Keywords

1 Introduction

In commerce, Customer Experience (CX) is the relationship between a firm and its customers for the duration of their interaction [2]. Considering this definition, there exists significant overlap with Human Computer Interaction (HCI) in terms of goals: to create products and experiences that are usable, useful, and enjoyable [2]. However, there exists several discrepancies in methods between the two disciplines.

CX primarily focuses on gathering self-reported user feedback, most commonly gathered through surveys and questionnaires. The difference between firms often arise in the channel through which these surveys are administered; web applications, location-based triggered mobile phone notifications, text messages and experience kiosks to name a few [2]. However, HCI professionals understand that self-reported feedback alone carries a strong bias and is insufficient for informing design decisions that best support users. When self-reported data is coupled with observational data it provides a much more robust model for understanding user needs and desires.

Observational data gathering is expensive as it traditionally relies on human intervention and intuition. However, with advances in Artificial Intelligence (AI) machines are gaining the capacity in making sense of the unstructured data observational methods yield [3]. With these advances in technology there is a new push to create innovative ways of gathering observational data to give context to self-reported user feedback.

2 Methodology

The participants in this study were anonymized students carrying a variety of devices who frequently visited a specific building on campus. No identifying information was collected. All tracking and load testing was performed using MAC addresses and the research team had no way of tying this hardware identifier with the individual owners. Methods of discovery around the specific problems to solve and the associated pain points for CX professionals included an academic literature review on tracking technologies, user observations by the team, and expert interviews with CX professionals. Based on this literature review, the team concluded that WiFi tracking is currently an underutilized method in the CX field for cross validating self report data and understanding user habits and choices.

2.1 Observational Tracking Methods

Currently, there are two leaders in gathering unstructured, observational data in the CX field. The first and most widely used method is bluetooth tracking. Bluetooth tracking requires customers to download an application to their mobile device and leave the bluetooth turned on while in the store. The systems observe user movements through the store through a series of strategically placed Bluetooth Beacons. While the bluetooth is fairly accurate in tracking device location it quickly drains the battery of mobile devices, and only updates the users location as long as the app is in the foreground [4].

Another more recent method relies on AI to recognize and tag customers actively through security camera footage. With this method firms are able to create accurate heat maps of customer activity in the store and gain valuable insights observing customer shopping habits. The drawback is that these systems are expensive. They are often required to be installed and calibrated by trained technical personnel and require some effort to maintain accuracy [5].

Finally, one of the newest observational data gathering methods leverages WiFi protocols to understand a users location. The benefits to using WiFi tracking are that it requires no additional software on user devices, however, it provides those conducting the tracking a universally unique identifier (the anonymized device’s mac address) and it does not take any more than the devices normal battery usage [6]. For these innovative reasons, exploring WiFi protocols to observe user experiences is an exciting possibility for gathering observational data.

2.2 WiFi Probe Requests

As part of 802.11 WiFi protocol, WiFi enabled devices emit a signal called a probe request. Each probe request contains the devices MAC address, a universally unique identifier of the device, and the signal strength of the router with relation to the device. Probe requests are designed to find available networks the device is able to join. The router then returns a similar probe request to be received by all the proximal WiFi devices communicating their connection status [7].

2.3 Our Setup

In order to capture and analyze probe requests the team used three Raspberry Pi micro computers with 150 Mbps WiFi adapters (TP-LINK TL-WN722 N). Each Pi (or node) had a unique title and sent its logged probe requests to a central server where they were permanently stored and displayed with a web application. Each node sent its newly gathered probe request data to the server every ten seconds.

figure a

2.4 Time-Based Tracking

With the Raspberry Pi and server set up, gathering and analyzing time-based data was as easy as plugging the Pi into the wall and running a short Python script. There is significant value in gathering time-based data alone. Depending on the frequency of probe requests, CX researchers would gain contextual insight into when a user physically visited the store location and how long they spent there. Coupling this data with an unstructured survey administered to the user later yields valuable insights to the relationship of time spent in store and customer purchasing activity.

However, such insights rely on data being gathered accurately. To test the accuracy of time-based data tracking the team walked in and out of the WiFi range of the Raspberry Pi with two mobile devices. The first was actively recording a video with in order to log the time and location-based of the user. The second device was asleep in the testers pocket. The test yielded these findings.

Devices send 50% more probe requests when they are actively being used than when they are in standby mode. Further, android devices, specifically those manufactured by Samsung, consistently send 30% more probe requests than the next highest manufacturer. Devices manufactured by Apple send the least amount of probe requests from the devices measured. However, all devices correctly reflected the time spent in the target location with a deviation of only four minutes.

figure b

2.5 Location-Based Tracking

One of the most exciting possibilities of WiFi tracking is the ability to determine a customer’s physical location within a store. A customer’s location is identified using the logged signal strength from three receiving nodes. A node is a router that based on how far away the device is from the node (router) provides a signal strength or level. Current WiFi location-based tracking solutions use triangulation algorithms, however, these rely on a very predictable signal and tend to only work in more open spaces. Attenuation through walls or exhibits is difficult to predict and delivers unreliable results.

In interviews with CX professionals it became apparent that what they wanted to know was simply whether a user had visited a specific location or not. For example whether they had visited the checkout or a featured exhibit. The team chose to test a more simple method of tracking which would only log a users location in predetermined positions. With a simple machine learning algorithm and a learning interface on the web application store managers would train the algorithm by providing their own mac address and a specific location tag and then using their mobile device at that location. Once they have been there for 5 min the machine learning algorithm delivers a classifier which is used to measure the locations of all other devices.

figure c

While the machine learning method cannot provide continuous location-based information, it does provides predetermined location-based information at roughly half of the computing cost. This method still requires a significant amount of modeling and reliability testing, however, early results are promising.

2.6 Privacy Concerns

The ethical considerations of implementing a technology solution with the capacity to track all WiFi enabled devices are significant. However, they are comparable to the use of security cameras in a business. To combat possible privacy concerns, business users will be made aware that they are being tracked by WiFi upon entering the store. Users will also be informed to disable the WiFi on their device if they wish not to participate. Awareness and transparency to the user are key to the ethical implementation of this technology.

3 Conclusion

Coupling self-reported user feedback with insights from unstructured tracking data may provide the necessary context for CX professionals to make impactful decisions for their products and brand experiences. Further, with the consent of the user there are valuable HCI research applications in understanding users habits and interactions in a variety of physical spaces at scale. While the concept has significant limitations as it is still in its infancy, it is ripe for further innovation, development and research.