"I just hated it and I want my money back": Data-driven Understanding of Mobile VPN Service Switching Preferences in The Wild
Rohit Raj, Mridul Newar, Mainack Mondal
TL;DR
This study investigates why users switch mobile VPN providers by analyzing 1.3 million reviews across 20 VPN apps and 376 VPN-review blogs. It combines qualitative open coding with scalable ML pipelines (DeBERTa for switch classification and BART for theme labeling) to create an extended dataset of 185,399 VPN-switching reviews and to identify a hierarchical set of switching reasons and desired features. The analysis reveals that 45% of reviews mention switching, with a clear hierarchy of factors (security/privacy, speed, usability, geography, and finances) and distinct feature co-occurrence patterns, highlighting substantial differences from VPN adoption motives. It also uncovers biases and gaps in VPN review blogs, emphasizing the need for more design- and research-driven tools to help users switch based on substantive requirements rather than promotional content. The work provides concrete implications for VPN providers, review sites, and the security community, including recommendations to tailor features by geography, reduce blog bias, and develop evaluative tools for user-driven VPN switching.
Abstract
Virtual Private Networks (VPNs) are a crucial Privacy-Enhancing Technology (PET) leveraged by millions of users and catered by multiple VPN providers worldwide; thus, understanding the user preferences for the choice of VPN apps should be of importance and interest to the security community. To that end, prior studies looked into the usage, awareness and adoption of VPN users and the perceptions of providers. However, no study so far has looked into the user preferences and underlying reasons for switching among VPN providers and identified features that presumably enhance users' VPN experience. This work aims to bridge this gap and shed light on the underlying factors that drive existing users when they switch from one VPN to another. In this work, we analyzed over 1.3 million reviews from 20 leading VPN apps, identifying 1,305 explicit mentions and intents to switch. Our NLP-based analysis unveiled distinct clusters of factors motivating users to switch. An examination of 376 blogs from six popular VPN recommendation sites revealed biases in the content, and we found ignorance towards user preferences. We conclude by identifying the key implications of our work for different stakeholders. The data and code for this work is available at https://github.com/Mainack/switch-vpn-datacode-sec24.