Legal
      Back to home
      1. Home
      2. Legal

      Questions relating to the questionnaires, AI and algorithms

      Do you use Artificial Intelligence to provide your services?

      AssessFirst integrates artificial intelligence (AI) into several of its functionalities:

      • Creation of predictive models: the functions for creating predictive models based on job offers and contact analysis use AI models to identify, objectively, the dimensions (personality, motivations, reasoning, skills) that enable success in a position. The models offered by default, known as "Benchmark", do not include functionalities based on AI.
      • VOICE: VOICE is an AI agent designed to transform the way companies assess talent. By integrating advanced machine learning models, VOICE autonomously orchestrates the entire skills assessment process.
      • Development plan: this feature allows you to create tailor-made development paths for each employee. Based on the latest research in behavioural sciences, this advanced feature of the Competency Framework combines generative AI, contextualisation of objectives and a variety of learning supports to offer an integrated and personalised approach to skills development.
      • AI Assistant: an intelligent chatbot that enables the Recruiter to explore profiles dynamically and obtain personalised insights that can be used directly. Thanks to the integration of language models, the chatbot responds instantly to the most specific questions with relevant, contextualised and actionable information.

      Are you affected by the European "IA Act" regulation?

      Yes, the European regulation on artificial intelligence (IA Act) applies to all suppliers of AI systems marketed in the European Union, including AssessFirst. However, the obligations imposed on AssessFirst by this regulation will only apply from 2 August 2026.

      We have therefore anticipated the requirements of the future European AI Regulation (AI Act) and aligned our practices in terms of security, transparency, risk management and model governance. Our systems are designed to ensure the ethical and reliable use of AI, with strict data and model control processes. In addition, we have carried out a full audit with Holistic AI, a recognised independent organisation, which confirmed the absence of gender or origin bias in our systems, reinforcing the compliance and reliability of our solution. The detailed results of the audit can be consulted on this page.

      What is the risk level of your AI systems?

      According to the IA Act, the services offered by AssessFirst are considered to be high-risk AI systems, as they are "intended to be used for the recruitment or selection of natural persons, in particular for (...) analysing and filtering applications and evaluating candidates" (appendix III 4. (a) of the IA Act).

      What is your role with regard to the AI systems used?

      For certain functions such as VOICE, the Development Plan or the AI Assistant, AssessFirst uses GenAI systems and modifies their destination. In accordance with the IA Act, AssessFirst therefore qualifies as a supplier of AI systems (section 25 of the IA Act).

      Have you set up governance for the ethical risks of AI?

      Yes, we have set up specific governance for managing these risks, which is ensured by a multidisciplinary internal committee made up of the Head of Legal, the Head of Science and the Chief Technology Officer. This committee oversees the development, deployment and evolution of AI systems in order to guarantee compliance with ethical principles and local and international regulations, and to assess risks on a regular basis.

      What ethical commitments do you make in relation to AI?

      AssessFirst is committed to the ethical, responsible and transparent use of artificial intelligence. This commitment is based on six main principles:

      • Respect for privacy: we apply strict data minimisation principles, exclude all sensitive data, and anonymise or pseudonymise the data used for AI. A privacy impact assessment is available on request.
      • Fairness and avoidance of bias: our models are trained on a variety of datasets and are regularly analysed for potential bias. Our questionnaires comply with APA and ITC standards.
      • Transparency and accessibility: explanations are provided on how our tools work. Our interfaces are designed to be inclusive (accessibility, multilingualism), and our results include clear and understandable justifications.
      • Security and robustness: our systems are hosted on ISO 27001-certified infrastructures, protected by advanced cybersecurity measures (encryption, audits, MFA), and are subject to continuous monitoring.
      • Continuous improvement: pre-deployment tests and post-deployment audits are carried out regularly. Our users can also contribute to improving the tools via supervised feedback.
      • Human supervision: no decision is automated. Users always retain control, and AI managers are identified and trained in the technical, behavioural and ethical issues involved.
      • External audit: we carried out a full audit with Holistic AI, a recognised independent organisation, which confirmed the absence of gender or origin bias in our systems. The results are available here.

      These commitments are updated regularly to keep pace with technological and regulatory developments. We also have a charter on the responsible use of AI, which is available on request.

      Are your employees trained in AI ethics?

      Our employees, in particular those involved in the development, deployment and supervision of our platform, receive specific training in AI ethics. This training covers the principles of fairness, transparency, non-discrimination, data confidentiality and security. It is regularly updated to take account of regulatory developments and international best practice, particularly those arising from European and international recommendations.

