The Protection of Personal Information Act changed what a defensible analytics platform looks like in South Africa. Yet most BI estates we assess still treat compliance as a report-level concern — masking a column here, restricting a dashboard there — while the warehouse underneath holds everything, forever, for everyone.
Four patterns that hold up
Minimise at ingestion. The cheapest personal information to protect is the field you never load. Most analytical questions need behaviour and amounts, not identities; pseudonymise customer keys at the pipeline and keep the lookup table under separate, logged access.
Scope access at the data layer. Row-level security in the warehouse means every tool inherits the same rules — a broker sees their book, a manager their region — instead of re-implementing (and eventually forgetting) restrictions per dashboard.
Keep a lawful-basis register per model. A one-page record of what each dataset contains, why it is processed and under which POPIA condition turns an audit from archaeology into a lookup. It also forces the useful question: do we actually need this data?
Automate retention. Data that should expire must expire without a human remembering. Retention rules belong in the pipeline code, tested like any other logic.
The payoff
Teams adopt governed platforms faster because access is safe by default — nobody wonders whether opening a dashboard breaches something. And when AI use-cases arrive, the foundation is already consentable, minimised and auditable, which is precisely what makes AI projects shippable rather than stallable.