Marketing Research Seminar Series

Uncertainty around content monetization poses a persistent challenge for independent creators, highlighting the importance to understand and manage reward uncertainty in content creation. This paper addresses the challenge and studies how monetary reward and reward fluctuations (risk) affect creators' content choices. We build and estimate a model that features the ''learning of learning'': to resolve uncertainty in reward, creators learn about their reward evolvement through their past creation and consumption experience, and utilize the expected reward and reward variance to make content choices. In the empirical context of a live streaming platform, we quantify the impacts of risk and find that, at the mean level, risk brings disutility that amounts to about 1/3 of the average reward. Managerially, we propose an income smoothing policy in a hope to help creators cope with risk and encourage content creation. Our counterfactual simulations reveal that, by collecting a moderate level of reward from each creation and fully subsidizing creators when their reward levels are low, the platform could achieve additional revenue gain while promoting content creation. Interestingly, creators not only produce more under the policy, but will tap into content categories that are less popular, thereby enhancing content diversity on the platform.

Promotions for digital goods, such as free samples, have typically focused on enticing users to accelerate their consumption. Here, we investigate the implications of a novel consumption-decelerating type of promotion, ''Wait For Free'' (WFF), applied to serialized digital content sequences of interconnected episodes monetized via episode-level paywalls. Specifically, customers can sample early episodes of promoted series for free, and can continue to do so by waiting a pre-specified time or immediately by paying. Because it defies one of the traditional roles of promotions consumption acceleration such a policy may appear counterproductive. However, when applied to serialized content, some users may get ''hooked'' into the plotline, and those who are unwilling to wait may elect to pay in order to progress more quickly. We analyze episode-level viewership data from an international digital comics platform while accounting for the time trend and observed heterogeneity in promotional lifts, as well as the impact of promoting a series on others by constructing dyadic behavioral similarity measures between series. Using a combinatorial Genetic Algorithm, we efficiently search through potential promotion sets, finding that this seemingly counterintuitive type of promotion can in fact boost paid viewership at the platform level, net of cannibalization on unpromoted comics, when applied to an appropriate set of comics. We also discover that the genres and overall popularity within the solution set change considerably depending on the planning horizon. Finally, to understand the individual user-level impact of WFF, we estimate a Cox proportional hazards model to compare pre-and post-promotion the degree of within-series inertia in content consumption and across-series switching behavior, finding evidence of trial effects among users and conversion of some free users into paying ones.

I study how rating system design affects innovation incentives. In settings in which product quality cannot be observed prior to purchase, online ratings serve as a signal of product quality for consumers and affect demand. Owing to their impact on sales, ratings also motivate firms to innovate. If firms use displayed ratings to guide their investments in improving product quality, then platform rating aggregation policies can play a key role in increasing or decreasing firms’ innovation incentives. I study the impact of online rating systems on innovation incentives and, more importantly, the implications of the design of the rating aggregation policy. After collecting a unique firm-level dataset from a mobile game app platform, I combined reduced-form analysis and a structural model to show how rating systems can be optimized for innovation. I show that innovation has a positive impact on all key rating system metrics and that a lower rating significantly increases innovation incentives. Building on empirical evidence, I develop a dynamic structural model to represent firms' forward-looking behavior and estimate innovation cost. I then evaluate the impact of alternative rating aggregation policies on innovation incentives. The counterfactual analysis shows that placing greater weight on recent ratings can increase the innovation rate substantially.

We study the deployment of an algorithmic pricing tool, Smart Pricing (SP), on Airbnb's platform. SP is a machine learning algorithm that uses past data to predict demand and employs proxies that are correlated with the host's marginal cost to set prices for listings. The success of such deployments depends on how good the performance of the algorithm is and how sellers use the tool for their business decisions. Our analyses suggest that adopting SP is associated with higher benefits for hosts who rarely change their prices compared to those who flexibly adjust their prices before adoption. However, hosts who rarely change their prices are surprisingly less likely to adopt SP. To understand how the platform can overcome this challenge, we propose and estimate a dynamic structural model in which hosts make adoption decisions based on their expectations of the algorithm's behavior. Our estimation results identify a gap between the actual performance of the SP algorithm and the host's prior belief about it. Specifically, hosts with a pessimistic prior belief about SP think they will need to manually correct algorithmic prices if they adopt SP, and this belief is disproportionately stronger for hosts with higher adjustment costs, making the SP adoption less attractive to them. Our counterfactual simulations indicate that the introduction of SP has had a small positive impact on the average host profit and the total platform revenue. But this boost can be significantly raised if Airbnb helps hosts to correct their beliefs about the SP algorithm. This highlights the need for proper communication of how the algorithm works and its benefits in order to successfully deploy a machine learning tool. The counterfactual analyses also demonstrate that, since the platform does not fully capture the host's private marginal cost in training the algorithm, using the estimated costs from the structural estimation to re-train the algorithm can significantly increase the profitability of SP for both hosts and the platform. It suggests that combining the results of structural models and machine learning tools can help platforms design better algorithms.

