Accurate testing requires accurate test data. Yet many teams struggle to achieve this balance.
Production data cannot be used due to compliance, while synthetic data often breaks application logic and lacks the details needed to expose real defects. Anonymisation promises a middle ground, but in practice it is highly technical, resource-intensive, and demands deep knowledge of the system under test.
The result is a persistent paradox: either data is safe but unrealistic, or realistic but unsafe.
In this talk, I will draw on my experience leading development and testing in regulated environments, working with cryptography, anonymisation, and compliance. I have seen how lack of realistic test data leads to false confidence: software appears to work in test but fails in production.
I have also seen how compliance, while necessary, can become an obstacle that slows delivery and frustrates teams.
I will present a different approach: generating test data that is safe, valid, and practical without requiring major resources or deep specialist skills. The key is automation: production data is scanned to profile structures, distributions, and sensitivity; results are combined with the organisation’s risk appetite and policies; and transformation rules are applied.
Non-sensitive fields can be reused directly, sensitive but relationally important fields pseudonymised or encrypted, and non-critical filler fields generated synthetically. This layered, policy-driven method preserves the quirks and relationships of production data while remaining compliant and secure.
The approach is in use with organisations from finance to government. Previously, teams either had non-compliant test data or datasets so artificial they were meaningless. Now they work with compliant data that looks like production, where systems function correctly and all relations are intact.
Because profiling and transformation are automated, teams no longer need application specialists to prepare datasets. In one large financial institution it is maintained with only about 0.25 FTE.
Participants will gain practical insights into how to establish a sustainable test data strategy – one that reflects reality, satisfies compliance, and avoids becoming a costly bottleneck.
This talk is principle-based, tool-agnostic, and grounded in industry patterns. It will draw on personal experiences and anecdotes. It is not a vendor presentation.
