Legal
Data Protection Impact Assessment
Version 0.1 (draft) · Last updated 2026-11-20
Draft — not yet in force
This text has not yet been reviewed by a lawyer and may change. Details marked [to be completed] will be filled in, and a Danish version will follow. Questions: privacy@haiya.dk.
This is a summary of Haiya's data protection impact assessment (DPIA) for the Haiya
hiring platform at app.haiya.dk, made under Article 35 of the GDPR. It describes what
the platform does with personal data, the risks this carries for the people involved,
and what reduces them. Our Privacy Policy and Terms of Use have
the full detail of what we collect and why.
1. About this assessment
Controller: HAIYA ApS, CVR 46424832, Denmark. Where a company uses Haiya for its own hiring, that company is the controller and Haiya its processor; the company can rely on this assessment for its own (see Privacy Policy, section 1).
What it covers: candidate profiles and applications, the work-style and personality screening, team-member culture profiles, AI-assisted matching, interview transcripts and messaging.
Status: a working draft. It will be reviewed by counsel, and the conclusion in section 7 is completed once the actions it lists are done.
Review: at least once a year, and before any change that adds a new kind of data, a new AI use or a new provider (for example Haiya-hosted video interviews).
2. Why a DPIA is needed
A DPIA is required when processing is likely to result in a high risk to people's rights and freedoms. Several of the European Data Protection Board's criteria apply to Haiya:
- Evaluation and scoring. The platform weighs applicants against a position and a team, and suggests a match.
- Personal and sensitive assessments. Candidates and team members answer personality and work-style statements; team members also take a short reasoning test.
- People in a weaker position. Job applicants and employees cannot freely refuse what an employer asks of them.
- New technology. AI models read CVs and transcripts and write assessments.
3. What the platform does
| Who | What is processed | Why |
|---|---|---|
| Candidates | Account, profile and CV; applications, answers and motivation; work-style, personality and motivation statements; what the hiring team records (interview bookings, notes, transcripts, match score and its reasons) | To apply for jobs and let a company assess the application |
| Team members | Name, email, culture-profile answers, scores, reasoning-test result and an AI-written profile | So a company can describe how its team works and judge fit against the real team |
| Company users | Account, role and activity | To run the company's hiring |
| Everyone | Messages, consent history, security logs | To communicate, to show what was agreed, and to keep the platform secure |
AI use. AI models read CVs into structured profiles, weigh an application against a position and the team, summarise interview transcripts and write team-member profiles. The output is a suggestion with reasons. A person makes every hiring decision. Candidates never see their match score.
Recipients and providers. Only the company a candidate applies to sees the application. Providers process data on our instructions: hosting (Railway, EU; Vercel), file storage and bot check (Cloudflare), email (Resend), AI (Anthropic, USA) and search embeddings (OpenAI, USA). The transfer basis for the US providers is [to be completed].
4. Necessity and proportionality
- Legal basis. Contract for accounts; the company's legitimate interest, or steps before a contract, for assessing applications; the company's legitimate interest for team profiles, with two recorded consents that make taking part voluntary and informed; consent for optional emails.
- Only what is needed. No special categories of data are asked for. The AI is told not to consider names, gender, age, ethnicity, health and similar characteristics, and signals that mention them are filtered out.
- Transparency. The Privacy Policy and Terms of Use explain the processing, and every consent is recorded with the version of the terms in force.
- Limited storage. Interview transcripts are deleted automatically 6 months after the hiring process ends, and after 24 months at the latest. Retention periods for the other kinds of data are set out in the Terms of Use, section 10.
- Your rights. You can see, correct and delete your data, object, and ask for a person to review a decision (see Privacy Policy, section 10).
5. The main risks
| Risk to the people involved | How serious |
|---|---|
| A candidate is judged unfairly or with bias, by the AI or by the personality statements | High |
| The suggested match decides the outcome in practice, with no real human judgement | High |
| Answers to personality statements reveal, or are read as revealing, something about health or other sensitive matters | High |
| A team member feels they cannot say no to the culture profile | High |
| Data is kept longer than it is needed | Medium |
| People cannot easily use their rights, such as getting a copy of their data | Medium |
| Unauthorised access to profiles, applications or transcripts | High |
| Data sent to AI providers outside the EU | Medium |
6. What reduces the risks
In place now:
- AI output is advisory. Nothing in the platform rejects or advances a candidate on a score, and the hiring team sees the reasons next to the evidence they come from.
- The AI is instructed to ignore protected characteristics, and statements that mention them are removed from what the hiring team sees.
- Optional blind screening ("Anonymous apply"): when a company creates a job post, it can choose to hide candidates' names, photos, locations, contact details and CVs from its hiring team, so applicants are first assessed on skills and fit. A reviewer can reveal them at a later stage.
- The weight of culture fit in the match can be switched off.
- Candidates never see a score about themselves.
- Team members give two separate, recorded consents before starting a profile, and their company sees a written profile, not their individual answers.
- Interview transcripts are encrypted with a separate key and deleted on a fixed schedule.
- Each company sees only its own data. Sign-in uses short-lived tokens and a secure cookie, sign-in attempts are rate-limited, and logs are cleaned of personal content.
- AI prompts are logged only as a fingerprint, never as the text.
- Candidates can delete their own account and data.
Your rights in practice: candidates can delete their own account and data at any time. Anyone can ask us at privacy@haiya.dk for a copy of their data, a correction, deletion, or to withdraw a consent; team members, who have no account, use this route or ask their company.
Being completed: [to be completed]
7. Conclusion
[to be completed]
Questions about this assessment: privacy@haiya.dk.