Zeiss links DORA metrics and FinOps through Backstage and Argo CD

Zeiss has successfully integrated DORA metrics with FinOps practices using Backstage and Argo CD, marking a significant step toward unified operational visibility. The company addressed the challenge of attributing deployments, services, and costs accurately to teams, systems, and organizational units. This integration enables reliable performance assessments of software delivery and transparent cost analysis, essential for both DevOps efficiency and financial accountability. The initiative began with the recognition that cloud platforms generate vast amounts of telemetry, logs, metrics, and cost data. However, without clear attribution mechanisms, these data points fail to deliver actionable insights. Alexander Troppmann, during his presentation at the Mastering GitOps conference on June 25, 2026, outlined how Zeiss tackled this issue through metadata management. He emphasized the need for consistent attribute assignment, such as deployment ownership, system domain, environment, and cost responsibility, to ensure meaningful analytics. Initially, Zeiss operated in three distinct domains: Backstage as a service catalog with ownership information, GitOps repositories defining desired deployment states, and Kubernetes/Argo CD managing actual runtime environments. Without seamless metadata propagation across these layers, inconsistencies arose. These led to unreliable DORA metrics and ambiguous cost reports, making it difficult to trace failures or allocate expenses accurately. To resolve this, Zeiss leveraged Backstage as its central source of truth for attribution. It maintains dimensions such as team, system, component, environment, cost user, and cost center. These attributes form the foundation of all analytical reporting. Future enhancements include adding SRE owners, criticality ratings, and scalability profiles. While some metadata, like team and component assignments, are crucial for DORA metrics, others, like cost user and tenant, are more vital for FinOps, supporting internal billing, budgeting, and cost clustering. To bridge the gap between these domains, Zeiss developed custom tooling. The Argo CD Annotator runs as a Kubernetes CronJob, cloning GitOps repositories to identify Argo CD application manifests. It retrieves metadata from Backstage and injects them back into the manifests via Kustomize components and JSON patches. This approach ensures that GitOps configurations remain untouched, preserving clarity in version control while maintaining platform-level attribution. Another key tool is the DORA Metrics Collector for Argo CD, which operates as a Go-based microservice within the cloud environment. It monitors Argo CD’s Application CRDs through Kubernetes informers, classifying deployments, failures, and recoveries deterministically. The collector exports Prometheus metrics and provides REST endpoints for Backstage, Grafana, and management dashboards. Its codebase includes over 15,000 lines of production code, nearly equal amount of test code, more than 600 test functions, 13 architecture decision records, and nine specifications. The implementation remains fully deterministic, free from reliance on artificial intelligence. Defining a successful deployment at Zeiss is particularly precise. The company has established clear criteria to determine whether a deployment meets its standards, ensuring consistency and reliability in its operations. This level of precision supports continuous improvement and aligns with broader goals of operational excellence and financial transparency.