AI in Employee Conversations: How SME Leaders Finally Handle Difficult Conversations With Confidence
Every service company has conversations that managers keep putting off. The overdue feedback for a senior who has been underperforming for months. Clearing the air between two strong personalities on the same team. The talk with a key person who has hinted at looking elsewhere.
These conversations are rarely avoided out of incompetence. Usually something more mundane is missing: an overview of the past few weeks, concrete examples, and confidence that whatever is agreed will still hold months later. The audit disaster adds a traceability perspective on reviewed conversation content and approvals. For SMEs, appropriately restricted knowledge preservation keeps agreements available, while knowledge validation requires someone responsible to check the record before it shapes a sensitive conversation.
My view: AI can do a lot here, but almost all of it before and after the meeting. In the conversation itself it should do as little as possible, because a manager who gets prompted from a screen loses exactly the composure that a conversation like this needs. The other person notices.
The surveillance objection is fair as long as there are no rules
Many managers, and even more employees, associate AI in employee conversations with surveillance. From a change and usability perspective, I think the cause lies less in the technology than in a lack of transparency. Three rules take away much of the ground for the objection:
- The employee sees the same version of the notes as the manager and can comment before they are filed.
- The AI does not assess people. It records what was said and agreed.
- It is clear in advance which conversations are documented and who may read the notes.
The legal side, from consent to the works council, is covered in Documenting employee conversations.
Preparation: what belongs on the table beforehand
Before a sensitive conversation, RECO summarizes what came up in recent conversations, what was agreed and what is still open. The manager no longer walks in cold.
A typical pattern looks like this. A team lead wants to talk to a senior consultant about missed deadlines, and her gut says it happens all the time. The notes from recent weeks can show a more precise picture. Perhaps deadlines were pushed back in team meetings because another project took priority. Perhaps there is an open commitment from the team lead herself to sort out priorities with the department head. "You don't meet deadlines" then turns into a different, more honest conversation.
Whatever tool you use, I recommend answering three questions in writing before any difficult conversation:
- Which two situations do I want to raise, with dates?
- What should be different afterwards, in one sentence?
- What have I promised myself and not yet delivered?
The last question is uncomfortable. On purpose. That is exactly why it changes the conversation. To open, I suggest naming the first situation and your intention straight away, for example: "I'd like to talk about the deliverables on the project, because I want to understand where things are getting stuck."
During the conversation: use prompts sparingly
RECO can flag points from the preparation that haven't been raised yet. In routine meetings that's handy. In a critical or conflict conversation, I would turn the prompts down to a minimum or switch them off. Every glance at the screen costs eye contact, and the other person notices immediately.
What gets said, agreed or escalated is always a human decision. The AI suggests, the manager approves, and the post on human-in-the-loop describes in more detail what that principle looks like day to day.
After the conversation: one version for both sides
Afterwards there are notes with agreements, responsibilities and follow-up dates, without a long write-up. They only become valuable once both sides have seen them. Not before. Notes only the manager knows about are a memo. Notes both have confirmed are an agreement you can still refer back to months later.
How AI helps by type of conversation
- For overdue feedback, preparation matters most, as in the pattern above.
- For a retention conversation, the view across several months helps. If the same development wish comes up in several team meetings without ever making it into an agreement, that is a signal long before a resignation lands on the desk.
- For resolving conflict in a team, a traceable record of what was agreed when, and what has happened since, helps. Restraint matters here too: notes from conflict conversations don't belong in the personnel file, and in deeply entrenched conflicts no tool replaces neutral facilitation.
How management can tell whether it works
Don't decide whether this pays off based on vendor promises, ours included. Three numbers are easy to track:
- share of employee conversations with notes confirmed by both sides
- share of agreements marked as done or deliberately postponed at the follow-up meeting
- time managers report spending on preparation and follow-up per conversation, before and after
If none of these moves after a quarter, the first thing I would check is whether anyone actually opens the notes in the follow-up conversation. As a side effect, the knowledge from these conversations stays in the company even when a manager moves on.
If you change only one thing tomorrow: before your next difficult conversation, answer the question of what you still owe the other person.
Sources
Frequently asked questions
Can AI analyze emotions or mood in employee conversations?+
No. Since 2 February 2025, Article 5(1)(f) of the EU AI Act has prohibited AI systems that infer people's emotions in the workplace, except for medical or safety reasons. According to the European Commission's guidelines, the ban also applies to the employer deploying such a system, even if the provider excludes that use. So ask every vendor explicitly whether their tool analyzes tone of voice, facial expressions or mood. Not legal advice.
Does the employee have to agree when the conversation is documented with AI?+
For an audio recording, yes. In Germany, recording the privately spoken word without authorization is a criminal offense under section 201 of the Criminal Code (StGB), and consent under data protection law must be voluntary and revocable at any time (Art. 7 GDPR). Saying no must not lead to any disadvantage; the conversation then simply goes ahead with handwritten notes. Minutes of outcomes alone need no consent, but they do need a legal basis under Art. 6 GDPR. If there is a works council, it must be involved when the system is introduced. Not legal advice.
Related articles
In a 30 minute software demo we show you how RECO secures and transfers knowledge in your company. Request a demo now and we'll get back to you with a proposed time.
