Artificial Intelligence: How It Works and Use Safeguards

1.1AI is used to practice a sales conversation in a fictional situation, prepare scenarios, and analyze the content of a practice session. Depending on the features included in their plan and the configured permissions, managers can track their teams’ progress.

1.2The Service does not hold a conversation with a real prospect on the User’s behalf. The identities, voices, and situations of simulated customers are used for the exercise. Behaviors assigned to a persona are simulation settings; they do not correspond to an emotion detected in the User. Website demos use fictional personas and results.

1.3Scores and advice are learning aids. They are not a diploma or professional qualification recognized by the state, a certain prediction of success, or a complete assessment of a person's professional aptitude.

2.1Before each call, a notice informs the User that they are speaking with an AI-simulated customer. The initial workflow also asks the User to acknowledge the use of AI and transcription. This step may be repeated when the information version changes.

2.2Call reports and coach responses are identified as AI-generated. This indication distinguishes automated output from a comment written by a person.

2.3An acceptance checkbox confirms that the information has been presented in the Service. By itself, it does not constitute the legal basis for all processing decided by the employer, nor does it replace the employer’s obligation to inform employees in advance.

3.1The scenario, practice instructions, and authorized context are used to configure the fictional customer. The voice provider processes the User’s speech and generates the responses needed for the conversation. Quality depends in particular on the microphone, connection, language used, and relevance of the scenario.

3.2The conversation is transcribed into text. The application does not retain any audio recordings of calls. In the voicemail feature, temporary in-memory processing enables transcription, after which the audio is erased from memory.

3.3Artificial intelligence model providers are involved depending on the features used. Voice data may be processed in the United States; location is not guaranteed for all providers. Provider categories, locations, and safeguards are listed in the list of Sub-processors and the DPA; the list by name is disclosed to customers. The absence of an in-app recording does not mean that these providers do not process audio.

4.1The practice analysis uses the text transcript, the exercise context, and the applicable scorecard. It searches the text for elements related to the exercise criteria and produces scores, observations, excerpts, or advice.

4.2The Service does not score the acoustic characteristics of the voice. It does not infer a person’s emotions, stress, personality, or health status. It does not perform biometric identification by voice. Scorecards must not include criteria intended for such analyses.

4.3A transcription error may affect the result. A criterion or weighting may not be appropriate for the context. The call report should therefore be read alongside the exercise and scorecard settings; a number on its own is not enough to explain performance.

5.1Supplied documents and instructions may be used to extract contextual details or suggest scenarios. Any generated suggestion must be reviewed before activation, in particular to correct facts, remove unnecessary data, and adapt the wording to the relevant profession.

5.2The coach suggests rephrasing, ideas, and exercises based on the available context. Its responses remain suggestions; they do not constitute validated legal, medical, financial, or regulatory advice. Users should have information that requires expertise reviewed by a qualified professional. Each message sent to the coach is deducted from the number of minutes, at a rate of twenty seconds per message.

5.3Generated content may resemble content produced for another user. Its exclusivity, originality, or freedom from third-party rights is not guaranteed. The Customer must verify rights and accuracy before distributing it beyond its training use.

6.1Generative systems may produce a false statement, omit information, wrongly attribute a statement, or interpret an exchange inappropriately. Confident wording is not proof of reliability.

6.2Results may be affected by text quality, language, accents in transcription, cultural context, the scenario, or the criteria selected. Results for different people or periods are comparable only if the conditions of their exercises are sufficiently similar.

6.3The Service is not a tamper-proof means of proving what was said. Transcripts and analyses must be assessed alongside other available information. They are not presented as sufficient proof of actual behavior with a customer.

6.4The Customer reviews persistent discrepancies and results that appear discriminatory or inconsistent. The Customer suspends use of the result in question for decision-making and seeks an explanation from the person concerned and, if necessary, from support.

7.1The User may compare the call report with their transcript, report an error, and discuss it with the person responsible for the practice session. The Customer appoints a contact who can review challenges, explain the criteria, and decide not to rely on an unsuitable result.

7.2Sales managers verify information they are legitimately authorized to access, while respecting private feedback and access permissions. When individual access is limited, human review is arranged through a discussion with the person and does not justify circumventing privacy settings.

