Skip to main content

How Heroify uses artificial intelligence

Effective from 22 September 2026. The full AI Transparency Note for Clients is available on request: contact@heroify.co.

1. Provider and status of the system

Provider: Heroify sp. z o.o., ul. Padewska 23/7, 00-777 Warsaw, Poland. Contact: contact@heroify.co (product), gdpr@heroify.co (personal data).

The Heroify AI system is used to evaluate candidates in recruitment and employees in internal assessments, and it is a high-risk AI system within the meaning of Annex III point 4 of the AI Act. Heroify applies the provisions of the AI Act to the extent they apply to the system at a given time.

2. Intended purpose

The system supports the User in evaluating the competencies, cognitive abilities, values fit and professional attitudes of people who complete Assessments. It provides structured information for an interview and for a decision made by a human.

The system must not be used to:

  • automatically reject or accept candidates;
  • evaluate minors;
  • infer legally protected characteristics;
  • evaluate employees for purposes other than recruitment, development and HR purposes permitted under employment law;
  • purposes other than professional evaluation.

3. How the system works

The system consists of several components. We describe each of them separately.

  • Scoring of answers. Closed-ended answers are scored algorithmically against an answer key. The system does not score open-ended answers automatically.
  • Tailoring and selection of questions. A language model tailors the wording of selected questions to the job or role context provided by the User; the structure of the question, the competency measured and the scoring key stay the same, so results remain comparable within the same test. For some tests, a language model selects questions from a question pool designed and validated by the psychometric team, following rules set by that team (including the competency measured and the difficulty level), so that question sets in different assessments are equivalent and, at the same time, do not repeat; norms are calculated for the test as a whole. The question pool and the selection rules are created and verified by humans; the model does not create new evaluation criteria.
  • Percentile norms. A result is compared with the population of people who have completed the same test. Norms for a test are made available once the comparison population is large enough.
  • AI summary. At the User's request, a language model generates a summary based on the job context and the numerical results, without identifying data and without the content of the candidate's answers. In the interface it is labeled as generated by AI.
  • Session trust indicator (high/medium/low), calculated from technical data about the session.
  • Ranking by overall score.

Beyond candidate evaluation. A language model also supports the User in setting up an Assessment (suggestions for the job level, competencies and open-ended questions, and mapping the organization's values to scales) based on the content entered by the User; these features do not process candidate data and are not part of the evaluation.

4. Human oversight

No component of the system changes a candidate's status, rejects a candidate or hides a result; statuses are set by the User. The AI summary consists of hypotheses to check in an interview, not findings. The trust indicator refers to the session, not to the candidate's competencies; it is a signal to verify, not proof of cheating or grounds for rejection. Unusual patterns may have causes unrelated to cheating (a break, technical problems, assistive technologies, an individual working style).

5. Limitations

The AI summary is statistically generated text: it may generalize, miss a nuance or state a conclusion too confidently, and it may echo wording from the job description provided by the User. Norms depend on the population that completed the test; for new tests they may not be available. The User is responsible for the content of the Assessment setup (including the job description). The system does not recognize emotions, does not process biometric data and does not use candidate data to train models.

6. Obligations of the Client as deployer

A Client using the system:

  • informs candidates and employees that they are being evaluated with the use of a high-risk AI system (Heroify provides the wording: Section 5 of the Privacy Policy for Candidates/Participants) and, where applicable, also informs employee representatives;
  • ensures human oversight and the competence of the people who use the results;
  • uses the system in line with its intended purpose;
  • at a candidate's request, explains the role of the system in the decision (Heroify provides a description of the factors taken into account in the result);
  • reports any suspected malfunction to Heroify;
  • where required to do so (Article 27 of the AI Act), carries out a fundamental rights impact assessment using the information made available by Heroify.

Detailed rules: the full AI Transparency Note and Section 11 of the DPA.

7. Data, changes and contact

The scope of data, the roles and the retention periods are described in the DPA and the Privacy Policy for Candidates/Participants. Questions about how the system works: contact@heroify.co. Questions about personal data: gdpr@heroify.co.