Co-op Legal Services uses AI to rate workers' phone calls with customers

According to The Guardian, Co-op Legal Services is using an AI model to record, analyze and give percentage scores to interactions between agents and customers concerning probate, wills and estate consultations. An OpenAI model was trained to assess more than 50 separate aspects of each call and provide managers with pass-or-fail scores, which managers then use to analyze employee performance.

The system is believed to have been introduced over the past few months into the work of dozens of legal services employees, and is partly being used to identify ways for employees to improve sales performance. One employee said the system could mean being monitored by AI for hours each day while handling consultations—consultations that often come from customers who have just lost a loved one.

An anonymous whistleblower described the system as 'oppressive and dystopian,' saying that every word and every utterance by employees is monitored, and that this creates a sense of distrust, entirely different from occasionally being listened in on for compliance purposes.

The Co-op, for its part, said it does not accept the criticism and described its probate consultants as highly committed. The company said the AI is used only as a support tool, not a decision-maker, helping leaders support colleagues and continuously improve the customer experience; it is also 'a key aspect of its quality assurance process,' helping ensure customers consistently receive empathetic, professional guidance.

Caoilionn Hurley, Managing Director of Co-op Life Services, further said that the efficiency of AI relative to traditional methods allows it to review all customer conversations while providing colleagues with more informed support, coaching and development. Here a key disagreement emerges: the company says the AI 'is not a decision-maker,' yet the same system is used to analyze employee performance and identify ways to improve sales—how the boundary between these two is defined is not answered by the report.

This is not an isolated incident. The Guardian revealed last week that Euan Blair's AI training company Multiverse has begun using AI to conduct full-coverage monitoring of online classrooms, with teachers calling it 'relentless' and 'unsettling.' In February this year, Burger King in the United States announced it would use the technology to track restaurant employees' customer interactions; this summer, Meta paused tracking employees' computer keystrokes due to data privacy concerns and employee opposition.

Unions have already begun to push back. John Chadfield, a national officer of the Communication Workers Union, said irresponsible computer systems should not become people's managers, and that these hard-working employees need more respect. Hayfa Mohdzaini, senior policy and practice adviser at the UK's Chartered Institute of Personnel and Development, said that as performance feedback can be given in a more structured way, offices are becoming more like factories, and clearly some people have a problem with that.

The International Labour Organization warned this year about the 'psychosocial' risks brought by AI monitoring, including increased work-related stress, and said AI systems are being deployed in workplaces without a comprehensive regulatory framework. Experts also worry that growing AI employee monitoring may increase stress.

Putting these threads together, the Co-op case has drawn attention not just because of the technology itself, but because it touches on a question that has yet to be answered: when AI can evaluate word by word an employee handling probate consultations, who judges whether the scores are accurate, whether employees can appeal, and whether these scores will ultimately affect employment decisions. The report provides no public information about the system's accuracy rate, error rate or appeal mechanism.

It should be made clear that this report is based on a single source, The Guardian, and provides no independent third-party verification. The Co-op's response to the criticism and the whistleblower's account contradict each other, and neither side has been externally verified. The specific timing of the AI monitoring's deployment is described only vaguely as 'the past few months'; the specific name, version and training details of the OpenAI model have not been disclosed; and the system's actual impact on employees is conveyed only through an anonymous employee and the union.

In addition, it remains unclear whether the AI scores are actually used for employment decisions such as dismissal, promotion or pay; the exact number of monitored employees and the scope of coverage have not been confirmed; whether the system obtained employee consent or underwent a data protection assessment has not been stated; and the report gives no regulatory conclusion on whether such AI monitoring is legal and compliant in the UK. The whistleblower's description of 'oppressive and dystopian' is a subjective feeling that cannot be objectively verified.