      Are AI results supervised by a human?


      Yes, the results produced are systematically supervised by humans:

      • Client side: recruiters have full and transparent access to the assessments, which ensures that a human user always reads and validates the results before any decision is made.
      • AssessFirst side: we maintain full control over the system by carrying out regular checks on the performance of the AI models, and make continuous improvements to ensure the quality, reliability and compliance of the system.

      Have you implemented an AI risk management system?

      Yes, AssessFirst has implemented a risk management system dedicated to artificial intelligence systems, which is documented, regularly updated and applied to all stages of the model lifecycle.

      This structured framework enables us to identify, assess and mitigate the risks associated with the use of AI, whether in terms of bias, fairness, confidentiality, data security, performance or ethical impact. It applies from the design phase, throughout development, during deployment and during the operational phase.

      We carry out regular reviews to take account of technological developments, user feedback and regulatory requirements. This process includes internal audits, validation checks and, if necessary, impact assessments.

      Are there protocols in place to detect and report incidents quickly?


      Yes, AssessFirst has implemented advanced security protocols to protect its AI systems and infrastructure, and to ensure a rapid response in the event of an incident.

      Our measures include data encryption, rigorous access control, intrusion detection systems and continuous monitoring to prevent and mitigate cyber threats, data breaches or unauthorised access. We carry out regular security audits and penetration tests and apply industry best practice in cyber security.

      We also have a strict data governance policy in place, guaranteeing the confidentiality, integrity and availability of AI systems. Finally, we are currently preparing for ISO 27001 certification, demonstrating our commitment to maintaining a high level of security in line with evolving standards.

      Is documentation available to help understand how AI systems work?

      Yes, full documentation is available to help understand how our systems work and how they operate. This includes the technical manual, the fairness guide, AI module self-assessment reports, impact and return on investment (ROI) studies, scientific publications and Holistic AI's external audit report. This documentation is accessible via our website from the Sciences Knowledge Hub or can be provided on request.

       

      If an Individual refuses to allow the AI system to process their data, do they benefit from the same advantages and opportunities as those who use it?

      Yes, the AI system is designed to assist the Client in decision making, but does not replace human judgement or create exclusive benefits for participants.

      Other non-AI based methods are available to ensure fairness, inclusivity and equal access to opportunities. Also, the basic questionnaires (SWIPE, DRIVE, BRAIN) do not use AI, allowing individuals who wish to benefit from our services to do so without interacting with artificial intelligence systems. No one will be disadvantaged for choosing not to use AI-based functionalities.

      Do AI-based features use personal data?

      For the personality questionnaires, the Development Plan and the AI Assistant, certain AssessFirst functions - which are offered after the person's profile has been established by the calculation algorithm - such as the job recommendation to the Individual, known as "orientation", use information such as "current job title" in order to offer jobs that are consistent with the Individual's profile.

      For VOICE, AssessFirst collects voice recordings of the answers given by Individuals.

      Is personal data used to train your AI models?

      No. When training our models, we only use anonymised data, in accordance with CNIL recommendations, and separated from production environments. This data enables us to :

      • Carry out statistical studies, for example to refine skills profiles and assessment models,
      • scientifically validate our services and guarantee the robustness and relevance of assessments,
      • more generally, to continuously improve our services, for example to develop and validate new functionalities or methodologies.

      Are your AI models protected against unauthorised access?

      Yes, our models are protected by several levels of security. Access to the training and deployment environments is strictly controlled via role-based access management rules, limiting access to authorised people only. All accesses and modifications are recorded in audit logs to ensure full traceability. Finally, we apply rigorous network segmentation to isolate critical environments and prevent unauthorised access.

      How do you eliminate cognitive bias?

      AssessFirst's rigorous methodology and expertise enable us to neutralise potential bias at every stage of the process:

      • At the level of the psychometric questionnaires: the psychometric qualities of AssessFirst questionnaires demonstrate the absence of bias in the questionnaires. Our Science team makes every effort to ensure the scientific robustness of the questionnaires, and is an affiliated member of the ITC (International Test Commission), the benchmark for standards in the development of assessment tools.
      • Calculating scores: our algorithms only take into account 'neutral' variables, i.e. questionnaire scores. Under no circumstances do these algorithms take into account other variables that would be discriminatory, such as gender, age or origin.