How does the way companies elicit ratings from consumers affect the ratings that they receive? In 11 pre-registered experiments, we find that consumers rate subpar experiences more positively overall when they are also asked to rate specific aspects of those experiences (e.g., a restaurant's food, service, and ambiance). Studies 1-4 established the basic effect across different scenarios and experiences. Study 5 found that the effect is limited to being asked to rate specific features of an experience, rather than providing open-ended comments about those features. Study 6 found that the effect holds even when consumers are told that they are going to rate specific aspects on a subsequent page. Studies 7-10 provided evidence that the effect does not emerge because rating positive aspects of a subpar experience reminds consumers that their experiences had some good features. Rather, it emerges because consumers give less weight to negative aspects of an experience in their overall evaluation when they are invited to directly rate those aspects. Lastly, study 11 found that asking consumers to rate attributes of a subpar experience reduces the predictive validity of their overall rating. We discuss implications of this work and reconcile it with conflicting findings in the literature.

We explore an unintended consequence of using personalized recommendations, that is, recommendations that are targeted to an individual consumer (e.g., personalized music playlists). We conceptualize that personalized recommender systems are seen as having the ability to replace the self. Therefore, using these systems can decrease people's preference clarity, which is defined as certainty about their own preferences. For example, people may feel less certain about their own music preferences after listening to auto-generated personalized playlists. This reduced preference clarity, in turn, reduces consumer willingness to generate word-of-mouth (WOM) about their consumption experiences, such as their intent to talk about music they listened to with others, or to post social media content on their favorite musicians. Six studies, using correlational and experimental designs and conducted with consumers who actively use personalization services (in the fashion and music domains), support this theorization. We find that listening to one's own (vs. a matched other's) Spotify-generated music playlist increases satisfaction but decreases WOM due to changes in preference clarity. We end with a discussion of the potential theoretical extensions of this novel finding, as well as its practical implications.

Advertising effects can have substantial heterogeneity, and identifying the moderating factors is important. We study how own and rival retail advertising effects depend on consumer-store spatial relationships and ad content (promotion vs. non-promotion ads). We integrate TV ad viewership data with new large-scale microdata on mobile locations, which track consumer traffic to home improvement stores and their home census block groups. We find that customers' relative proximity to own store compared to rival store moderates their ad effects. This moderator is more important than customers'''' absolute proximity and suggests fierce geo-competition. The ad effect heterogeneity comes from differences in both visit frequency and ad responsiveness. While non-promotion ads expand demand, promotion ads steal rival traffic, especially among customers in proximity to both stores. The results are primarily driven by non-professional customers and not contractors. Our findings have implications for retail advertising targeting strategy and the equilibrium impact of improved targeting technology.

We study the effectiveness of influencer marketing in the video game industry. We construct a novel dataset on video game streaming by continuously monitoring Twitch.tv, the largest gaming platform in the world, once every 10 minutes. The high-frequency nature of the data allows us to leverage plausibly exogenous variation in within-day streaming schedules to estimate the effect of stream viewership on video game usage. We find a modest and short-lived elasticity of 0.033 of primarily organic streams, and that sponsored content is only 25% as effective—an elasticity of 0.007. Finally, using generalized random forests to examine heterogeneous returns to streaming, we find that Twitch influencers primarily benefit new games, games from small publishers, and games with mixed user reviews. These results suggest that, whereas streaming does not unequivocally benefit all firms, this new, low-cost marketing channel does help promote products that would otherwise face significant information frictions.

While prior research strongly suggests that performance-based, variable compensation improves sales performance, several companies have recently reduced the amount of variable compensation because they blame it for causing myopic selling behavior. However, the effects of such incentive reductions are not well understood. Therefore, we examine an industrial wholesaler that reduced the share of variable compensation for its sales force in one sales division (from 80% to 20%) while keeping it unchanged in another division. We differentiate between effects of the incentive reduction on salespeople's performance quantity and quality. While quantity reflects salespeople's effectiveness in terms of sales output, quality reflects how well salespeople advise their customers. Using difference-in-differences models, we show that salespeople's performance quantity declined after the incentive reduction while performance quality increased. The strength of these effects depends on the prior performance in the two performance facets: reducing incentives especially harms the performance quantity of salespeople with high prior performance quantity, while performance quality increases particularly for salespeople with low prior performance quality. Thus, managers contemplating the reduction of variable compensation need to carefully balance the trade-off between performance quantity and performance quality, given their sales force’s current performance.