Skip to content
DT Nova

Data protection impact assessment (DPIA): when you need one and how to do it

A data protection impact assessment, or DPIA, is the written risk assessment the GDPR requires before you start processing personal data in a way that is likely to put people at high risk. It describes the processing, tests whether it is necessary and proportionate, assesses the risks to the people concerned and records the measures that bring those risks down. This guide explains when a DPIA is required, what it must contain, how to run one in practice and when you must ask the supervisory authority first.

The legal text quoted here is Regulation (EU) 2016/679, the GDPR, read on 12 November 2026. The criteria come from the European guidelines on DPIAs (WP248 rev.01), which the national data protection authorities apply. This is a practical guide, not legal advice: for a decision on a specific project, involve your data protection officer or lawyer.

What a DPIA is, in one paragraph

Article 35(1) of the GDPR says that where a type of processing, in particular using new technologies, is likely to result in a high risk to the rights and freedoms of natural persons, the controller must assess the impact of the envisaged processing on the protection of personal data, and must do so before the processing starts. One assessment may cover a set of similar processing operations with similar high risks. The DPIA is therefore a planning document: it belongs in the design phase of a system, not in the archive after launch.

When a DPIA is required

Article 35(3) names three cases where a DPIA is always required:

  • a systematic and extensive evaluation of personal aspects based on automated processing, including profiling, on which decisions are based that produce legal effects or similarly significantly affect the person;
  • processing on a large scale of special categories of data (such as health, biometric or ethnic data) or of data on criminal convictions and offences;
  • systematic monitoring of a publicly accessible area on a large scale.

Beyond those three, each national supervisory authority must publish a list of processing types that require a DPIA (Article 35(4)) and may publish a list of types that do not (Article 35(5)). Check the list of the authority where you are established, and of the countries where you offer services, before you decide.

The nine criteria from the European guidelines

For everything not on a list, the Article 29 Working Party guidelines WP248 rev.01 (adopted 4 April 2017, last revised 4 October 2017, read on 12 November 2026) set out nine criteria. As a rule of thumb, the guidelines say that processing which meets two criteria will in most cases require a DPIA, and that in some cases one is enough.

CriterionTypical example in a business system
1. Evaluation or scoring, including profiling and predictingCredit or fraud scoring, lead scoring of private persons, performance prediction
2. Automated decisions with legal or similarly significant effectAutomatic rejection of an application or a contract
3. Systematic monitoringEmployee activity tracking, vehicle location tracking, camera systems
4. Sensitive data or data of a highly personal natureHealth records, care notes, financial details, location history
5. Data processed on a large scaleA national customer base, all residents of a region
6. Matching or combining datasetsJoining CRM, web analytics and purchase data from different sources
7. Data about vulnerable peopleEmployees, patients, children, elderly clients, asylum seekers
8. Innovative use or new technological solutionsFace recognition, connected devices, AI models trained on personal data
9. Processing that prevents people from using a right, service or contractScreening customers against a database before offering a loan

Two examples show how quickly a common project reaches the threshold. A fleet management system that tracks company vehicles driven by employees meets criteria 3 and 7. A home care app with care notes about elderly clients meets criteria 4 and 7. Both normally need a DPIA.

What a DPIA must contain

Article 35(7) sets the minimum content. The assessment must contain at least:

  1. A systematic description of the envisaged processing operations and their purposes, including, where applicable, the legitimate interest pursued by the controller.
  2. An assessment of necessity and proportionality of the processing in relation to the purposes.
  3. An assessment of the risks to the rights and freedoms of the people concerned.
  4. The measures envisaged to address the risks, including safeguards, security measures and mechanisms to protect personal data and demonstrate compliance.

The controller must seek the advice of its data protection officer, where one is designated (Article 35(2)), and, where appropriate, the views of the people concerned or their representatives (Article 35(9)). The annex of the WP248 guidelines lists the criteria an acceptable DPIA should meet, and is a useful checklist when you review your own document.

How to run a DPIA in practice

Most DPIAs that fail do so because they are written after the system is built, by someone who was not in the design meetings. The order below keeps the assessment useful:

  1. Screen early. At the idea stage, check the national list and the nine criteria. Record the result, including a decision that no DPIA is needed and why.
  2. Describe the data flows. Which data, about whom, from which sources, stored where, shared with which processors and countries, kept for how long. A diagram of the flows is worth more than pages of text. Our process mapping guide shows how to draw them.
  3. Test necessity. For each data field and each purpose, ask whether the purpose can be reached with less data, shorter retention or pseudonymised data. This step usually removes risk cheaply.
  4. Assess the risks from the person's point of view. Not the risk to the company, but to the people in the data: unauthorised access, loss, wrong decisions, discrimination, loss of control, chilling effects. Rate likelihood and severity.
  5. Decide the measures. Access control by role, encryption, logging, retention rules, human review of automated decisions, clear information to the people concerned, and contracts with processors.
  6. Record residual risk and sign off. The controller decides whether the remaining risk is acceptable, with the advice of the data protection officer recorded.
  7. Review on change. Article 35(11) requires a review at least when the risk represented by the processing changes.