      AssessFirst enables companies to limit cognitive bias by collecting high-quality, highly standardised data (through our questionnaires) and applying systematic processing (through our algorithms).

      Where do the SWIPE, DRIVE and BRAIN personality questionnaires come from?


      1. Personality questionnaire : SWIPE

      Mesure: the SWIPE personality questionnaire helps us to understand the behaviours that individuals naturally tend to display. 

      Theoretical model:

      • Big Five: this is the most widely used personality model.
      •  We also add a Humility trait, as proposed by new research (Denissen et al., 2022; Lee et al., 2022). 

      Dimensions: 20 personality facets measured: in order to provide a detailed description of how an individual functions, each of the 5 traits has been broken down into 4 facets. Examples of facets measured: Takes the lead over others, Seeks to convince others, Approaches others spontaneously, Demonstrates diplomacy, Is emotionally involved, Is open to others' ideas, Accepts criticism, Consults before making decisions, Is attracted by varied tasks, Is interested in abstract things, Is inventive, Adapts to change, Organises systematically, Attaches importance to detail, Perseveres in the face of obstacles, Goes beyond prescribed tasks, Is relaxed, Attaches importance to positive aspects, Controls emotions, Seeks stability. 

      Scoring:  IRT scoring, which is the best method of calculating scores in psychometrics.

      Format: 

      • 75 forced-choice questions. This question format makes it possible to: (1) neutralise the social desirability bias, (2) make the questionnaire easier to complete.
      • Accessible on all platforms (desktop, mobile, tablet) and available in 23 languages.
      • Audio help available. 

      Quality of experience:

      • Fast: the duration of SWIPE has been designed to maximise the balance between meeting user expectations, the perceived seriousness of the test and the validity of the assessment.
      • Mobile-first: SWIPE has been designed to be mobile-first in order to respond to the ever-increasing desire of users to complete all assessments via mobile. 70% of our users complete SWIPE on their mobile.
      • Engaging: image gamification maximises the quality of information while optimising user engagement and satisfaction (average rating of 4.6/5).

      Duration: Approximately 5 minutes. 

      Publications: SWIPE has been the subject of 15 scientific publications.

       

      2. Motivation questionnaire : DRIVE

      Mesure: the DRIVE motivational questionnaire helps us to understand what determines people's satisfaction and commitment. 

      Theoretical model:  self-determination and person-environment fit theories. 

      Dimensions: 20 motivational dimensions are measured. Examples of dimensions measured: Creating new things, Surpassing oneself on a daily basis, Being concerned about aesthetics, Analysing data, Meeting new people, Having clearly defined tasks, Being concerned about quality, Having influence, Having autonomy, Working in a team, Having a positive impact on the world, Working in a fun atmosphere, Evolving in a secure environment, Preserving one's personal balance, Working in a disciplined way, Receiving rewards, Having an attractive remuneration, Regularly achieving success, Helping others, Being recognised by others. 

      Scoring:  IRT scoring.

      Format: 

      • 90 forced-choice questions (two answer choices per question).
      • Accessible on all platforms (desktop, mobile, tablet) and available in 23 languages.
      • Audio help available. 

      Duration: between 8 et 12 minutes.

       

      3. Cognitive reasoning test: BRAIN

      Mesure: the BRAIN questionnaire measures an individual's general reasoning ability.

      Theoretical model:  hierarchical models of intelligence and general intelligence factor theories. 

      Dimensions: overall score for reasoning potential, decision speed, type of task preferred, learning style, ability to manage complexity, precision of decisions, risk factors (haste, excessive caution, inaccurate deductions, indecision). 

      Scoring:  IRT scoring.

      Format: 

      • Adaptive format: BRAIN's adaptive models enable a tailor-made test for each user, as the items proposed are adapted to the user's actual performance on previous items. This makes for: (1) a faster test, (2) an improved user experience, as the individual does not feel they are failing thanks to items adapted to their level, and (3) less leakage of answers and cheating.
      • Gamified format: BRAIN, without being a game, uses the same interaction levers: real-time feedback, construction of answers, adapted level of items. This format helps to develop candidate commitment while ensuring validity.
      • Fairness: The neutral material used in BRAIN (i.e. material free of verbal elements and based on colours unrelated to the ability to solve the items) makes it accessible to people with disabilities.
      • Accessible on all platforms (desktop, mobile, tablet) and available in 23 languages. 

      Duration: Between 8 and 12 minutes. 

      How do your personality questionnaires work?

