What CEOs Don't Say

Emerging technologies could be used to detect contradictions between business leaders’ verbal and nonverbal expressions.

What CEOs Don't Say

ARTIFICIAL INTELLIGENCE AND and machine learning could offer valuable insights into the link between a CEO’s communication style and an organization’s performance. Researchers examining the implications of this possibility include Prithwiraj Choudhury, the Lumry Family Associate Professor of Business Administration, and Tarun Khanna, the Jorge Paulo Lemann Professor, both of Harvard Business School (HBS) in Boston, Massachusetts. Their co-authors include Dan Wang, the David W. Zalaznick Associate Professor of Business, and doctoral student Natalie Carlson, both of Columbia Business School in New York City.

The team used machine learning to analyze 130 video interviews with CEOs. The videos were part of Creating Emerging Markets, an oral history project conducted at HBS by Khanna and Geoffrey Jones, the Isidor Straus Professor of Business History. The team examined the vocabulary and speech patterns of the CEOs, from the number of times they used certain words to the extent to which they moved from subject to subject (what the researchers call “topic entropy”). Next, the researchers analyzed the emotional content of each speaker’s words— in particular, how often he or she expressed positive and negative emotions. Finally, they rated each speaker’s facial expressions across eight emotions: anger, contempt, disgust, fear, happiness, neutrality, sadness, and surprise.

Because machine-learning technology allowed the researchers to examine words, tone, and facial expressions simultaneously, they could better detect what the CEOs didn’t say. For instance, the technology revealed contradictions between verbal and nonverbal expression. As Choudhury explains in an article in HBS Working Knowledge, “You might say something positive, but a negative facial expression may create the opposite meaning.”

Using this data, the researchers identified five communication styles among these leaders. These include positive-leaning styles such as excitable and dramatic, and negative-leaning styles such as stern, rambling, and melancholy. The team then looked for patterns in performance at each leader’s organization. For example, the higher a leader’s dramatic score, the fewer mergers and acquisitions took place at that person’s company in the year following the interview.

“With computing power going up stratospherically in the last 20 years, there are now many opportunities for better analyzing many of the things going on in business,” says Khanna in HBS Working Knowledge. For example, a leader’s voice patterns during calls with shareholders could be analyzed for clues about the firm’s performance. “There is a whole ocean of data out there,” Khanna adds, “that people aren’t using.”

Read “Machine Learning Approaches to Facial and Text Analysis: Discovering CEO Oral Communication Styles.”  The paper is forthcoming in Strategic Management Journal.