Should an AI call coach score an employee's emotions?
An AI call coach that finds a missed refund explanation gives a support lead something to teach. A label saying the employee sounded unhappy asks the manager to judge the person instead. The EU AI Act prohibits workplace emotion inference from biometric data, with exceptions for medical or safety uses. Article 3 defines emotion recognition around biometric data; this is not a blanket ban on transcription or every text-analysis tool. Recital 44 points to reliability problems and the power imbalance at work. These are existing rules, not a new enforcement announcement. Before buying a call-coaching tool, I would ask the vendor to show exactly what it scores. Does it point to a missing explanation in the call, or infer an attitude from someone's voice or face? I would keep coaching tied to things the employee can inspect and change, and get legal review before enabling any emotion feature. What should a support lead refuse to score, even when the software offers it?
Comments
Even a ‘missed explanation’ flag needs a way for the employee to show the tool was wrong. I would check a sample against the actual call before using those flags in coaching, including calls where the customer interrupted or already knew the answer. How many flagged omissions were real, and how long did employees spend contesting the others? Keep unverified flags out of performance records. Otherwise the time saved by a support lead becomes time spent by staff defending a conversation the lead never listened to.
Recital 18 draws a useful line: detecting a raised voice is not itself emotion recognition, unless the system uses it to infer an emotion. ‘Volume increased here’ and ‘this employee was angry’ are different claims. I would ask the vendor to show the step between the recording and the label, including what lands in the manager’s report. That distinction is not a blanket permission to monitor staff; other rules still need review. But it helps an employee challenge what was actually asserted about them, rather than having to defend their character against a score.
A perfectly accurate script-compliance score can still reward a bad call. If a customer wants to cancel, does skipping the retention pitch count against the employee? I would ask a vendor to score that example before buying. The manager should be able to mark the omission as good judgment and change the scoring rule for everyone—not make each worker appeal the same penalty. Some scripts deserve to be followed less faithfully.