Conversation Analytics: Can Machines Read between the Lines in Real-Time Strategic Conversations?

Abstract

Strategic conversations involve one party with an informational advantage and the other with an interest in the information. This paper proposes machine-learning based measures to quantify the degrees of evasiveness and incoherence of the informed party during real-time strategic conversations. The specific empirical context is the questions and answers (Q&A) part of earnings conference calls during which managers endure high pressure as they face analysts’ scrutinizing questions. Being reluctant to disclose adverse information, managers may resort to evasive answers and sometimes respond less coherently due to increased cognitive load. Using data from the earnings calls of the S&P 500 companies from 2006 to 2018, we show that the proposed measures predict worse next-quarter earnings. Moreover, the stock market perceives incoherence as a negative signal. This paper contributes methodologically by developing two novel machine-powered measures to automatically evaluate behavioral cues during real-time strategic conversations. The proposed analytical tools are particularly beneficial to resource-constrained and informationally disadvantaged parties such as retail investors who may not be able to effectively trade on signals buried deep in unstructured conversational data.