7.3The Service must not be used to make an exclusively automated decision about hiring, promotion, compensation, discipline, or termination. Human involvement must be meaningful: the qualified person must have the context and the ability to challenge the AI output.

7.4The Service lets the Customer enable a ranking of sales reps based on practice session scores. This setting is the Customer’s responsibility. The Customer assesses its proportionality, informs employees and, where applicable, the CSE (works council), and does not use it as a disciplinary tool. The Customer must not turn a training indicator into an automatic disciplinary ranking. A certificate related to practice sessions does not replace human assessment of skills in real-life situations.

8.1Before deployment, the employer informs individuals, in accordance with Article L1222-4 of the French Labor Code (no information concerning an employee personally may be collected using a system that has not been brought to the employee’s attention in advance) and Article L1222-3 (information about assessment methods and techniques, which must be relevant to the intended purpose), of the system’s purpose, how the exercises work, transcription, text-based scoring, who receives the results, and any potential use of those results in professional monitoring.

8.2The employer communicates the legal basis relied on, retention periods, processing and transfer locations, privacy settings, rights, and contact details. The employer specifies whether exercises are required by the employer or available at each person’s initiative, and explains the consequences of that choice within the organization.

8.3The information distinguishes AI outputs from human assessments. It explains how a person can report an error, request a review of their case, and exercise their rights. Documents made available through the Service support this process but do not replace information tailored to the company’s context.

9.1Before introducing the system and making significant changes to it, the Customer assesses its obligations to inform and consult employee representatives. When legal conditions are met, particularly for systems that enable activity monitoring or affect working conditions, the Customer completes the required steps with the CSE (works council) before implementation, in accordance with Articles L2312-8 (introduction of new technologies) and L2312-38 (means or techniques enabling the monitoring of employee activity) of the French Labor Code.

9.2The existence of a notice or documentation provided by the Publisher does not constitute an opinion from the CSE (works council) or approval of the system by employee representatives. The Customer describes the uses it actually intends to make, access permissions, and safeguards provided to employees.

9.3The Customer assesses the necessity and proportionality of the processing and whether a data protection impact assessment is required. Under the DPA, the Publisher assists the Customer with the information it has about the features, data, providers, and security measures.

10.1Regulation (EU) 2024/1689 on artificial intelligence, known as the AI Act, provides in particular for transparency obligations (Article 50), AI literacy obligations for staff (Article 4), and rules relating to certain uses in the workplace. Before each call, the Service informs users that they are interacting with AI and identifies its call reports and coach as AI-generated.

10.2The Service is designed to provide text-based training feedback. It does not perform emotion recognition in the workplace, a practice prohibited by Article 5(1)(f) of the Regulation. The Customer must not seek to add such criteria or repurpose the results for this purpose.

10.3A system’s regulatory classification depends on its intended purpose and actual uses. Use for workplace assessment or monitoring, recruitment, or employment decisions may be subject to specific requirements, including those for high-risk systems (Annex III, point 4, of the Regulation). A training purpose alone does not rule out these requirements for every conceivable use.

10.4The Customer must not change the Service’s intended purpose to turn it into an automated HR decision-making tool. The Customer ensures that users and managers have an appropriate level of AI literacy, including an understanding of how AI works, its limitations, critical interpretation of scores, privacy, and available recourse. This notice does not constitute general certification of AI Act compliance or authorization for every use within a company.

11.1By default, transcripts are retained for 180 days and call reports for 365 days. Starting with the Business plan, the Customer can set these periods to between 30 days and three years. Coach conversations follow the transcript retention period. Detailed retention periods, exports, and deletion procedures are set out in the Privacy Policy and the DPA.

11.2Users should avoid stating contact details or confidential information about real prospects and should use fictional examples. Documents submitted must be limited to what is necessary, with the required authorizations.

11.3The Customer sets access permissions for results and regularly reviews whether access is still necessary. A score or training comment may constitute personal data; the fact that it was generated by AI does not exempt it from the person’s rights.

12.1To report an error in a transcript, call report, or the way the Service works: support@clozing.example. For questions about personal data: privacy@clozing.example. To report a security incident: security@clozing.example. Specify the exercise and the issue without sending unnecessary data.

12.2This notice is updated when there are significant changes to the features or uses described. The Terms of Service, the General Terms of Sale, the Privacy Policy, the DPA, and the sub-processors supplement this information.