For over a decade, a team of MIT scientists has been testing special stress‑tracking sensors on drivers in the Boston area. The wrist‑mounted devices monitor perspiration levels and other physical indicators of stress, then upload data after each commute.
One takeaway from multiple trials: People are frequently off base when reporting their own emotional states. Researchers frequently interviewed drivers about their biggest stressors in traffic, but didn’t glean many specifics beyond the usual: merging onto on‑ramps or getting stuck in gridlock.
Instead, drivers showed the most stress in response to situations that cropped up out of nowhere—for example, when a mom and a baby stroller crossed in front of one driver while cars from behind bore down.
The driving tests, led by Rosalind Picard, director of MIT’s Affective Computing group, are actually a useful proxy for workplace stress, a growing problem that is the subject of Picard’s current research.
Turns out people are equally bad at identifying and predicting stress at the office. The health consequences are alarming: Workplace stress is now the fifth‑leading cause of death in the U.S., according to research by Jeffrey Pfeffer, professor at Stanford University’s Graduate School of Business.
That’s one reason why academic researchers and a number of startups are pairing an established technology called emotion recognition with new forms of machine intelligence that can detect and reduce worker stress in ways that not even the most empathetic managers could do on their own.
Businesses have begun tapping Picard’s team and others to supply new tools that can identify stress and other emotional states. The use cases cover a wide spectrum, from managing customer service calls, to improving productivity in teams with low morale, to helping overextended employees maintain work‑life balance.
“A great way to find the inefficiencies in a business process,” says Picard, “is to find out where people are most frustrated and stressed.”
Sentiment detectors>stress detectors
In the early 2010s, market researchers adopted the first wave of emotion recognition tools. One common application was video analysis of consumer expressions as they tried new products.
Other apps crunched through millions of tweets to evaluate consumer sentiment about product launches. Today consumer sentiment analysis is a multi‑billion‑dollar industry that should grow rapidly over the next five years, according to market research firm Tractica.
Many businesses are exploring whether more advanced tools can help diagnose emotion‑triggered problems in customer service training or worker productivity. Practitioners believe they can reduce risk via closer tracking of employee behavior, lower employee burnout rates, improve productivity and boost morale.
To be sure, there’s a creep factor associated with emotion‑recognition tools. Some workers may find them invasive or not trust their employers to use the data responsibly. The practice could easily become coercive if used for critical HR decisions on individuals. Companies that jump into the field with less than transparent motives could face lawsuits and walkouts.
Picard acknowledges these hurdles, but believes that third‑party data banks could securely handle aggregated results for employers, and offer employees their own data as feedback when requested. If accomplished through trust and employee consent, such data could be seen as a valuable professional development benefit. “These tools help us get insight into the stressors that people don’t even themselves consciously acknowledge,” says Picard.
Boston‑based software company Cogito has built an app based on machine‑learning algorithms that listen in on customer‑service calls for Humana, MetLife and others. The goal is to identify stress in employee‑customer interactions. When agents grow frazzled after long hours of explaining insurance policies to customers, an upbeat voice tells the agent that her own stress is coming through on the call.
For the banking industry, U.K.‑based Behavox monitors account reps for tone and speed to help managers track employee performance. And emotion‑tech firm Affectiva recently announced AI tools for ride‑sharing services that analyze driver facial expressions for signs of stress and burnout.
Other companies are exploring new ways of understanding the subtle mechanisms of employee performance. Humanyze, a emotion analytics firm, asks employees to wear a small microphone‑fitted sensor that captures voice data and analyzes the tone and frequency of worker interactions throughout the day. (The system is strictly opt‑in, and employees who don’t wear the badge are given dummy look‑alikes.)
One client, a major U.S. bank, worked with Humanyze to solve a staff morale problem. One call center location had higher turnover rates and burnout levels than others, and bank managers couldn’t put their finger on the cause.
Humanyze sensors and software determined that call centers where employees took breaks together lowered stress and built cohesion. At the underperforming branch, managers had enforced a policy of staggered breaks, preventing service reps from socializing. A new break policy helped turn things around: Within months, worker productivity at the site rose 23%; employee retention rates increased 28%.
Turning off the rational brain
Researchers have always struggled to determine how subjects really feel, says Jeremy Pincus, a psychologist who co‑invented a tool called MindSight for marketing intelligence firm Isobar. Pincus designed the device to distract test subjects into providing more accurate emotional feedback.
The test flashes seemingly random, surprising images—a skydiver in flight, car driving off a cliff, a butterfly trapped in a cage—while asking employees to tap those images that represent how their jobs make them feel.
Pincus has gathered evidence over the years that shows how the speed of the images forces subjects to blurt out their true feelings without thinking. When volunteer employees at a major CPG company recently took the MindSight survey to comment on the company’s new mission statement, managers were surprised by the results.
“The statement said management wanted to give employees more ownership over their jobs,” says Pincus. “But this cuts both ways: It pushed workers toward higher achievement, but also made them more anxious. Subjects connected the added responsibilities associated with self‑direction to operating all alone without a safety net.”
This is precisely the kind of workplace stress that MIT’s Picard wants to detect and alleviate. Her latest venture, Empatica, makes a wristwatch that tracks stress through the skin. It also serves as an alert for people experiencing epileptic seizures. The devices have been used to test emotional states in research subjects at NASA, Sony and Microsoft.
If high levels of emotional intelligence are considered an important differentiator of successful leaders, tools like Picard’s suggest that a company’s collective EQ isn’t just measurable but could eventually become an important benchmark of long‑term success.