DPIA under GDPR: when it is required and how to run one

A data protection impact assessment is one of the few GDPR tools that has to be done before processing starts. Done well, it is a short, structured decision record that shapes a project. Done badly, it is a long document written after go-live that nobody reads. This guide explains when a DPIA is mandatory and gives a practical structure you can reuse as a template.

GDPR DPIA Privacy by design Published: 8 min read

What Article 35 requires

Article 35(1) of the GDPR requires the controller to carry out an assessment of the impact of envisaged processing operations on the protection of personal data 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 assessment must happen prior to the processing.

The obligation sits with the controller, not with the processor or the software vendor. A vendor can supply useful input, such as a security description or its own assessment of the product, but the controller must assess its own specific use: its purposes, its data subjects and its context. If a Data Protection Officer has been designated, Article 35(2) requires the controller to seek the DPO's advice.

When a DPIA is mandatory

Article 35(3) names three cases where a DPIA is always required. Beyond these, each supervisory authority publishes a national list of processing operations that require a DPIA, under Article 35(4). Those lists differ between countries and often include examples such as large-scale employee monitoring, processing of location data, or combining datasets from different sources. Check the list of the authority responsible for you.

For everything else, the EDPB-endorsed guidelines of the Article 29 Working Party (WP248 rev.01) offer nine criteria. As a rule of thumb, processing that meets two or more of them is likely to require a DPIA, although a single criterion can be enough in some cases.

  • Systematic and extensive evaluation of personal aspects based on automated processing, including profiling, with legal or similarly significant effects
  • Large-scale processing of special categories of data or criminal convictions data
  • Systematic monitoring of a publicly accessible area on a large scale

The nine criteria in practice

The criteria are easiest to use as a screening checklist at the start of every project that touches personal data. Record the answers even when the result is that no DPIA is needed, because that screening record shows accountability under Article 5(2).

Example: a company introduces a tool that scores job applicants automatically. It involves evaluation or scoring, automated decision-making with significant effect, and vulnerable data subjects in the sense of an imbalance of power. Three criteria are met, so a DPIA is clearly needed. If the same tool also uses video interviews with emotion analysis, innovative technology is added, and the EU AI Act may also apply.

  • Evaluation or scoring, including profiling and predicting
  • Automated decision-making with legal or similarly significant effect
  • Systematic monitoring
  • Sensitive data or data of a highly personal nature
  • Data processed on a large scale
  • Matching or combining datasets
  • Data concerning vulnerable data subjects, such as employees, children or patients
  • Innovative use or application of new technological or organisational solutions
  • Processing that prevents individuals from exercising a right or using a service or contract

Check your readiness for free

10–15 minutes, no account needed: readiness score, gap report and AI recommendations.

A practical DPIA structure you can reuse

Article 35(7) sets the minimum content: a systematic description of the processing and its purposes, an assessment of necessity and proportionality, an assessment of risks to individuals, and the measures envisaged to address those risks. The structure below follows that logic and works well for SMEs. Each section should be short and specific to the project.

Several authorities publish their own templates and tools, such as the open-source PIA software from the French CNIL. Using a recognised structure helps if an authority ever reviews your assessment, but the quality of the content matters more than the template.

  • 1. Context: project owner, date, version, whether the DPO was consulted
  • 2. Description: data categories, data subjects, sources, recipients, processors, international transfers, retention, systems and data flows
  • 3. Purposes and legal basis for each purpose, including any Article 9 condition
  • 4. Necessity and proportionality: data minimisation, accuracy, storage limitation, transparency, data subject rights
  • 5. Risk assessment: risks to individuals, likelihood and severity before measures
  • 6. Measures: technical and organisational controls linked to each risk
  • 7. Residual risk and conclusion: acceptable or not, sign-off by the accountable manager
  • 8. Review date and triggers for reassessment

Assessing risk the right way

The most common conceptual mistake is to assess risks to the company instead of risks to people. A DPIA is not an information security risk assessment, even though it uses some of the same inputs. The question is what could happen to the individuals: discrimination, identity theft, financial loss, reputational damage, loss of confidentiality of professional secrets, loss of control over their data, or being unable to exercise their rights.

For each risk, estimate likelihood and severity, describe the measures and estimate the residual level. Link measures to specific risks rather than listing generic controls such as we use encryption. A useful test is whether a reader can see why each measure was chosen and what it changes.

Security measures matter, but so do privacy measures: shorter retention, pseudonymisation, removing unnecessary fields, human review of automated decisions, clear notices and simple ways to object.

When you must consult the supervisory authority

Article 36 requires prior consultation with the supervisory authority when the DPIA indicates that the processing would result in a high risk in the absence of measures taken by the controller to mitigate it. In practice, this means a high residual risk that you cannot reduce.

The authority has up to eight weeks to give written advice, extendable by six weeks for complex processing. Processing should not start before the consultation is complete. Prior consultations are rare, because most projects can be redesigned to bring residual risk down, which is exactly what the DPIA is for.

Common mistakes and what reviewers look for

When an authority or an auditor reviews a DPIA, it looks at timing, ownership and substance. A DPIA dated after go-live, a document copied from a vendor without adaptation, or a risk table with every risk rated low without explanation all suggest that the assessment did not influence the project.

Article 35(11) requires a review where necessary, at least when the risk represented by the processing changes. New data sources, new purposes, a new processor or a move to a different country should trigger a review. A free self-assessment on baltum.ai shows whether your DPIA process, records of processing and other GDPR basics are in place.

  • DPIA started after the contract was signed and the system configured
  • No screening record for projects where a DPIA was judged unnecessary
  • Risks described only from the company perspective
  • No DPO advice recorded, although a DPO is designated
  • No owner sign-off for residual risk
  • No review date and no link to change management

Frequently asked questions

Is a DPIA the same as a PIA?

The terms are often used interchangeably. Privacy impact assessment is the older and more general term, while DPIA is the specific assessment required by Article 35 of the GDPR. What matters is that the assessment covers the minimum content listed in Article 35(7).

Do we need a DPIA for processing that started before the GDPR?

The guidelines expect a DPIA for existing processing when risks change, for example because of new technology, new purposes or new data. It is good practice to review older high-risk processing and document an assessment, even if nothing has changed.

Can one DPIA cover several similar processing operations?

Yes. Article 35(1) allows a single assessment for a set of similar processing operations that present similar high risks. A retail chain using the same CCTV setup in all stores, for example, can use one DPIA and note local differences.

Do we have to publish our DPIA?

The GDPR does not require publication. Publishing a summary can support transparency and trust, especially for public bodies. The full assessment must be available to the supervisory authority on request.

How does a DPIA relate to the EU AI Act?

They are separate obligations. A deployer of a high-risk AI system that processes personal data may need both a DPIA under the GDPR and, in certain cases, a fundamental rights impact assessment under Article 27 of Regulation (EU) 2024/1689. The AI Act allows the two to build on each other.

Check your readiness for free

10–15 minutes, no account needed: readiness score, gap report and AI recommendations.

This article is for general information and is not legal advice. baltum.ai is independent software, not a certification body.