A DPIA Is an Engineering Review, Not a Privacy Paperwork Exercise
A practical DPIA starts with data flows and risk controls. Here is how developers can turn GDPR Article 35 into decisions that shape an analytics system before launch.
A data protection impact assessment is most useful before code and vendor choices make the risky design expensive to change. GDPR Article 35 requires a DPIA before processing that is likely to result in a high risk to people’s rights and freedoms. For developers, that makes the DPIA an engineering review of data flows, purpose, necessity, and safeguards—not a form to complete after deployment.
Not every analytics implementation automatically requires a DPIA. The GDPR identifies examples such as systematic and extensive evaluation of people based on automated processing that produces significant effects, large-scale processing of special-category data, and large-scale systematic monitoring of publicly accessible areas. Supervisory authorities can also publish lists of processing types that do or do not require one.
Start with the processing, not the vendor
Describe the actual processing operation in plain language. Identify whose data is involved, what events and identifiers are collected, where the data travels, who can query it, how long it remains, and what decisions depend on it. A vendor’s “privacy-friendly” label or a short SDK does not answer those questions.
Draw the data flow from browser or server to collection endpoint, queue, storage, reporting, export, and deletion. Include fallback paths, debugging tools, backups, and support access. The diagram should show where personal data enters, where it is transformed, and which systems receive a copy; otherwise the assessment can miss the secondary paths that create risk.
Make high risk a testable decision
The DPIA must assess necessity and proportionality as well as risks to individuals. Write down the product purpose, then test whether every field and retention period contributes to it. If aggregate page counts answer the product question, a persistent cross-site identifier or full URL may be unnecessary. If a custom event is needed, constrain its schema so arbitrary text cannot become an unreviewed data channel.
Assess harms in concrete terms rather than listing abstract categories. Could a path reveal a medical condition? Could a small team infer a person’s behavior from a rare event? Could support staff access data unrelated to their task? Could a breach expose a long-lived profile? For each risk, name the affected people, a plausible failure, its severity and likelihood, and the control that reduces it.
Turn safeguards into implementation work
A useful mitigation is verifiable. Examples include dropping query strings before persistence, using short retention with automated deletion, restricting access by role, separating production from test data, and logging administrative access without copying event payloads into the audit log. Add the control to the system design and assign an owner; “the vendor is secure” is not a mitigation plan.
Record residual risk after controls, including why the remaining collection is necessary. A DPIA should describe processing purposes, operations, and legitimate interests where applicable; assess necessity and proportionality; evaluate risks; and specify measures to address them. Treat each item as a requirement that can be checked in code, infrastructure configuration, or an operating procedure.
Reassess when the system changes
A DPIA is not a one-time approval. Revisit it when an event schema expands, a new audience or use is introduced, data is combined with another source, a new processor is added, or the retention period changes. These changes can alter both the likelihood and impact of harm even when the SDK code barely changes.
Keep the assessment close to the system: link its data-flow diagram, field inventory, threat model, deletion tests, and vendor configuration. A developer should be able to trace a DPIA statement—such as “no raw IP address is stored”—to the collection path and verify it with a test or query. That turns accountability into evidence rather than a promise.
If a DPIA indicates high residual risk that cannot be mitigated, GDPR Article 36 requires prior consultation with the supervisory authority before processing begins. That is a decision point, not a reason to lower the risk rating. Building the assessment into design review gives the team time to remove unnecessary collection or change the workflow before that point.
Comments
Loading comments…