> ## Documentation Index
> Fetch the complete documentation index at: https://docs.botbrains.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Grading employees

> What you can analyze in your team's conversations, and how to grade your team

[Insights](https://docs.botbrains.io/concepts/insights.md) and [ServiceLens](https://docs.botbrains.io/concepts/servicelens.md) analyze your customer conversations, including the ones only your team handled (AI involvement **Not involved**). Almost everything you do with them is ordinary service analytics. The line is AI judging how well an individual employee works: that's [high-risk under the EU AI Act](#ai-evaluation-of-employees-is-high-risk), and botBrains doesn't allow it. In Germany, the works council also has a say once you analyze your team's conversations at all.

> [!NOTE]
>
> For questions, email [legal@botbrains.io](mailto:legal@botbrains.io). This page explains how botBrains assesses these uses and doesn't replace legal advice.

## What you can do

| Use | Example |
| - | - |
| [Topics](https://docs.botbrains.io/concepts/topics.md) and [metrics](https://docs.botbrains.io/concepts/metrics.md) | See what customers contact you about, which topics grow, and how resolution differs by channel. |
| [Incident detection](https://docs.botbrains.io/concepts/incidents.md) | Spot a sudden spike in a topic, such as failing payments, and confirm it with customer quotes. |
| [Suggestions](https://docs.botbrains.io/concepts/suggestions.md) | Learn from past tickets. Group the questions your AI agent couldn't answer and reuse the replies your team already sent. |
| Customer signals with [conversation attributes](https://docs.botbrains.io/concepts/conversation-attributes.md) | Record what customers want and how they feel, such as a refund request, the product, or frustration about a late delivery. |
| Process analysis | Find procedures that customers misunderstand, such as a confusing returns policy, or steps that give contradictory instructions. |
| AI agent quality with [scorecards](https://docs.botbrains.io/concepts/scorecards.md) and [monitors](https://docs.botbrains.io/concepts/monitors.md) | Check your AI agent's answers, handoffs, and adherence to your procedures, and flag unsupported answers or refunds it can't authorize. |
| Business signals with [monitors](https://docs.botbrains.io/concepts/monitors.md) | Collect product feedback, buying signals, or competitor mentions from every conversation. |
| Manual quality review in the [review queue](https://docs.botbrains.io/concepts/review-queue.md) | Let people grade conversations with a scorecard whose items all use the fill mode **Manual**. See [Grading your team by hand](#grading-your-team-by-hand). |

## Customer, procedure, or employee

Ask what the result describes. Results about customers, procedures, or your AI agent are fine. A result that judges how well a person works is an employee evaluation.

| What the result describes | Example | Assessment |
| - | - | - |
| A customer | The customer expresses frustration about a late delivery. | Customer analysis |
| A procedure | Many customers misunderstand the returns policy. | Process analysis |
| Your AI agent | The AI agent gives contradictory instructions. | AI quality review |
| An employee | A particular employee gives unclear explanations or fails to calm customers. | Employee evaluation, high-risk |

The test is whether you evaluate **the company's procedure** or **how well a person follows or explains it**. "30% of requests are about delivery delays" analyzes customer issues. "Anna follows the guidelines in 82% of her tickets" evaluates Anna. Comparing sentiment, quality scores, or policy failures by employee crosses the same line.

That's why scoring conversations your AI agent wasn't involved in is difficult. Only a person answered in those conversations, so a scorecard or monitor that judges the quality of the answer evaluates that person, even without an employee profile. Attributes with the fill mode **On demand** or **After the conversation**, monitors with flag criteria, and scorecards built on such attributes all let the AI make that judgment.

## AI evaluation of employees is high-risk

Where the EU AI Act applies, AI that monitors or evaluates the performance or behavior of workers is high-risk under Art. 6(2) and [Annex III, point 4(b)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3). This covers employees and contractors, and it doesn't take an automated decision about pay, promotion, or termination.

This evaluation is profiling of people ([Art. 3(52)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-3)), so the exceptions in [Art. 6(3)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-6) don't apply. The classification stays the same when a person checks the result or when you call it coaching. Adding a manual item to an AI scorecard doesn't change it either, and neither does hiding names while tickets, timestamps, or shift plans still identify the employee.

So don't use AI attributes, monitors, or scorecards built on AI attributes to evaluate or compare individual employees, to pick them for coaching, or to prepare HR decisions.

## Why AI employee evaluation is currently not allowed

botBrains doesn't offer Insights or ServiceLens as a high-risk AI system. For employee evaluation, botBrains hasn't carried out a conformity assessment, set up a quality management system under Art. 17, drawn up technical documentation under Annex IV, issued an EU declaration of conformity, affixed CE marking, or registered the system in the EU database. A person reviewing the results doesn't close that gap.

