Measuring Seller Response to Buyer-initiated Disintermediation: Evidence from a Field Experiment on a Service Platform

Abstract

Online platform businesses have thrived and created enormous economic value in recent decades. However, disintermediation, which occurs when platform users bypass the platform to complete transactions directly, has become the Achilles’ heel of the platform business model, hurting platforms' revenues and even survival in the long run. Despite being a prevalent issue, quantifying the extent of disintermediation has been a challenge as it is unobserved in most secondary data, so researchers have had to formulate approaches to infer it. In this study, we overcome this measurement challenge by designing a randomized field experiment to investigate sellers' response to buyer-initiated disintermediation on a service platform. Independent contractors were hired as undercover buyers to reach out to sellers under different experimental conditions. Specifically, the experiment varies (1) if the communication could be monitored by the platform and (2) the magnitude of commission savings from disintermediation. Our results show that on average, 51.7% of sellers across the four experimental conditions were willing to disintermediate. Even when communications could be monitored by the platform, the number was over 40% and communications being unmonitored significantly increased the level of seller willingness by up to 21.3 percentage points. Meanwhile, the average treatment effect of higher commission savings is statistically insignificant. Next, we apply the multi-arm causal forest algorithm to estimate the conditional average treatment effects (CATE) for each individual observation. The CATE estimates reveal salient treatment heterogeneity across sellers: Communications being unmonitored ubiquitously induces higher levels of disintermediation for all sellers, albeit with different magnitudes. However, high commission savings lead to opposing effects across individuals, a finding that helps reconcile the insignificant average treatment effects. Finally, second-stage analyses reveal a U-shaped effect of sellers' platform loyalty tier status on CATEs. In addition, sellers with shorter tenure and higher recent sales are more motivated to disintermediate while being unmonitored or facing higher commission savings. Our results provide important managerial implications for online platforms on whom to target for more effective disintermediation management.

Keywords: Platform design, disintermediation, randomized field experiment, causal forest

JEL Classification: C93, M20, M30