Commerzbank

As one of Germany's largest financial institutions, Commerzbank AG faced mounting pressure to close the gap between rising customer expectations for instant, personalized digital service and the constraints of fragmented legacy infrastructure. Customers demanded round-the-clock access to account services — from credit card management to complex business payments — while the bank simultaneously navigated stringent European financial regulations, evolving fraud threats, and the need for full auditability in every interaction. Existing systems could not deliver the responsiveness customers expected without compromising compliance or security, creating a structural ceiling on the quality of digital customer service the bank could sustainably provide.

Commerzbank AG built Ava, a conversational AI agent powered by Microsoft Azure AI Foundry Agent Service and Azure OpenAI, implementing an agentic AI and autonomous workflows architecture designed specifically for regulated financial environments. Ava uses a distributed, multi-agent orchestration model running on Azure Kubernetes Service and Azure Container Apps, enabling elastic scaling across thousands of simultaneous interactions. Azure Cosmos DB maintains persistent memory and context synchronization across sessions, while Azure AI Content Safety with customized filters screens every interaction for sensitive data and fraud signals before reaching the model. Azure AI Search grounds all responses in trusted internal data sources. GitHub Copilot and Azure Foundry tooling were used throughout development, accelerating the build-and-release cycle with automated testing integrated directly into the DevOps pipeline. Ava progressed from prototype to production managing the full credit card lifecycle, balance inquiries, savings account requests, and business payments — with structured escalation paths to human agents for complex cases.

Ava now handles more than 30,000 customer conversations per month, resolving approximately 75% of requests fully autonomously with 24/7 availability — eliminating the resolution bottleneck that previously required human agent intervention for the majority of routine inquiries. Development velocity doubled compared to prior approaches, attributed to AI-assisted coding, streamlined CI/CD pipelines, and automated testing. Qualitatively, the deployment freed human agents to focus on relationship-driven and high-complexity cases rather than routine service tasks. Key outcomes include:

• 30,000+ monthly conversations handled end-to-end by Ava
• 75% autonomous resolution rate across credit card, savings, and payment workflows
• 2× development speed improvement via AI-assisted DevOps
• Established a reusable, auditable AI framework being extended to lending, business banking, and internal operations

• Compliance-first architecture pays dividends: Embedding Azure AI Content Safety and customized filters from day one — rather than retrofitting — allowed Commerzbank to meet regulatory requirements without slowing deployment or limiting capability.
• Modular, multi-agent design enables incremental scaling: Specialized sub-agents handling distinct business domains (credit cards, savings, payments) let the team expand scope without rearchitecting the core system.
• Persistent memory is table stakes for banking AI: Azure Cosmos DB context synchronization ensured conversation continuity across sessions — critical for customer trust in high-stakes financial interactions.
• Persona and empathetic design drive adoption: Investing in Ava's tone and personality alongside technical capabilities was a deliberate strategic choice that directly influenced customer engagement rates.