Why it matters: Stories about AI’s capabilities being overhyped are becoming increasingly common. After recently hearing that many agentic AI projects fail to live up to expectations, customer service reps in a call center say even their AI assistants are proving more of a hindrance than a help.
A study carried out by researchers from several Chinese universities and a Chinese power company looked at what impact AI assistants were having on the plant’s customer service reps (CSRs). The results did not paint AI as the miracle assistive technology its creators often portray it as.
One of the biggest complaints was that when transcribing customer audio calls into text, the AI was filled with inaccuracies due to callers’ accents, pronunciation, and speech speed. The AI also struggled whenever it had to turn audio consisting of sequences of numbers into text, often getting the likes of phone numbers wrong.
One CSR who took part in the study said, “The AI assistant isn’t that smart in reality,” adding that “It gives phone numbers in bits and pieces, so I have to manually enter them.”
Homophones, words that have the same pronunciation but different meanings, such as new and knew, were another problem area for the AI assistant.
Emotion recognition technology, something we’ve seen several reports about – most of them not good – is also criticized by those interviewed. It often misclassified normal speech as being a negative emotion, had too few categories for the range of emotions people expressed, and often associated a high volume level as someone being angry or upset, even if it was just a person who naturally talks loudly. As a result, most CSRs ignored the emotional tags that the system assigned to callers, saying they were able to understand a caller’s tone and emotions themselves.
Ultimately, while the AI assistant did reduce the amount of basic typing required by CSRs, the content it produced was often filled with errors and redundancies. This required workers to go through the call summaries, correcting mistakes and deleting sections. Moreover, the AI often failed to record key information from customers.
“While the AI enhances work efficiency, it simultaneously increases CSRs’ learning burdens due to the need for extra adaptation and correction,” the report concludes. “The mismatch between technological expectations and actual implementation reflects a common oversight among technology designers, who overestimate efficiency gains while underestimating the implicit learning burdens of adapting to new systems.”
The report highlights other problems that AI integration is facing, including employee pushback against the technology’s use in call centers and the stress it causes over feared job losses. There’s also the customer factor, with many people refusing to use a company that relies so heavily on AI for customer service.
In June, a survey from Gartner found that 50 percent of organizations that had planned to replace CSRs with AI were expected to reverse their decision. More recently, the research firm predicted that over 40 percent of agentic AI projects will be canceled by 2027.
