Analytics on PII data demand a paradox: a shared dataset, many audiences, and zero tolerance on sensitive data leakage.
In this session, we present how dynamic row & column-level security is implemented on Databricks, driven entirely by the identity of the user at runtime.
Multiple permission groups (C-level, managers) access the same gold tables, while seeing different rows and column values. Table duplication is avoided and all sensitive data stored encrypted.
Data is ingested and transformed via dbt using a service principal with elevated privileges. Governance datasets are produced mapping users to permissions. An organizational hierarchy defining which employee records each user may access is also built.
Databricks policies evaluate the runtime user, permission group, and managed hierarchy to control row visibility and column unmasking.
Dashboards are built once, accessed via Entra ID SSO, and adapt dynamically,– the same SQL, the same tables but different results are presented.