      The reports we produce following the Swipe, Drive and Brain questionnaires contain two types of information: quantitative (in the form of scores) and qualitative (in the form of comments).

      The aim of the quantitative information is to situate the Individual Candidate in relation to a reference population used to calibrate our questionnaires. This reference population was determined by AssessFirst. This population is made up of several hundred thousand anonymous profiles, of all ages, genders, levels of education, levels of responsibility and functions.

      Initially, a user's responses to the questionnaires enable us to calculate an initial "raw score" on the various traits and dimensions measured by our questionnaires. These raw scores are calculated by taking into account all the responses linked to a personality or motivation criterion. Each criterion is evaluated several times using our questionnaires, to ensure the reliability of the measurement. For the 3 questionnaires, our scoring methodologies are based on an IRT (Item Response Theory) approach.

      However, these raw scores cannot be interpreted as they stand. They therefore need to be 'translated' using another scale, which will give them meaning. Most of our raw scores are calibrated on a scale of 1 to 10, and it is this score that is presented directly on the application, and which enables us to understand a person's positioning.

      In concrete terms, the person's raw score is calibrated using a Gaussian calibration. Each score from 1 to 10 corresponds to a position relative to a reference population: a score of 1 corresponds to the 2% of the population with the lowest raw scores, while a score of 10 corresponds to the 2% of the population with the highest raw scores.

      It should also be noted that there are no "good" or "bad" scores per se. It is the responsibility of the human resources professional who interprets it to relate it to a professional context.

      For example, obtaining a score of 2 for the personality trait "Demonstrates diplomacy" reflects a direct, straightforward mode of communication. In certain professional contexts, this mode of communication is valued, in others it can be limiting, and finally it can also have no positive or negative impact on the situation.

      Alongside this quantitative information and scores, AssessFirst also provides qualitative information in text format. The content generated is based on the scores obtained by the individual. In this way, they provide a textual interpretation of the score, offer explanations and make it easier to understand the profile.

      What information do you record when an individual takes the personality questionnaires?

      By answering personality questionnaires, AssessFirst records :

      • the answer to each question,
      • the time taken to respond,
      • the number and duration of breaks taken,
      • whether the questionnaire was carried out in good conditions,
      • whether the questionnaire was completed in the respondent's mother tongue.

      For the Brain questionnaire, other elements are also recorded:

      • stages in the construction of the final response,
      • validation that the Brain instruction has been correctly understood,
      • the number of resets.

      AssessFirst's data processing algorithm only takes into account responses to questionnaires and the time taken to answer questions.

      Does the calculation algorithm used in your personality questionnaires use personal data?

      No. AssessFirst does not use personal data in its calculation algorithm for its questionnaires.

      Our Swipe, Drive and Brain questionnaires are based on data that has been anonymised as defined by the CNIL, i.e. data that is no longer considered to be personal data.

      Only the answers to the questions are kept, without any link to the surname, first name, gender, IP address or any other personal data collected by AssessFirst relating to the person concerned.

      How do you create your personalised job recommendations?

      The personalised job recommendations feature allows you to discover the jobs that best match the characteristics of the individual's AssessFirst profile.

      This list is not intended to lock the Individud into a particular direction, but rather to allow each individual to broaden his or her thinking.

      The recommendations are based on a number of factors:

      • Matching the Individual to the business predictive model ;
      • The individual's career objective: to explore new occupations close to their own, to move into a position with a higher hierarchical status, or to change career path.

      How do your predictive models work?

      AssessFirst creates and makes available "predictive models": these models define the personality traits, motivations and levels of reasoning that make it possible to anticipate success and fulfilment in a target position. We have three types of model.


      1. Benchmark model

      This is the default model. The "Benchmark" model includes jobs that have already been configured by AssessFirst to serve as a reference (around 400 jobs at the moment, with over 3,000 targeted). Each model incorporates the personality traits, motivational dimensions and levels of reasoning expected in the job.


      2. Customised model

      For this type of model (optional), the Client completes a questionnaire proposed by AssessFirst to qualify the objective and the professional context associated with the predictive model. The Client can also directly select the dimensions they wish to include in their model. AssessFirst then generates a predictive model based on the answers provided by the Client.


      3. Artificial Intelligence model

      In this optional model, AssessFirst takes into account indicators relating to Individuals identified by the Client in its contacts as well as the scores of these same Individuals in AssessFirst questionnaires.

