The Connection Between Machine Learning And CEO Leadership Skill
Studies indicate that CEOs spend 85 percent of their time in communication-related actions, including speeches, meetings, and telephone calls with people both inside and outside the firm.
Now, a new study utilizing machine learning is trying a deep dive into the words and facial expressions of chief executives to find out if leadership style could be related to a firm’s performance. The researchers believe their work may open new directions in big data evaluation, combining graphic and textual analysis to make a more complete picture of how a primary executive influences firm performance.
“Machine learning can use data that are both large in size, but also in a different form than that which would traditionally fit into an Excel spreadsheet,” says Harvard Business School’s Prithwiraj (Raj) Choudhury, the Lumry Family Associate Professor in the Technology and Operations Management Unit. “We are now able to operate on all these rich new sources of visual data.”
In a newspaper in the Strategic Management Journal, Machine Learning Approaches to Facial and Language Analysis: Finding CEO Oral Communication Styles (pdf), the writers use an assortment of techniques to group CEOs into different communication styles which, among other matters, seem to be correlated with the financial performance of their businesses.
The study was conducted by Choudhury and Tarun Khanna, the Jorge Paulo Lemann Professor in the Strategy Unit in HBS; Columbia Business School professor Dan Wang; and doctoral student Natalie Carlson.
The abilities that a CEO needs to control business are diverse, however, the ability to communicate effectively is close to the very top. “The task of being a leader in a business is to pull together different tools to achieve something effective,” says Khanna. “That means you are motivating people to do things, and the only way to motivate people is to communicate together.”
As crucial as communication is to running an organization, however, it’s hard to get an accurate read on differences in CEO communication style. The words that they select are crucial to meaning, naturally, but they also express themselves through tone and nonverbal cues like facial gestures. Further, both these verbal and nonverbal cues differ across cultures and geographies.
“You may say something positive, but a negative facial expression can create the opposite significance,” says Choudhury.
Locating CEOs to test
But where could the researchers find enough CEOs to conduct their analysis?
It ends up that Khanna was familiar with the response. He and Geoffrey Jones, the Isidor Straus Professor of Business History at HBS, happen to be compiling an oral history project named Creating Emerging Markets, 130 video interviews of iconic small business leaders from emerging markets talking their careers and associations in an unstructured format.
Those videos were analyzed using three machine-learning techniques.
First, the investigators looked at the words which CEOs chose, using statistical inference to eventually create 100 topics as diverse as advertising, company boards, and personal family history. Each CEO was scored based on their inclination to remain on a specific topic, compared to bouncing around from subject to subject, a step they called “topic entropy.”
The next machine-learning technique also looked at words, but this time split them by positive or negative valence, or just how much the speaker vacillated between negative and positive emotions.
In the end, the researchers analyzed the nonverbal communication of CEOs by analyzing their facial expressions utilizing a computer-vision program that rated them based on eight emotions: anger, contempt, disgust, fear, happiness, neutral, sadness, and surprise. This sort of investigation had typically been completed by individual coders, but the researchers found that computer analysis proved to be a robust, yet quick and cheap alternative to creating emotional data.
Can the CEO style correlate with performance?
With that information in hand, the researchers piled CEOs into one of five different communication styles. Those who showed anger, contempt, and disgust but also a fair amount of neutral expressions were labeled Stern. CEOs with high subject entropy and that seemed to utilize joyful and contemptuous facial expressed were known as Rambling. People that have a range of expressions were known as Dramatic, while those characterized by despair and negativity were termed Melancholy.
The various styles just appeared to emerge, state Khanna and Choudhury, when both text and facial features were examined together. “We were able to take these multiple dimensions and assemble those styles which were not possible previously,” Choudhury says.
The investigators then looked for correlations between communications fashion and other aspects of the enterprise. For example, people who have a greater Dramatic score revealed merger and acquisition activity in the year following the meeting. Also, the communication styles that emerged from collectively utilizing text and image data had more statistical power in explaining variability in the data, as compared to using text or image data independently.
Choudhury and Khanna stress those results are only illustrative. Their primary purpose in writing the newspaper, the state, is a proof of concept in opening up the concept of what type of data machine learning can effectively analyze.
“There are many new kinds of data on the market, and with calculating power going up stratospherically in the last twenty years, there are currently many opportunities for greater assessing many of the things going on in the business,” Khanna says. YouTube, for instance, could provide a rich trove of data to analyze CEO communication from addresses.
Additional uses for machine learning research
These additional information sources are especially helpful in studying emerging markets where traditional information sources might be less easily available and researchers’ understanding of these markets is more rudimentary. Also, many of these geographies are abundant in technology-based sources of data that can be studied, like the cell phone and Internet of Things explosions across many fast-growing Asian markets.
Beyond academic queries, business analysts may use these machine-learning methods, which the researcher’s state are comparatively simple and easy-to-use, to examine voice patterns in earnings calls or nonverbal communication at shareholder meetings to include thickness to their evaluation of business performance.
“There is an entire ocean of information out there that people aren’t using,” says Khanna. “Using it could help examine many questions that are essential to business.”