We all know how it feels to be low on energy at the end of a long work day. Some call-center agents at insurer MetLife are watched over by software that knows how it sounds.
A program called Cogito presents a cheery notification when the toll of hours discussing maternity or bereavement benefits show in a worker’s voice. “It’s represented by a cute little coffee cup,” says Emily Baker, who supervises a group fielding calls about disability claims at MetLife.
Her team reports that the cartoon cup is a helpful nudge to sit up straight and speak like the engaged helper MetLife wants them to be. The voice-analysis algorithms also track customer reactions. When call agents see a heart icon, they know the software has detected a heightened emotional state, either positive or negative. Baker says that emotional sixth sense can be helpful when talking with people adjusting to life’s most stressful events. “If a call becomes not so positive, it lets the associate to know to offer a little bit of hope,” she says.
Voice-controlled virtual assistants such as Siri and Alexa are becoming common in homes. MetLife is part of a quieter experiment using the same underlying technology to create superhuman helpers that are still part human. In workplaces, though, deploying smarter software will yield consequences more complex than just “Algorithms in, humans out.” A McKinsey report last year concluded that many more jobs will be transformed than eliminated by new technology in coming years.
Call centers have often been on the frontline of changes in labor and technology. A wave of US companies outsourced call centers to cheaper countries such as India and the Philippines beginning in the 1980s. More recently, voice recognition technology has been enthusiastically embraced to automate simple tasks that once required a human on both ends of the line, such as checking a bank balance or confirming a medical appointment.
Anyone who’s spent time with Siri or Alexa knows that computers are far from good enough with language to replace a human customer service agent. But MetLife and other early adopters say call center workers become more empathetic and efficient when they have a machine-learning powered wingman that can recognize words and traces of emotion.
MetLife’s empathy adviser was developed by MIT spin-out Cogito. CEO Josh Feast says his software can detect signs of distress and other emotions in a customer’s voice thanks to Pentagon-funded research at the university’s Media Lab.
In a project intended to help veterans with PTSD, the Department of Defense paid for medical staff to interview patients with psychological problems and annotate audio files of the data to mark changes in emotional state. That provided the perfect feedstock for machine-learning software now used in industries including healthcare and financial services, says Feast. Cogito applied deep-learning algorithms like those behind the improved speech recognition in assistants like Alexa to the Pentagon’s data, and audio from call centers and other sources.
In addition to nudging call agents to pep up their tone, or respond to distress in a caller’s voice, Cogito’s software listens to the pace and pattern of calls. Agents see a notification if they start speaking more quickly, a caller is silent for a long time, or the caller and agent talk over each other. Humans can notice all those things, but struggle to do so consistently, says Feast. “We’re trying to help someone doing 60 calls a day, and who may be tired,” he says.
Eavesdropping by corporate, emotionally aware software, may bother some consumers, even those used to being watched by cameras and online tracking. Analyzing a customer’s voice doesn’t require additional disclosure beyond the familiar line advising that calls may be monitored. “I’m habituated to that warning but this feels different,” says Elaine Sedenberg, a graduate researcher and co-director of the Center for Technology, Society & Policy at the University of California Berkeley. “I’m not expecting that extra layer of analysis.”
Sedenberg also questions whether technology like Cogito’s works equally well across different groups of people, which could lead to disparities in service between different social or ethnic groups. The company says it has tested the software on a range of demographics, and that non-verbal cues are more reliable across languages than analyzing the words people say. As well as US customers like MetLife and Humana, Cogito is used by Zurich Financial, where the primary language is German. Cogito does not provide guidance to customers on what it would consider inappropriate uses of the product, but says it is designed to only deliver insights that improve customer relations.
MetLife says it’s already seeing Cogito’s insights paying off. The company has deployed the technology in three customer-call teams in different parts of the business, and is planning a wider rollout. Kristine Poznanski, executive vice president of customer solutions, says that the initial deployments have driven customer satisfaction up, and the average duration of calls down.
State Collection Service, which handles collections for many healthcare companies from three call centers in Wisconsin, reports similar gains from deploying a competing product. The company has customized a real-time call monitoring tool from CallMiner, based in Waltham, Massachusetts. It is based on speech analysis technology from Nuance, which for a time powered the speech recognition capabilities of Apple's Siri.
Employees at State Collection see messages of congratulation and cute animal photos when software suggests a customer is satisfied, for example. When tone and language suggest a caller is getting worked up, employees see a suggestion to “Calm down” and a list of soothing talking points. If a worker omits legally required disclaimers, the system sends a reminder.
Chief operating officer Tracy Dudek estimates that a worker might see three to five notifications a minute during a typical call. That might sound like a lot, but she says call agents like it because it helps them in their job, boosting their chances of a performance-related bonus. The system has helped bring about a significant jump in the number of cases resolved in a single call, and increased revenue per call agent, Dudek says.
As you might expect, making some workers more productive can mean there’s less work for others to do. A few years ago, State Collection introduced a system that uses speech recognition to retrospectively transcribe every call. Since then, Dudek says the company has reduced the number of people and amount of time dedicated to reviewing calls for quality assurance. The newer, real-time voice analysis has helped the company increase the number of call agents overseen by a single supervisor, she says.
The technology has also led to new hires. The CallMiner system can spot shifts in tone or stress in a person’s voice, but it doesn’t automatically decide which phrases are most important to a specific business. So State Collection hired a new employee to identify signals and phrases for the software to watch for. One finding: A person saying “This is ridiculous” is the most reliable indicator they are becoming dissatisfied. Dudek also credits the call monitoring system with helping State Collections grow its business, which has led to more hiring.
As AI-assisted call agents become more common, expect companies to find new ways to use their powers. MetLife is deploying Cogito into a sales team pitching auto and home insurance. Poznanski says she’s interested in using the technology to spot when a person is about to say something signaling a lack of interest so the salesperson can get in first. “When a customer is potentially uninterested we could pick that up sooner to help position the value that MetLife brings,” Poznanski says.
Cogito is also working with the Veterans Administration to test an app that analyzes voice recordings and tries to give health staff a forewarning of major shifts in a patient’s mental health. Some of those who have experienced the technology in the call center could imagine it being useful in more casual situations, too. MetLife’s Baker says she’s thought it could help during presentations. When WIRED asked if she was using the software during a recent interview, she laughed, then turned thoughtful. “I wish I could, I’d love to know how I was doing.”
- Google Clips is a tiny camera that uses an on-board AI chip to suggest moments to snap a photo or record a video.
- The Pentagon envies Silicon Valley's AI chops. A contest for algorithms that can understand satellite photos aims to help the agency catch up.
- Images from Google's Street View cars are being mined by image-analysis algorithms to update the company's maps.
