My research investigates how tech businesses alter society by engaging with their consumers in online and mobile spaces. In the first chapter, I examine the effect of a user deterrence policy - a policy that adds friction to information sharing and promotion to curb malicious activities of abnormal users - on regular users in an online platform. According to a quasi-experiment on an online news site, the deterrence policy increases regular users’ feedback activities by 0.11 percent by upvoting or downvoting others’ comments. To examine the underlying rationale for effect, I use the rational voting decisions framework and find that the deterrent policy enhances the subjective likelihood of affecting the popularity of comments among regular users by casting a decisive vote, due to the expected drop in abnormal user activity. Additionally, I observe that the deterrence policy strongly affects active users and political extremes. As disinformation poses a substantial threat to the sustainability of online platforms, this study provides valuable insights for businesses and policymakers as they assess the consequences of abnormal user deterrence policies. In the second chapter, I develop a conceptual model that shows how the efficacy of artificial intelligence positively affects firm performance, which is mediated by improved distributive equity. Using a dataset of 45 million users from 35 countries across 5 continents collected from a global artificial intelligence-powered education app, I find strong support for distributive equity as a mediator between the perceived efficacy of artificial intelligence and firm performance. I also find that the mediating effect of distributive equity is more pronounced in severe environmental conditions regarding political regimes, economic development, socio-cultural aspect, and technological resources, hence aiding in the actualization of artificial intelligence in such contexts. This research contributes to the growing popularity of the blended value proposition, such as the simultaneous pursuit of financial and social interests, by expanding a business model from a bottom line to a double bottom line approach. I discuss how this study extends the literature and provides managerial implications of artificial intelligence firms for global expansion, recruitment, and an inclusive information system.

In two independent and thematically relevant chapters, I empirically investigate consumers’ mobile channel usage behaviors. In the first chapter, I examine the impact of mobile use in online higher education. With the prevalence of affordable mobile devices, higher education institutions anticipate that learning facilitated through mobile access can make education more accessible and effective, while some critics of mobile learning worry about the efficacy of small screens and possible distraction factors. I analyze individual-level data from Massive Open Online Courses. To resolve self-selection issues in mobile use, I exploit changes in the number of mobile-friendly, short video lectures in one course (“non-focal course”) as an instrumental variable for a learner’s mobile intensity in the other course (“focal course”), and vice versa, among learners who have taken both courses during the same semester. Results indicate that high mobile intensity impedes, or at most does not improve course engagement due mainly to mobile distractions from doing activities unrelated to learning. Finally, I discuss practical implications for researchers and higher education institutions to improve the effectiveness of mobile learning. In the second chapter, I investigate the impact of mobile users’ popular app adoption on their app usage behaviors. The adoption of popular apps can serve as a barrier to the use of other apps given popular apps’ addictive nature and users’ limited time resources, while it can stimulate the exploration of other apps by inspiring interest in experimentation with similar technologies. I use individual-level app usage data and develop a joint model of the number of apps used and app usage duration. Results indicate that popular app adoption stimulates users to explore new apps at app stores and allocate more time to them such that it increases both the number of apps used and app usage duration for apps excluding the popular app. Such positive spillover effects are heterogeneous across app categories and user characteristics. I draw insights for app developers, app platforms, and media planners by determining which new apps to release in line with the launch of popular apps, when to release such apps, and to whom distribution should be targeted.

Customer lifetime value has been a popular topic within the marketing field with which many researchers and marketing managers have been dwelling. The topic plays an important role in customer segmentation and has been studied and applied in a variety of business areas. The main objective of customer lifetime value and customer segmentation is to classify the importance level of each client to a company and compare it to other clients. Questions, such as which marketing strategies should be implemented for which customers and how much should be invested in a certain group of customers can all be answered by customer lifetime value and customer segmentation. However, the related literature is missing comparative research on assessing the amount of time from the initial point of acquisition that a client needs to be with the company in order to accurately predict its customer segment. This paper intends to provide a clarification to the problem. Purpose: Analyze customer profitability with clustering analysis and identify how many years does a customer need to be with a company to accurately predict its customer segment. By determining this number, managers can understand their clients better and establish which clients will most likely yield a greater profit at an early stage of the relationship. Methodology: Using data mining to clean and prepare our financial services dataset, we selected young clients who were less than ten years old. Linear regression and K-means clustering analyses then returned five clusters of clients. Next, we predicted the accuracy levels of customers with two to seven data points against the "correct" segment. Lastly, we validated our overall prediction accuracy levels using the chance probability and the desired classification accuracy, calculated from a discriminant analysis. Findings: We found that using five data points or more to cluster returned percentage accuracies greater than the desired classification accuracy. However, this desired classification accuracy and the percentage accuracies were fairly low and not sufficient to use as a base for business decisions and other managerial purposes.