Prior consultation with the supervisory authority

If the DPIA shows that the processing would result in a high risk in the absence of measures taken by the controller to mitigate it, the controller must consult the supervisory authority before processing (Article 36(1)). The authority then has up to eight weeks to give written advice, extendable by six weeks for complex processing (Article 36(2)). Article 36(3) lists what to send, including the DPIA itself, the responsibilities of controllers and processors, the purposes and means, and the safeguards. In practice, most projects avoid prior consultation by finding measures that bring the residual risk down to an acceptable level, which is the purpose of the exercise.

What happens if you skip it

Failing to carry out a required DPIA, or to consult the authority when required, is an infringement of the controller's obligations under Articles 25 to 39. Article 83(4) allows administrative fines for those infringements of up to 10 million euro or, for an undertaking, up to 2 % of total worldwide annual turnover of the preceding financial year, whichever is higher. The more common cost is practical: a system that has to be redesigned after launch because a basic privacy measure was missing.

DPIAs, processors and AI systems

A DPIA is the controller's duty, but processors must help: Article 28(3)(f) requires the processor to assist the controller with Articles 32 to 36. That assistance belongs in the data processing agreement. For AI systems, the DPIA sits next to the AI Act. Article 27 of the AI Act, as amended by Regulation (EU) 2026/1744, lets deployers that must carry out a fundamental rights impact assessment refer to the relevant sections of their DPIA rather than repeat them (read on 12 November 2026). Our AI governance guide explains where the two assessments meet.

A DPIA template outline

  • Project name, owner, version, date, next review date.
  • Screening result: national list, Article 35(3), the nine criteria.
  • Description: purposes, legal basis, data categories, people concerned, sources, recipients, processors, transfers, retention.
  • Data flow diagram.
  • Necessity and proportionality: data minimisation, retention, information to people, rights handling.
  • Risk register: risk, likelihood, severity, measures, residual risk.
  • Advice of the data protection officer, and views of the people concerned where sought.
  • Decision and sign-off by the controller; prior consultation yes or no.

How we help

We build software with the DPIA in the design phase. When we build a system for you, we deliver the material your DPIA needs: data flow diagrams, the list of data fields and their purposes, the processors and hosting regions (EU by default), the access model, logging and retention. Measures from the DPIA become requirements in the backlog, not notes in a binder. For an existing system, a security audit checks whether the measures described in your DPIA are actually in place. You own the code and the documentation. Planning an AI use that processes personal data? Our AI consulting services include the written data rule.

Questions people ask

Is a DPIA mandatory?

Only for processing that is likely to result in a high risk to people's rights and freedoms. Article 35(3) of the GDPR names three cases where it is always required, and each national supervisory authority publishes a list of further processing types that need one. For processing outside those lists, the controller decides, and the European guidelines suggest a DPIA is usually needed when two or more of their nine criteria apply.

Who has to carry out the DPIA?

The controller, meaning the organisation that decides why and how the data is processed. The controller must seek the advice of its data protection officer, where one is designated, and its processors must help with the information they hold. A software supplier or consultant can write much of the document, but the controller owns the conclusion.

What is the difference between a DPIA and a privacy impact assessment?

Privacy impact assessment, or PIA, is the older and wider term used in many countries and standards. A DPIA is the specific assessment required by Article 35 of the GDPR, with a minimum content set by law. A PIA that covers the four elements of Article 35(7) can serve as the DPIA.

Do we have to publish our DPIA?

The GDPR does not require it. The European guidelines say publishing a summary can help build trust, especially for processing that affects the public. The full DPIA must be available to the supervisory authority on request, and it must be sent to the authority in a prior consultation.

How often should a DPIA be reviewed?

Whenever the risk changes: a new purpose, new data, a new supplier, a new country or a new technology. Article 35(11) requires a review at least when there is a change in the risk, and the guidelines recommend continuous review as good practice. Put a review date in the document and link it to the release process of the system it describes.

Does the AI Act replace the DPIA for AI systems?

No. A DPIA is still needed when an AI system processes personal data in a high-risk way. Some deployers of high-risk AI systems must also carry out a fundamental rights impact assessment under Article 27 of the AI Act, and that article allows them to refer to the relevant sections of the DPIA instead of repeating them.

Sources