“Diversity means lots of things,” says Amir Goldberg, an associate professor of organizational behavior at Stanford Graduate School of Business. “These days, it evokes the idea of race or gender, but it’s also about how people think.“
Beyond their demographic differences, people working in a group will likely think differently about a collaborative task. That cognitive diversity can be helpful — or not. “In an organization, there’s tension between people who have incongruent ways of thinking about a specific problem to solve,” Goldberg says.
It’s widely thought that brainstorming how to create a better app or product requires a range of ideas and perspectives, while getting down to the business of execution is best handled with greater alignment around how to proceed. “The assumption,” Goldberg explains, “is that intellectual diversity is good for creating novelty and creative problem-solving, but not necessarily good for efficient coordination.”
On the surface, it may seem like a team will be good at only one of those things, depending on its level of intellectual diversity. “But that’s a blatant simplification,” Goldberg says, “because teams can also modulate” how they apply their variety or consistency of thought to a given task.
To better understand this dynamic, Goldberg and collaborators Melissa Valentineopen in new window and Katharina Lixopen in new window of Stanford and Sameer Srivastavaopen in new window of Berkeley Haas studied hundreds of thousands of Slack messages sent by software development teams working remotely. Using computational linguistics tools, the researchers measured how team members’ ideas diverged and converged over time.
They found that more successful teams modulated their cognitive diversity to fit the task at hand. “The diversity level doesn’t jump around on these teams but undulates,” Goldberg says. “It changes by phase of the software project. Teams that become cognitively divergent for ideation but more convergent for coordination are the ones most successful in delivering their projects on time and to the satisfaction of the customer.”
Dialing divergence up or down
To get this data, the researchers worked with Gigster, an online platform where freelancers from around the world collaborate on tech projects. “Some of the projects are well into the hundreds of thousands of dollars,” Goldberg says. “And many customers are leading S&P 500 companies. So the stakes of team interactions are high.”
The researchers looked at Gigster teams’ interactions on Slack, using an AI-based algorithmic tool to analyze anonymized conversations. The study evaluated more than 800,000 messages generated by more than 400 people across 117 teams.
Diversity of thought was detected through a word-embedding model that predicted words that tended to go together based on context. The model was trained on the language that coders, designers, and project managers are likely to use, such as “bug” to refer to a coding error rather than an insect.
“Eventually, every word gets represented in a multidimensional space related to the underlying meanings of people’s language,” Goldberg explains. The researchers then modeled each Slack conversation in terms of how aligned speakers were — based on the distance between the core types of words they used — to generate a semantics-based linguistic reflection of a team’s cognitive diversity.
Goldberg and his colleagues associated this measure of cognitive diversity with the likelihood a team would reach development milestones on time — the best measure of success. “If the customer is unhappy with the product delivered, it goes back,” he says. “So it’s not just about timely delivery, but timely, successful delivery.
A team’s average level of cognitive diversity over time was not predictive of its performance. Yet higher diversity in the ideation stage and lower diversity in the coordination stage were associated with more timely — and successful — project delivery.
In contrast, teams with more cognitive divergence during periods of coordination or lower divergence during periods of ideation had lower success rates. In the early coordination stage, teams with cognitive diversity levels one standard deviation above the mean had a lower predicted probability of success (37%) compared to teams with diversity levels one standard deviation below the mean (46%).
The results have meaningful implications for all kinds of teams. Yet Goldberg cautions anyone eager to “monetize” cognitive diversity through, say, a dashboard monitoring team members’ convergence and divergence in real time: “These people-analytics tools are more useful as self-empowering tools, as opposed to supervisory tools.”
Cognitive diversity is “just one thing that we’re measuring imperfectly that affects the performance of teams,” Goldberg says. “We’re still a long way. If people were to think, ‘OK, we don’t need team leaders now that we have this,’ we’re so not there yet.”
Indeed, the person overseeing a project team plays a critical role in modulating its cognitive diversity for success. “If the team leader is divergent or convergent at the right times, it’s important,” Goldberg says.
It may make sense for leaders to run thought experiments to promote a range of ideas in the ideation stage and then corral the cognitive diversity by seeking common ground once the team transitions to coordination and delivery. “It’s something we need to study further,” Goldberg says, “to understand what leaders do and don’t do to help with this.”
“In the end,” he concludes, “something like this not meant to replace human judgment but to augment it. Team coordination is a problem orders of magnitude more complex than parking in reverse.”