A data protection impact assessment (DPIA) is a structured look at a new project's privacy risks before it goes live. Done well, it changes the design: less data, shorter retention, better access controls. Done badly, it is a form filled in the week before launch to tick a box.
The GDPR requires DPIAs for high-risk processing. India's DPDP Rules require Significant Data Fiduciaries to do one every year. Vietnam requires impact assessment dossiers to be filed with the regulator. Even where the law doesn't require one, the method is the cheapest way to avoid expensive privacy mistakes.
When to do one
Run a DPIA, or at least a short screening, whenever a project:
- uses sensitive data (health, financial, biometric, children's, location);
- monitors people systematically, for example employee monitoring, CCTV analytics or behavioural tracking;
- uses data at large scale or combines datasets in new ways;
- makes automated decisions that significantly affect people, such as credit, hiring or pricing;
- uses new technology, such as AI models, facial recognition or IoT;
- transfers data to a new country or a new vendor.
Timing matters more than format. Start during design, when the cost of change is low.
The seven steps
1. Screen
Answer five or six yes/no questions based on the list above. If any answer is yes, do the full assessment. Record the screening either way; it shows you thought about it.
2. Describe the processing
Draw the data flow: what data enters, from whom, which systems touch it, who can access it, where it goes, and when it is deleted. Include vendors. One diagram does more than five pages of prose.
3. Check necessity and proportionality
For each data item, ask: do we need this for the stated purpose? Then:
- Could we use less (date of birth versus age band, exact location versus city)?
- Could we keep it for less time?
- Could we use pseudonymised or aggregated data instead?
- What is the legal basis, and is it the right one?
This step usually removes 20–30% of the planned data. That alone justifies the exercise.
4. Identify the risks to people
Think about harm to individuals, not just to the company: discrimination, financial loss, identity theft, loss of confidentiality, embarrassment, loss of control over their data, and chilling effects. Consider both security failures (breach, misuse by insiders) and intended processing that goes too far (profiling people didn't expect).
5. Score the risks
Keep it simple. Rate likelihood (remote, possible, likely) and severity (minimal, significant, severe) and combine them into low, medium or high. Agree the scale before scoring so different assessors produce comparable results.
| Minimal | Significant | Severe | |
|---|---|---|---|
| Likely | Medium | High | High |
| Possible | Low | Medium | High |
| Remote | Low | Low | Medium |
6. Decide on measures
For each medium or high risk, choose measures that reduce it: minimisation, pseudonymisation, encryption, access restrictions, shorter retention, opt-outs, human review of automated decisions, transparency notices, vendor terms. Record the residual risk after the measures.
If a high risk remains, escalate it. Under the GDPR you may need to consult the regulator before going ahead.
7. Sign off and revisit
A named owner accepts the residual risk. Put a review date in the diary, or a trigger such as a new data source, new country, new vendor or new purpose. A DPIA is a living document.
Tips that make it work
- Pair a privacy person with an engineer. The engineer knows what the system really does.
- Use the DPIA to drive tickets. Each measure becomes a backlog item with an owner and a due date.
- Keep a library. Reuse earlier assessments for similar projects so teams aren't starting from scratch.
- Measure. Count how many DPIAs changed a design. That number tells you whether the process is real.
Learn more
DPIAs and privacy by design are covered in our free Data Protection & Privacy Fundamentals course. For AI projects, combine this with our five-step AI risk assessment.
Written and reviewed by the DigiTrust Asia editorial team. How we research, review and correct articles.
General information, not legal advice. Laws and official guidance change; check the current text before acting.
Learn privacy properly
Data Protection & Privacy Fundamentals is a free, law-neutral course with a final exam and certificate.
Create a free account