      These elements will be analysed to create a predictive model for the Client recruiter, who can then highlight certain criteria that are important in a particular sector or specific to the Client (it is possible to personalise the criteria for a given post on the basis of 41 criteria determined by AssessFirst).

      Are a Client's personalised predictive models isolated from other customer environments?

      Yes, the customised templates created for one Customer are completely isolated from the environments of other Customers. Each model is generated from the Customer's job card data and stored independently, without any sharing or cross-referencing of information between customers. This approach guarantees the confidentiality of the data and results specific to each organisation using our solution.

      How does the Competency Framework work?

      The Competency Framework is a function designed to assess and develop employees' key competencies, in line with the company's values and needs.

      Each Client can create a bespoke skills repository, organised into clusters (e.g. leadership, performance, adaptability) grouping together different specific skills. Each skill is linked to personality traits, motivational drivers and/or cognitive abilities measured using AssessFirst questionnaires.

      Once the framework has been shared with an employee, they can access their results directly in their personal space. They can then see their potential for success and their potential for development for each skill, accompanied by an explanatory report.

      When activated, the Development Plan module enables employees to choose up to three skills to develop. For each skill, a personalised action plan is generated based on the employee's objectives and declared level of autonomy. This plan includes habits to adopt, concrete actions to take and resources to support progress.

      Finally, monitoring tools are available on the recruiter side (notifications, statistics), and each plan can be adjusted at any time.

      How are the skills to be developed identified as part of the Competency Framework?

      The skills to be developed are selected on the basis of two factors: the potential for success (what the person can do) and the potential for fulfilment or development. The system highlights skills that are both motivating and sufficiently accessible to enable real development. The aim is to encourage concrete, sustainable progress that is in line with each individual's deepest aspirations.

      How does VOICE work?

      VOICE is an assessment function developed by AssessFirst which enables the technical skills and language levels of individuals to be assessed on the basis of oral responses. It is based on a processing chain entirely automated by artificial intelligence, without human intervention.

      When a recruiter wants to create a VOICE assessment, they start by importing the job description. Based on this description, key skills are automatically extracted, and then assessment questions are generated based on the job, the level sought, the company's communication tone and its business context. The recruiter chooses one of the proposed questions, which is then translated into the different languages available so that individuals can answer in their mother tongue.

      Individuals must then answer the questions in a simulated interview with an AI. The dimensions assessed vary according to the skills selected.

      VOICE assesses the following aspects of language skills: overall European level (A1 to C2), vocabulary, grammar, fluency, pronunciation, elocution and speaking speed.

      For business skills, VOICE assesses the following dimensions: overall level, relevance, consistency, business coverage, argumentation, development potential.

      The results are generated automatically, with no waiting time, and made available to the recruiter and the individual in the form of comprehensive summary reports, including more details for the recruiter. In this way, VOICE provides a reliable, contextualised and flexible assessment, adapted to all professional situations.

      What information do you record when an Individual takes the VOICE test?

      When an Individual uses VOICE, AssessFirst records :

      • the oral answers given by the Individual,
      • the written transcription,
      • the response time.

      Does the VOICE test use sensitive data within the meaning of the GDPR?

      When using VOICE, AssessFirst collects the voice of Individuals in order to verify that all responses come from the same Individual, without any biometric identification or authentication. In this context, voice does not constitute sensitive data within the meaning of the GDPR.

      Are voice recordings made using VOICE shared or available for consultation?

      No. The voice files and their transcriptions are not accessible to Clients. Only the evaluation report, detailing the level achieved for each skill and/or language, is made available to them.

      How does the AI Assistant work?

      The AI Assistant is an intelligent chatbot integrated into the AssessFirst platform, designed to help recruiters quickly explore an individual's profile using free-form questions. It uses natural language to obtain personalised, contextualised and directly usable insights based on assessment data and documents associated with the individual.

      When a Recruiter interacts with the chatbot, it queries a vector database enriched with profile information (questionnaire results, text analyses, suitability scores, skills entered, etc.). Using an RAG (Retrieval-Augmented Generation) mechanism, the most relevant documents and data are extracted, then a response is dynamically generated.

      The answer combines :

      • dynamic data about the individual (personality, motivations, reasoning, predictive model, business information, location, etc.)
      • and static resources (explanatory texts, guides, HR documents) integrated into the knowledge base.

      This Assistant is designed to make profile analysis faster, more intuitive and more strategic, while ensuring a high level of confidentiality and personalisation.