Although AI's implementation into CCaaS environments is progressing rapidly, recent research on its use in a utility company's operations is beginning to cast doubt on its supposed efficiencies.
The recent findings point to a disconnect between the theoretical benefits of AI and the practical challenges that affect workers using it.
Although major consultancies have been forecasting substantial workforce reductions through AI automation, the academic study's findings suggest these predictions may be overly optimistic.
The initial evidence from the study shows that the path to effective AI integration in contact centers is proving more complex than initially expected.
Issues with AI's Implementation
A study published on arXiv—a digital repository where researchers can share their latest findings - by researchers affiliated with a Chinese power utility and several Chinese universities examined how customer service representatives actually interact with AI assistance during their daily work.
Its aims were to examine AI being used to augment workers' experience instead of the customer's. Based on 13 semi-structured interviews with service representatives - including team leaders, shift supervisors, and those responsible for handling phone inquiries - the findings reveal significant gaps between AI capabilities and real-world performance.
A key finding of the study demonstrated that AI transcription accuracy remains a persistent challenge. Customer service representatives reported that AI systems frequently struggled with caller accents, pronunciation variations, and speech speed, leading to inaccurate text conversion. This becomes particularly problematic when dealing with critical information such as phone numbers, which AI systems often fragmented into incomplete segments that required manual reconstruction.
The research also highlighted problems with AI's ability to handle homophones - words that sound identical but have different meanings or spellings. Additionally, the AI's emotion recognition system demonstrated poor performance, misclassifying normal speech patterns as negative emotions and inappropriately linking volume levels to customer attitude. For AI systems that try to ascertain tone so that the agent can better mimic the customer's sentiment, this could result in widely different approaches to customer interactions.
This led researchers to conclude that while AI reduced basic typing tasks, it simultaneously introduced what they termed "inefficiencies in information processing." Representatives reported that AI-generated content frequently required manual correction or deletion, creating additional work rather than reducing it. Text summaries of calls, while potentially useful, often needed substantial editing and failed to capture key information consistently.
This issue is critical, as one of the main ways AI is being advertised to improve the agent side of the equation is by reducing the manual administration associated with calls. Not being able to accurately transcribe may force workers to revisit conversations to look for errors, eliminating the time savings AI promises.
This led the research to conclude that AI implementation actually increased the learning burden on customer service representatives, who must adapt to new systems while maintaining service quality, far from the supposed efficiencies AI is meant to bring to workers.
This mismatch between technological expectations and actual implementation reflects a common oversight among technology designers who overestimate efficiency gains while underestimating the human cost of system adaptation.