The language of leaving: Decoding defection via B2B customer language features

Abstract Customer language data (e.g., online reviews, call transcripts) is widely available. Yet, research in B2B contexts is still in its early stages of analyzing a full range of customer language features to derive insight. Using an empirics-first approach, this research sequentially identifies robust and parsimonious language features in B2B customer data that are diagnostic of customer experiences and predictive of forthcoming purchasing decisions. The authors examine five categories of how business customers speak: sentiment, temporality, certainty, concreteness, and personal pronouns. Results reveal that sentiment and temporality consistently reflect customer experience and provide incremental predictive value of customer defections within request-for-proposal (RFP) decisions. The authors demonstrate this pattern across customers’ written (Study 1), spoken (Study 2), and pre-RFP (Study 3) language. These studies leverage empirical regularities to advance early-stage theorizing around psychological distance and language-based models of customer defection. Illustrating the managerial relevance of customer language features, findings show that in a customer’s collective communication prior to an RFP decision, a 1% increase in past focus raises the odds of a multimillion-dollar RFP defection by 8.4%. The research also equips practitioners to compute sentiment and temporality in customer language via large-language-model prompts and statistical code, facilitating resource allocation and prioritization of customer accounts.