Swisscom Empowered Support and Sales Teams: How an AI Agent Tackled Routine and Scalability

Swisscom, Switzerland's leading telecommunications provider, faced the challenge of scaling AI solutions: traditional automation approaches hit what they called the "automation ceiling." Instead of creating disparate chatbots for each department, the company implemented an AI agent framework that allowed support and sales teams to develop new tools in just 3-4 weeks, while ensuring seamless integration and adherence to strict security and data privacy requirements.

In a large company, especially in telecom, where millions of requests and data are processed daily, the problem isn't just about launching AI, but about making it scalable, secure, and truly useful for different departments. Disparate solutions that don't communicate with each other quickly turn into chaos, and developing each new agent from scratch becomes an endless cycle of costs and delays. But now, frameworks exist that can solve this problem systematically.

The Reality of the Problem: "Automation Ceiling" and Siloed Solutions

Swisscom, like many large enterprises, was already using AI solutions, including conversational AI based on Rasa and fine-tuned large language models. However, as the scale grew, new challenges emerged. The main problem was that each new AI solution was created as a separate "silo." This led to duplicated efforts, difficulties in cross-departmental coordination, and inefficient data usage. For instance, when a customer had an internet connectivity issue, the cause could be a billing problem, a network outage, or a router misconfiguration. These issues fell under different departments, requiring complex interaction between agents and systems that individual chatbots couldn't provide.

The company encountered an "automation ceiling," where traditional approaches couldn't meet modern business demands, especially under Switzerland's strict data protection laws. A solution was needed that would allow AI agents to be scaled securely and efficiently across the organization, ensuring their interaction and reusability.

The Path to AI Agents: From Isolated Bots to Orchestration

Previously, Swisscom used isolated solutions for specific tasks, such as the SAM chatbot for customer interaction. This worked but did not allow for the creation of truly intelligent systems capable of solving complex tasks requiring cross-functional interaction. For example, when a customer called with a problem, it required one agent to retrieve data from the billing system, another to check network status, and a third to suggest troubleshooting steps for the router. The company needed not just automation of individual functions, but orchestration of multiple agents capable of communicating with each other and collaboratively solving tasks.

This is why Swisscom turned to an AI agent framework: not just another AI model, but an entire ecosystem of agents capable of complex interaction.

Designing the AI Agent Framework

Swisscom designed a framework that would allow for centralized management of AI agent deployment and interaction. Key elements included:

• Interaction Protocols. Model Context Protocol (MCP) and Agent2Agent (A2A) protocols were developed for seamless communication between agents from different domains. This allowed agents to exchange information and collaboratively solve tasks without being tied to a specific department.
• Security and Authentication. Under Switzerland's strict data protection laws, security was a priority. Each agent and every transaction required temporary access tokens that validated both the customer and agent context. This ensured the principle of least privilege and controlled resource sharing.
• Unified Runtime Environment. A runtime environment was chosen for hosting and scaling agents, providing secure and cost-efficient deployment of containerized agents while maintaining session-level isolation. This allowed teams to focus on agent logic rather than infrastructure.
• Memory Management. For B2C operations, understanding customer context across multiple interactions is critical. A memory management system was implemented to store both short-term and long-term user information, ensuring a personalized approach and maintaining data privacy.
• Integration with Existing Systems. The framework was designed to integrate with SAIL (Service and Interface Library) — Swisscom's central system for API hosting and service integration, as well as with the corporate network via AWS Direct Connect.

Implementation: From Pilot to Scale

Implementation began with two key B2C scenarios: generating personalized sales pitches and automated customer support for technical issues. These agents were integrated into Swisscom's existing SAM chatbot system, which required high-performance agent-to-agent communication protocols due to the large volume of customers and strict latency requirements.

Thanks to the framework, which handled security, scalability, and integration, development teams could focus on business logic. This allowed them to achieve their first stakeholder demos in just 3-4 weeks, even without prior experience with the chosen tools. Moreover, one project team successfully migrated from another implementation to the new framework, noting reduced complexity and faster development cycles. Integration with monitoring tools enabled performance tracking and rapid concept validation.

Results

By implementing the framework, Swisscom achieved significant results:

• Accelerated Development. Support and sales teams were able to develop and demonstrate new agents in just 3-4 weeks.
• Scalability. Agents successfully handle thousands of requests per month, maintaining low latency and optimizing costs.
• Improved Customer Interaction. The agents' long-term memory allows tracking and analyzing customer interactions, continuously enhancing their experience.
• Enhanced Security. Identity management provides precise access control, ensuring agents only access authorized resources for specific interactions.
• Reduced Complexity. It became easier for development teams to create and maintain agents, as evidenced by migration from more complex solutions.

How to Implement This in Your Business

The Swisscom case demonstrates that successful implementation of AI agents in a large company requires not just a set of tools, but a complete architectural foundation. To replicate this success in your own business, start with the following steps:

• Build an Architectural Foundation. Instead of creating disparate chatbots, focus on building a framework that provides secure cross-organizational authentication, standardized agent orchestration, and comprehensive observability.
• Ensure Seamless Integration and Agent Interaction. Develop protocols for agent-to-agent communication and centralize access to internal APIs and systems. This will allow agents to collaboratively solve complex tasks and exchange information.
• Prioritize Security and Privacy. Implement robust authentication and authorization mechanisms that ensure the principle of least privilege and comply with regulatory requirements. This is especially crucial for companies handling sensitive data.
• Choose a Framework That Accelerates Development. Use tools that simplify agent creation, accelerate development cycles, and integrate easily with existing monitoring and evaluation infrastructure. This will allow teams to quickly deliver business value.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager