In the era of distributed microservices, traditional approaches - such as anonymising production data - have reached their breaking point. As data ownership fragments across independent services, maintaining referential consistency becomes almost impossible task, turning staging environments into a bottleneck for every release.

While synthetic data is a safer alternative, generation alone is no longer the hardest part. The real challenge is provisioning: how do you create realistic, consistent datasets that span multiple teams, business flows and systems without breaking integrity? Local scripts and shared databases simply don’t scale.

This talk introduces Test Data as a Service, an architectural shift where each domain team exposes supported APIs to create and evolve its own data, strictly following real business logic and call order. On top of these domain APIs, the speaker enable discovery and self-service, allowing teams and pipelines to request test data without relying on shared databases or brittle scripts.

The key opportunity lies in composition. By orchestrating and chaining data-creation APIs across services, teams can build complex, end-to-end scenarios on demand while preserving consistency, isolation, and team ownership.Finally, the talk dives into a Just-in-Time test data approach, where pools of pre-created base entities are extended at execution time to fit specific scenarios - dramatically improving speed, parallelisation, and test reliability.

Based on real-world experience, this session provides a practical mental model for evolving test data management from static datasets to dynamic, API-driven services that scale with modern architectures.