2022 - Working Papers: Technology and Information Systems
Online news consumption, subscription and churn, 34 pp.
S. Bar-Gill
(Working Paper No. 6/2022)
Research No. 06022100
Churn is one of the main challenges for subscriber-based business models, specifically when conversion rates (the rate at which users are converted to paying subscribers) are low. Prior work shows that churn is associated with consumption patterns, yet the role of subscription utilization over time and its relevance to consumers’ decisions to unsubscribe from digital media services has not yet been explored. Using observational data from a U.S based news website, I investigate the dynamics of decreasing utilization of a paywall subscription leading up to churn, further evaluating the predictive power of different subscription utilization states on churn. Utilization is defined with respect to a metered paywall for news articles, which sets a threshold number of articles that can be accessed free of charge, while above threshold reading requires a paid subscription. Above threshold reading represents subscription utilization, whereas below threshold reading by subscribers implies underutilization of the subscription. The results show that utilization levels are informative for predicting churn, offering significant improvements over the predictive power of user attributes, and general consumption patterns. This finding is consistent across prediction methods (Logistic Regression, Classification Tree, and Random Forest).
Helping small businesses become more data-driven: A field experiment on eBay, 55 pp.
S. Bar-Gill, E. Brynjolfsson and N. Hak
(Working Paper No. 7/2022)
Research No. 06090100
As digitization sweeps across industries and markets, analytics and data-driven decision-making (DDD) are becoming increasingly important. The adoption of analytics and DDD has been slower in small-to-medium enterprises (SMEs) compared to large firms, and reliable causal estimates of the impacts of analytics tools for small businesses have been lacking. We derive experiment-based estimates of the impact of an analytics tool on SME outcomes, analyzing the randomized introduction of eBay’s Seller Hub (SH), a data-rich seller dashboard. We find that SH adoption is associated with increased DDD and find that access to SH increases e-retailers’ sales by 3.6% on average, as more items are transacted, more listings created, and service quality increases, without increases in average prices. Managerial practices and skills play an important role in reaping the benefits from the analytics dashboard, as over a third of the SH impact is driven by active performance monitoring, and a strong positive association exists between adopters’ aptitude for analytics and increases in DDD. Policies to support small businesses’ transition to the data era should therefore address analytics and DDD gaps by both ensuring access to tools like SH and improving managerial practices and analytical skills.
Predicting user performance using platform-owned discussion boards: Evidence from eBay’s seller hub forum, 36 pp.
S. Bar-Gill
(Working Paper No. 8/2022)
Research No. 06020100
NO ABSTRACT
The guide to content moderation: Introducing crowds to mitigate the challenges of the human moderator, 9 pp.
L. Zalmanson, I. Yahav and D. Yadin
(Working Paper No. 14/2022)
Research No.: 06322100
The increased reliance on user-generated content in the business models of online environments brings with it a challenge of setting up rules for content moderation. Namely, how should the firm create a fair and dynamic set of moderation guidelines that will be relevant and explainable. At present, firms are commonly setting the rules as they go and are in need to continuously assess, examine, and update with the help of content experts. Our work seeks to offer a different approach to generating content moderation rules in an economic, dynamic, and objective fashion. Specifically, we propose the use of crowdsourcing in symbiosis with NLP algorithms to compile and revise content moderation guidelines semi-automatically. We discuss how our proposed approach can be integrated in current moderation processes in practice.
