Some corporations have always cared how their employees feel—if only because happier workers are more productive than those who are miserable. Others have only recently begun to wake up to the fact that they need to address wellbeing in meaningful ways. This focus raises a question: How can a company tell whether the people who work there are happy?
A small Toronto-based company called Receptiviti is suggesting a tech solution. Unlike more traditional methods, like employee surveys, its method hinges entirely on analyzing the language used in employees’ everyday workplace communications, be that emails, Slack messages, or even voice. But what makes Receptiviti’s method interesting is that while it uses natural language processing, a branch of machine learning, to analyze language, it’s not sifting communications for sentiment. Rather, the company employs a branch of linguistic research that suggests passive parts of speech—the bits we use without thinking, like prepositions and pronouns—hold the key to how happy we are. Rather than looking at what people are saying to one another, it examines how they’re saying it.
But wait: The idea of one’s employer, or an outside service, monitoring all communications sounds terrifyingly invasive. Jonathan Kreindler, Receptiviti’s co-founder and CEO, is quick to note that the company doesn’t analyze results on an individual basis, but rather according to traits and teams. They might “cut” the data based on gender and department, for example, revealing that women on a company’s finance team are less happy than the men there. Or they might discover low morale among people who have been at a firm for more than three years, and work in North America. He also said that Receptiviti only works with companies that have active consent from all their employees to monitor the communications.
So how does the method actually work? In a 2013 TEDx talk, James W. Pennebaker, a social psychologist at the University of Texas at Austin and a co-founder of Receptiviti, explained that less than half the words we speak or write are actively designed to carry meaning. In the sentence “I’m very frustrated,” for example, the final word has been chosen to convey a specific sentiment. But, Pennebaker explains, the rest of speech, which native speakers use almost unconsciously, holds other subtle clues. People who are less happy tend to use personal pronouns (“I” and “me”) more than average, and words relating to other people (like he or she) less, he says. The sentence about frustration now looks a little different: The same sentiment could easily be communicated without reference to the speaker (“This is very frustrating.”)
Pennebaker made the discovery about the power of pronouns and similarly quiet clues when analyzing essays written as part of a study of people who had experienced trauma. In comparing the words used in the essays to other markers of a subject’s health, he found no meaningful difference when looking at the content. But when writing the computer program that analyzed the texts, he’d included analysis of the roughly 500 “function words” in the English language. Looking at that data led to a revelation.
The subjects’ mental states were reflected in their use of prepositions (to, the), pronouns (I, he, she), and auxiliary verbs (am, is.) In addition to using personal pronouns more than third-person pronouns (like he or she), they also tended to talk about the present much more than the future. These surrounding and supporting words are “profoundly social” Pennebaker explains. Analyzing them can reveal how a person is feeling, how they’re relating to the world around them, and what they think of themselves and others.
Pennebaker tried out his technique on a whole range of communication types, discovering that when groups or pairs of people are getting along well, they begin to mirror one another’s language. The use of pronouns and other function words can indicate what status a person feels themselves to be, for example in relation to someone they are emailing. When his team analyzed speed-date transcripts, Pennebaker says in the talk, they found they could predict whether the interlocutors would go on a second date with greater accuracy than could the people themselves.
The science behind Receptiviti’s approach isn’t brand new, but applying it to wellness at work is a development—one of many that are transforming the business of figuring out how employees feel. Isaak, a tool launched this year by the workplace analytics company StatusToday, promises to identify employees or teams that might be at risk of burnout by tracking metrics like email volume and responsiveness.
But while these kinds of services might gradually replace more traditional surveys, they’re not without issues. As Carolyn Axtell, senior lecturer in work psychology at the University of Sheffield in the UK, writes in The Conversation, using big data to provide insights into a workforce requires responsible handling, to make sure employees know what they’re signing up for, and so that they don’t feel they’re under surveillance.
But the more traditional strategies employers use to keep abreast of employee wellbeing also are subject to problems. Performance reviews are meant to channel concerns from workers to management, and part of the line managers’ responsibility is to support their reports. But those relationships can be complicated: managers also are responsible for those same workers’ productivity, and not every manager is talented at pastoral care. Even anonymous surveys, often administered by survey companies, have limitations. The surveys themselves need to have been designed well, probing in the right places to create an accurate picture of an employee’s feelings. Employees, meanwhile, need to take the time to fill the surveys in, they need to be honest, and they need to understand what they themselves are feeling in order accurately to express it. Then the data has to be sensitively interpreted. And all of this happens only at intervals, providing a temperature check that relates to a certain point in time.
Now, both more traditional survey companies and their tech-focused rivals are trying to solve an issue that’s becoming more pressing to businesses. How do you detox a culture? How do you find there’s a problem and head it off, before being derailed by the kind of PR crises faced by the likes of Uber and Amazon? Large corporations had a “very interesting year,” in 2017—the year #MeToo broke—says Kreindler. “Whether you are…a technology company that’s been around for five years and grown apace, or whether you’re a large bank that’s been around for over a century, these are issues that you deal with,” he said. “And understanding where these things are happening before they manifest into big problems means that you’re responding to the needs of your employees, rather than waiting for these things to blow up.”