Personalized vs. non-personalized engagement for at-risk new users in data-scarce contexts

Publication Date

1-1-2025

Document Type

Article

Publication Title

Information Technology and Management

DOI

10.1007/s10799-025-00446-5

Abstract

Recommendation systems are integral to online platforms, enhancing user engagement through personalized experiences. However, newly registered users (newcomers) often need to find a way to benefit from such systems due to a lack of traced user profiles and consumption patterns. This limitation is particularly critical in non-contractual platforms where users can disengage without notice, necessitating swift identification of at-risk users and the development of effective retention strategies. Despite its significance, prior research still needs to sufficiently explore the challenges of engaging at-risk newcomers. To address this gap, we collaborated with StumbleUpon, a non-contractual webpage recommendation platform, to examine whether providing personalized recommendations, even when imprecise due to limited data, is more effective than offering non-personalized content such as popular recommendations. Using a mixed-methods approach, we first employed statistical and machine learning models to identify at-risk users based on newly registered subscriber data. We then conducted a randomized field experiment comparing the effectiveness of personalized versus non-personalized email strategies. Results revealed that while personalized emails yielded higher open rates, non-personalized emails achieved significantly higher click-to-open ratios and greater platform engagement. Drawing on expectation-disconfirmation theory (EDT), we found that users exposed to personalized emails were more likely to show interest in personalized recommendations; however, they expressed lower satisfaction with the content compared to those who received non-personalized recommendations. Furthermore, demographic and usage characteristics, such as being older, English-speaking, male, and mobile-based, were associated with increased engagement. These findings provide actionable insights for optimizing recommendation systems and tailoring retention strategies in data-sparse environments.

Funding Number

P0025238

Funding Sponsor

Korea Institute for Advancement of Technology

Keywords

At-risk users, Customer engagement, Non-personalized-message, Recommendation systems

Department

Information Systems and Technology

Link to Full Text

Share

COinS