strict data protection laws — hitting what they called the 'automation ceiling' where traditional automation approaches could not meet modern business demands.

Workflow diagram · grounded in source
1
Customer request triggers agent
trigger
“The client sends a request to the Strands agent running on AgentCore Runtime, passing an authentication token from the Swisscom IdP.”
2
Token validation and downstream token generation
validation
“The client's token is validated and a new token for the agent's downstream tool usage is generated and passed back to the agent.”
3
Foundation model invocation and memory storage
ai_action
“The agent invokes the foundation model on Bedrock and stores the sessions in the AgentCore Memory. The traffic traverses the VPC endpoints for Bedrock and Bedrock AgentCore, keeping the traffic private.”
4
Cross-departmental API and agent access
integration
“The agent accesses internal APIs, MCP & A2A servers inside the shared VPC, authenticating with the temporary token from AgentCore Identity.”
5
Sales pitch or support resolution delivered
output
“generating personalized sales pitches, and 2) providing automated customer support for technical issues like self-service troubleshooting”
Reported outcome

Swisscom deployed two B2C agents — for personalized sales pitches and automated technical support — integrated into their existing SAM chatbot, handling thousands of requests per month each, with development teams delivering their first stakeholder demos within 3-4 weeks and one team migrating from LangGraph to Strands Agents citing reduced complexity.

Reported metrics
Time to first stakeholder demo3-4 weeks
Agent request volume per monththousands of requests per month each
Reported stack
Amazon Bedrock AgentCoreStrands Agents FrameworkAmazon SageMakerModel Context Protocol (MCP)Agent2Agent protocol (A2A)AWS Direct ConnectAmazon EKSRasaOpenTelemetryLangGraphSAM
Source
https://aws.amazon.com/blogs/machine-learning/how-swisscom-builds-enterprise-agentic-ai-for-customer-support-and-sales-using-amazon-bedrock-agentcore?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Swisscom deployed two B2C agents — for personalized sales pitches and automated technical support — integrated into their existing SAM chatbot, handling thousands of requests per month each, with development teams del…

What tools did this team use?

Amazon Bedrock AgentCore, Strands Agents Framework, Amazon SageMaker, Model Context Protocol (MCP), Agent2Agent protocol (A2A), AWS Direct Connect, Amazon EKS, Rasa, OpenTelemetry, LangGraph.

What results were reported?

Time to first stakeholder demo: 3-4 weeks; Agent request volume per month: thousands of requests per month each (source-reported, not independently verified).

How is this customer support AI workflow structured?

Customer request triggers agent → Token validation and downstream token generation → Foundation model invocation and memory storage → Cross-departmental API and agent access → Sales pitch or support resolution delivered.

Related customer support cases