Eavesdropping Whilst You're Shopping: Balancing Personalisation and Privacy in Connected Retail Spaces

Vasilios Mavroudis, Michael VealearXiv 2018 · arXiv

Introduction

The digital transformation of retail has moved beyond the screen. Physical stores are increasingly deploying sophisticated sensors to mirror the data-gathering capabilities of e-commerce giants. This "connected retail" environment utilizes everything from WiFi sniffers to affective computing systems that analyze facial expressions to gauge shopper sentiment. However, this surge in data collection creates a significant friction point between the desire for personalized shopping experiences and the requirements of data protection laws like the GDPR.

The Clash of Privacies

The core of the issue lies in a fundamental "clash of privacies." On one side is privacy-as-confidentiality, where technical measures (like MAC address randomization) are used to hide a user's identity from all observers. On the other is privacy-as-control, the legal right of individuals to decide who processes their data.

Ironically, current technical protections often undermine legal ones. When smartphones rotate their MAC addresses to maintain confidentiality, they inadvertently break the mechanisms for "opting out." If a retailer cannot identify a device, they cannot honor a request from that device's owner to be excluded from tracking. This creates a stalemate where neither technical nor legal solutions are fully effective on their own.

Tracking Methodologies and Data Analysis

The research categorizes tracking technologies into two main types: Active Signalling, where the user's device participates via apps or Bluetooth beacons, and Passive Signalling, where the environment monitors the user via WiFi logging or biometric imaging.

Figure 1: Heatmap visualization of shopper movement and density within a retail environment, illustrating the output of passive tracking systems.

The technical foundation of many passive systems relies on signal triangulation. By measuring the signal strength of a mobile device across multiple access points, retailers can pinpoint a shopper's location with high precision.

Figure 2: Diagram of WiFi trilateration, where multiple antennas are used to determine the precise location of a mobile device within a physical space.

The Hybrid Randomization Scheme

To resolve the conflict between confidentiality and control, the authors propose a Hybrid Randomization Scheme for WiFi. Instead of purely random MAC address rotation, a device uses a "Seed" provided by a trusted vendor. The device generates pseudorandom MAC addresses based on this seed and a timestamp. This allows a specific, trusted vendor to recognize a returning customer (enabling rewards and personalization) while the device remains an anonymous, rotating string of bits to all other "eavesdropping" retailers. This approach enables "Privacy by Default with Selective Consent."

The paper evaluates these technologies against the GDPR, specifically questioning the legality of "checkout-free" models like Amazon Go. A primary concern is whether consent can be considered "freely given" if tracking is a mandatory prerequisite for using the service. Furthermore, biometric tracking (facial or gait recognition) is particularly invasive because, unlike a phone, a person cannot "turn off" their physical characteristics. The authors argue that "legitimate interest" is an insufficient legal basis for such invasive tracking, necessitating a shift toward explicit, transparent opt-in models.

Conclusion and Future Work

While the hybrid scheme offers a path forward for WiFi-based tracking, biometric tracking remains a significant challenge. Future research must focus on developing "Do Not Track" signals for physical, multi-modal environments that do not place an undue burden on the shopper. Ultimately, technical fixes must be paired with efforts by retailers to become "worthy of the trust" of their customers, moving beyond mere legal compliance toward genuine transparency.

Summary updated

Abstract

Physical retailers, who once led the way in tracking with loyalty cards and `reverse appends', now lag behind online competitors. Yet we might be seeing these tables turn, as many increasingly deploy technologies ranging from simple sensors to advanced emotion detection systems, even enabling them to tailor prices and shopping experiences on a per-customer basis. Here, we examine these in-store tracking technologies in the retail context, and evaluate them from both technical and regulatory standpoints. We first introduce the relevant technologies in context, before considering privacy impacts, the current remedies individuals might seek through technology and the law, and those remedies' limitations. To illustrate challenging tensions in this space we consider the feasibility of technical and legal approaches to both a) the recent `Go' store concept from Amazon which requires fine-grained, multi-modal tracking to function as a shop, and b) current challenges in opting in or out of increasingly pervasive passive Wi-Fi tracking. The `Go' store presents significant challenges with its legality in Europe significantly unclear and unilateral, technical measures to avoid biometric tracking likely ineffective. In the case of MAC addresses, we see a difficult-to-reconcile clash between privacy-as-confidentiality and privacy-as-control, and suggest a technical framework which might help balance the two. Significant challenges exist when seeking to balance personalisation with privacy, and researchers must work together, including across the boundaries of preferred privacy definitions, to come up with solutions that draw on both technology and the legal frameworks to provide effective and proportionate protection. Retailers, simultaneously, must ensure that their tracking is not just legal, but worthy of the trust of concerned data subjects.
BibTeX
@misc{mavroudis2018eavesdropping,
  author = {Vasilios Mavroudis and Michael Veale},
  title = {{Eavesdropping Whilst You're Shopping: Balancing Personalisation and Privacy in Connected Retail Spaces}},
  year = {2018},
  eprint = {1807.05381},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1807.05381},
}
Figure