Under [§ 11 of the Terms of Service](https://docs.botbrains.io/terms-of-service.md#%C2%A7-11-ki-verordnung), any high-risk use requires a prior written agreement with botBrains. If you're interested in a high-risk use case, contact [legal@botbrains.io](mailto:legal@botbrains.io) before you configure anything. This restriction applies regardless of when the high-risk requirements of the EU AI Act take effect.

## Customers outside the EU

The EU AI Act can also apply to companies outside the EU, for example when someone in the EU uses the AI results or the results affect people there ([Art. 2](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-2)). Assess the actual deployment, not only where your company or your employees are. The restriction on AI employee evaluation applies to every botBrains customer, whatever that assessment shows.

## Fines and deadlines

| Law | What triggers the fine | Maximum fine |
| - | - | - |
| EU AI Act, [Art. 99(3)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99) | A prohibited practice under [Art. 5](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-5), such as recognizing employees' emotions at work | €35 million or 7% of worldwide annual turnover |
| EU AI Act, [Art. 99(4)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99) | Breaching the deployer obligations for a high-risk AI system under [Art. 26](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26), such as AI employee evaluation | €15 million or 3% of worldwide annual turnover |
| GDPR, Art. 83(5) ([Regulation (EU) 2016/679](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng)) | Processing employee data without a legal basis or beyond what's necessary | €20 million or 4% of worldwide annual turnover |

The higher of the two amounts sets the limit. Under the AI Act, small and medium-sized companies and start-ups face the lower one ([Art. 99(6)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99)).

The prohibitions in Art. 5 apply since February 2, 2025, and fines for them since August 2, 2025. The obligations for high-risk AI systems listed in Annex III, which includes employee evaluation, apply from **December 2, 2027**. The Digital Omnibus on AI moved this date from August 2, 2026 ([Regulation (EU) 2026/1744](https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng)). The GDPR has applied since May 2018, so data protection fines are possible today. botBrains' restriction on AI employee evaluation also applies today, whatever the AI Act dates.

Data protection authorities already fine employee monitoring. The Hamburg authority fined H\&M €35.3 million in 2020 for recording details of employees' private lives, such as illnesses and family issues ([EDPB](https://www.edpb.europa.eu/news/national-news/2020/hamburg-commissioner-fines-hm-353-million-euro-data-protection-violations_en)). The French data protection authority fined Amazon's French logistics company €32 million for tracking warehouse staff's activity through their scanners ([CNIL](https://www.cnil.fr/en/employee-monitoring-cnil-fined-amazon-france-logistique-eu32-million)). In December 2025, France's highest administrative court reduced that fine to €15 million ([decision 492830](https://www.conseil-etat.fr/fr/arianeweb/CE/decision/2025-12-23/492830)).

## Grading your team by hand

You can grade your team's conversations when people make every judgment. Use a scorecard whose items all use the fill mode **Manual**, and let reviewers answer them in the [review queue](https://docs.botbrains.io/concepts/review-queue.md). Monitors then only pick which conversations to review, for example by filtering **AI involvement** to **Not involved** or by topic, without judging how well someone worked. botBrains only calculates the weighted score.

Data protection law still applies to manual reviews, so you need a legal basis for processing employee data. In Germany, involve the works council first.

## Works councils in Germany

In Germany, the works council has a co-determination right over introducing and using technical systems that can monitor the behavior or performance of employees ([§ 87(1) no. 6 BetrVG](https://www.gesetze-im-internet.de/betrvg/__87.html)). The capability is enough. A system qualifies when it can collect and record behavior or performance data about employees, whatever the employer intends. That's how the Federal Labour Court reads the rule ([decision 1 ABR 7/15](https://www.bundesarbeitsgericht.de/entscheidung/1-abr-7-15/)).

Insights and ServiceLens analyze the conversations your team handled, and names, ticket assignments, or timestamps link these conversations to individual employees. So the co-determination right applies as a rule, even if you only look at topics and customer signals and never evaluate your team.

If your company has a works council, involve it before you analyze your team's conversations. Agree in a works agreement which analyses you run, who sees the results, and that you don't use them to evaluate individual employees. Because Insights and ServiceLens use AI, you also inform and consult the works council ([§ 90(1) no. 3 BetrVG](https://www.gesetze-im-internet.de/betrvg/__90.html)). It can bring in an expert to assess the AI ([§ 80(3) BetrVG](https://www.gesetze-im-internet.de/betrvg/__80.html)).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.