--- title: "Scoring and the AI Act | Natalia Analytics" description: "What the scoring of conversations and agents measures, human correction of scores, the articles of Regulation (EU) 2024/1689 and the French Labour Code." url: https://getnatalia.com/en/documentation/natalia-analytics/scoring-and-ai-act lang: en --- [Home](https://getnatalia.com/en/) / [Documentation](https://getnatalia.com/en/documentation/natalia-analytics) / Scoring and the AI Act Cloud Edition [See On-Premise Edition →](https://getnatalia.com/en/natalia-analytics-on-premise) # Scoring and the AI Act What the scoring measures, how your team corrects it, and what to tell your agents Natalia Analytics scores conversations, and through them the agents who led them. This page describes that scoring as the product implements it, and cites the texts that frame its use at work: Regulation (EU) 2024/1689 (the "AI Act") and the French Labour Code. Neither legal advice nor a declaration of conformity This page states facts about the product and cites the texts. It does not qualify the scoring under the AI Act and does not certify any conformity. ## What is scored - **Per conversation.** Each analysed conversation receives a score from 1 to 5 per criterion of its grid. Its aggregated score is the simple average of those scores, out of 5, to one decimal. The analysis starts by itself at the end of the conversation, or when the transcript of an imported recording arrives. - **Per agent.** Trends, Quality, "By agent" shows the average score of each agent over the period, for the six busiest. The weekly recap lists up to thirty agents with their average score. Every member of the organisation reads Trends, Quality. Your organisation chooses the criteria: the grid is described in [Analysis grid and reanalysis](https://getnatalia.com/en/documentation/natalia-analytics/analysis-grid). Natalia scores and displays; what your organisation does with the scores is its own decision. ## Human oversight in the console A person of your team can replace any score the AI gives. Correcting requires the right to manage the organisation (Administrator role). - **Correct a score.** On the conversation page, the aggregated score opens the score of each criterion. A click on a star opens "Correct the score", which asks for a reason (required, 500 characters at most). The correction replaces the AI's score in the aggregated score and in Trends, Quality, and survives a reanalysis. The page shows its author, its date and its reason. - **Withdraw a correction.** "Withdraw the correction" makes the AI's score apply again. The correction stays in the log, marked "Withdrawn". - **Corrections log.** The "Corrections" tab of each grid lists its corrections: conversation, criterion, score, reason, author, date and state (Current, Replaced, Withdrawn). - **Export.** "Export as CSV" downloads the log under the chosen filters. ## What the AI Act provides - **Annex III, point 4(b)** lists among high-risk systems those intended to monitor and evaluate the performance and behaviour of persons in work-related relationships. - **Article 6(3)** sets the conditions under which a system of Annex III is not high-risk, and adds that a system of Annex III that performs profiling of natural persons is always high-risk. **Article 6(4)** requires the provider who considers its system not high-risk to document that assessment. - **Article 26** sets the obligations of the deployer of a high-risk system, among them: use it according to its instructions (paragraph 1), assign human oversight to persons with the competence, training and authority required (paragraph 2), and, for an employer, inform the workers' representatives and the affected workers before putting it into service or using it at the workplace (paragraph 7). - **Article 113** sets the dates from which these provisions apply. ## Informing your agents in France Whatever the qualification under the AI Act, the French Labour Code already frames a tool that scores agents: - **Article L1222-4:** no information concerning an employee personally may be collected by a device that has not been brought to their knowledge beforehand. - **Article L2312-38:** in a company of at least fifty employees, the social and economic committee (CSE) is informed and consulted, before the decision to implement them, on the means or techniques that allow monitoring of employees' activity. These steps fall to your organisation, as employer and deployer. For a question about the scoring, [contact us](https://getnatalia.com/en/contact). ## Back to the documentation [Analysis grid](https://getnatalia.com/en/documentation/natalia-analytics/analysis-grid) [Back to documentation](https://getnatalia.com/en/documentation/natalia-analytics)