Essays in Platform Analytics and Customer Relationship Management

Date of Award

Spring 1-1-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Management

First Advisor

Sudhir, K.

Abstract

This dissertation examines digital platforms and customer relationship management (CRM) across three essays. In the first two essays, I investigate how the design of rating systems on platforms can inadvertently create and perpetuate biases, ultimately affecting quality and customer experience. In the third essay, I turn to a CRM question in marketing: whether integrating cross-selling with customer service enhances or undermines customer satisfaction and firm value. By employing a range of methodological approaches, including structural modeling and natural experiments, this dissertation provides insights into how design choices in various settings can generate sizable impacts on consumer perceptions, worker outcomes, and overall platform and firm performance. In the first essay, which is the joint work with K. Sudhir and Tristan Botelho, I examine how rater heterogeneity on platforms -- whether they are generous or harsh -- can have long-term effects on platform fairness and quality. Customer reviews and ratings help consumers make informed decisions by providing quality information and reducing uncertainty. However, in this paper, I demonstrate that ratings are not only influenced by the quality of the suppliers but also by the rating scales of the customers -- whether the customers are generous or harsh. Furthermore, I introduce a novel idea: the effect of the heterogeneity in the customer rating scale does not simply balance out as suppliers accumulate more ratings from a larger pool of customers. Instead, early differences in the mix of generous and harsh raters can result in long-term biased ratings and unfair outcomes. This occurs because platforms display past ratings to customers, whose ratings are subsequently influenced by this information. Additionally, platforms use these past ratings for their prioritization and recommendations, leading to path dependence. Using data from a gig economy platform, I estimate a structural model and find that the early rater mix significantly impacts future ratings, leading to persistent advantages in ratings and earnings for lucky workers who encounter more generous customers early on. To address this, I propose a neutral adjusted rating metric that can mitigate these effects. Counterfactuals show that using the metric enhances the accuracy of rating systems for customers, fairness in earnings for workers, and quality of supplier selection for the platform. In the second essay, jointly with K. Sudhir and Tristan Botelho, I investigate the conjecture that rating systems can lead to discriminatory spillovers and become ''discrimination amplifiers.'' Because rating systems memorialize differences in individual ratings (impacted by group-based statistical or taste-based discrimination) by serving as quality anchors for new customers, displaying average ratings on a platform can lead to discrimination spillovers for customers who do not discriminate and amplify discrimination for those who do, leading to greater inequity. After demonstrating the idea using a stylized analytical model, I investigate the question of discrimination amplification empirically using data from an online labor market platform that connects service workers with customer jobs. Using a model of customer's job cancellations and rating choice and allowing for unobserved heterogeneity in discriminatory behaviors, I identify three segments: one shows no difference in behavior towards minority and White workers; the second cancels minority workers at higher rates, hurting minority earnings; a third cancels minorities more and rates them lower than Whites. I find that customer discrimination leads to lower ratings and earnings for minority workers. Displaying ratings amplifies discrimination and increases the minority ratings and earnings gaps. In the third essay, jointly with K. Sudhir and Guofang Huang, I investigate whether firms should integrate cross-selling with customer service. Cross-selling, conducted after a service interaction, has been widely adopted across various industries involving customer service. However, there is little empirical research on how this combination affects customer satisfaction or whether cross-selling and customer service functions create synergies. The empirical tradeoff is as follows: indiscriminate cross-selling after service may ''irritate'' customers, lower service satisfaction, and ultimately diminish customer value. On the other hand, if service agents can effectively use information gathered during the service interaction to assess a customer’s likelihood of accepting a cross-sell offer and target only those who are more receptive, integrating cross-selling with customer service can create additional value. However, disentangling these effects is challenging, as the decision to cross-sell is endogenous to customer satisfaction with the service interaction -- an aspect that is unobservable. In this paper, I develop an empirical approach that leverages detailed internal administrative data and a natural experiment in a large financial services company to address these research questions. I find that extending cross-sell offers negatively impacts customer satisfaction. However, service agents mitigate this negative effect by selectively offering cross-sells to customers who are more likely to be receptive, using insights gained during the service interaction. This suggests that agents leverage information learned during service to optimize cross-selling while minimizing customer irritation.

This document is currently not available here.

Share

COinS