Who Designs the AI That Scores Your Call?

The next workplace AI is being decided in a meeting room, not a model card. Every scoring system embeds a design choice: who builds the rules, who can appeal them, who corrects the machine when it errs, and whether the people judged by the system get a say before it goes live. Pannini and colleagues' November 2025 survey of more than 700 UK telecoms and finance call-centre workers finds 81.3% of agents are scored on every call, typically by an AI that reads tone as well as words. Most coverage reads this as surveillance. The pattern underneath is governance: the four levers are transparency, appeal, manager override, and pre-deployment consultation. Pannini's own numbers prove the frame. Where managers can correct AI mistakes and workers can challenge scores, sentiment toward the tools is less negative. Where workers are consulted only after the system is live, fear of tone and accent penalties rises. The EU has read the same evidence and codified those four levers as a December 2027 human-oversight requirement. The UK has not. The reusable rule: a scoring system is a labor contract written in code, and the party that owns the design owns the workplace. The mechanism is named. The deadline is fourteen months away.