Shop theft is a multi-billion-dollar problem that has plagued retailing ever since the first artisans made their wares available to the public. Since then, retailers have tried a plethora of ways to try and contain shop thieves, including: putting products behind counters; attaching security tags; creating shelf fixtures that make it harder to take multiple products; using ‘safer’ cases; and changing store layouts to make it easier for staff to see errant behavior.
Retailers have also invested in video systems to try and address this issue, hoping to both detect and deter would-be thieves. Video deterrence is premised on the idea that would-be thieves do not want to be caught on camera and so will avoid surveilled areas. Using video to detect thieves is more involved and requires the monitoring of retail spaces, often in real time, in the hope of catching offenders in the act. However, while relatively common in the 1990s-early 2000s (and still employed by a few retailers today) for the most part it is now an approach rarely adopted.
This can be explained by at least two interrelated factors. First, asking humans to monitor dozens of camera feeds at the same time for many hours as potentially hundreds or thousands of shoppers go about their business is an unrealistic proposition. While some capable video watchers can successfully look in the right place, at the right time, and at the right person to spot theft, the reality is that filtering out the trivial many to expose the vital few in this type of environment is a tough proposition for most humans.
Secondly, building a return on investment (ROI) model for this approach has proved almost impossible for most retailers—the labor costs of monitoring and then responding to any events that are identified far outweigh the value of the goods recovered. It usually works out cheaper to not catch thieves this way.
The Rise of Video Analytics
However, developments in video analytics, which can be used to automatically identify predetermined events and generate an alert, have resurrected interest in using video to address shop theft. Some technology providers are now offering systems that claim to be able to automatically identify theft, generating an alert for store staff. In a recent ECR Retail Loss online session, one antipodean retailer using such a system suggested that it had successfully reduced their unknown stock loss significantly.
Balancing Clarity, Complexity, and Accuracy
While analytics are certainly beginning to transform the way video systems can be utilized, a recent ECR Retail Loss study highlighted the importance of ensuring clarity of purpose, consistency, and accuracy in identifying predetermined events, and recognizing the impact of environmental complexity on their successful operation.
Examples of video analytics working well are in relatively uncomplicated and binary environments, where the purpose is clearly defined. For instance, identifying when a burglar is entering the back of a building. In normal circumstances, nobody should be there and therefore when somebody is, the analytic can easily notice the difference and generate an alert.
When it comes to using video analytics to identify thieves in retail stores, these factors are important considerations to bear in mind. For example, deciding when an in-aisle event is shop theft is not always straightforward—uniformity of behavior is not a trait normally associated with the shopping public. In addition, as various types of self-scan systems become more commonplace—such as mobile scan and shop, where customers scan goods and legitimately put them in their own bags and pockets—differentiating between honest and dishonest shoppers will become trickier. Deciding, therefore, what constitutes suspicious behavior becomes key; make it too broad and staff will be overwhelmed with alerts, make it too narrow and criminals may slip through the net.
In addition, a busy retail store is a complicated environment, with lots of customers interacting with each other and with products, some with and without carts and bags, and sometimes very densely packed together. This can be a harsh operating environment for a video analytic to consistently operate successfully.
Another important consideration is who responds to the alert. At a time when retailers are evaluating how they can keep their retail staff safe from violence and verbal abuse, some may question who will receive the alert from these systems and to what extent will they be put in harm’s way should the outcome turn violent. We know from other studies that most retail violence is associated with incidents of shop theft. Could these systems possibly generate more harm, not less?
Finally, for retailers with very large stores with multiple aisles and potentially thousands of customers, is it realistic to think that this type of system is scalable. Can levels of accuracy be secured so that responding staff are not simply overwhelmed with potential events they are required to deal with?
As with many types of intervention designed to tackle the problem of retail loss, it is likely that in the right context, with clearly defined processes in place, incidents of in-aisle theft can be addressed with this type of approach. Indeed, it could well become another useful tool in the armory designed to address an issue that will no doubt forever plague the retail industry.
Video Watch is a monthly column written by Professor Adrian Beck sharing insights on the proactive use and impact of video technologies in retail. It reflects the latest research and monthly discussions of the Video Working Group of ECR Retail Loss, the leading global think tank on retail loss. The research commissioned by ECR Retail Loss is supported by independent research grants provided by Genetec and other leaders in retail loss prevention.