Generating Inventory Data Through Video Analytics
Generating Inventory Data Through Video Analytics: Development or Distraction?

May 11, 2021
By Adrian Beck

As part of the ECR Retail Loss Group’s regular sessions exploring the use of video technologies in retailing, the subject of utilising them to facilitate the collection of inventory data was recently the focus of attention. Presentations were made by representatives from Walgreens, the giant US pharmacy retailer, and Monoprix, a French grocery chain with over 650 stores. Both offered their experiences on different approaches to using a video-based system to capture out of stock events at the shelf edge.

At Walgreens, the focus of their experiment was to not only see whether existing ceiling cameras could be used to capture when categories of product were out of stock, but also measure the accuracy of the AI model designed to do this, including the number of images it required to be ‘trained’ to successfully recognise an out-of-stock event across a range of product types. For Monoprix, the focus was more upon analysing the capability of cameras fitted at the shelf edge and how this technology could be combined with an existing system of Electronic Shelf Labels to improve the overall accuracy of the programme. Again, the system was focussed upon trying to determine when a given SKU was out of stock at the shelf. These presentations and the subsequent discussion amongst 30 plus retailers generated a range of interesting points and observations.

Key Lessons

There is clearly an increasing recognition of the importance of inventory visibility and accuracy and how this can have a direct impact upon business profitability and customer satisfaction. As discussed in earlier blogs, the growing importance of omni-channel retailing undoubtedly shines an intense spotlight on inventory accuracy, especially when retail stores are utilised as fulfilment centres. Not being able to complete a customer’s online order due to an unforeseen out of stock situation is not good for a retailer’s reputation nor customer retention.

As the ECR Report on video in retailing highlighted, there is a growing range of potential use cases for this technology, moving beyond the traditional focus upon safety and security, and increasingly as a useful tool to improve a retailer’s overall performance. This has been driven by developments in digital systems that make possible a range of analytical capabilities that would be unimaginable with previous analogue technologies.

However, the Report also highlighted the very real challenges in getting many video analytics to work consistently in retail environments – they are complex, fast moving and constantly changing spaces. Moreover, for many video analytics to work as anticipated, they need to be finely tuned to the environment within which will operate – mundane factors such as the position of windows and the nature of surface materials can easily affect the capacity of these types of systems to reliably capture images capable of being analysed.

Regarding both retailer case studies, the extent to which these video-based systems could capture information about store inventory was relatively limited although not without value. They were only focussed upon capturing essentially binary data points – whether any given product is in stock on the shelf or not. There was no capacity to count the stock that was present. In many respects this is understandable, particularly given the Walgreen example, where they were utilising pre-existing cameras located on the store ceiling. It is hard to image how they would be able to ‘see’ into the back of a shelf, especially those nearest to the floor. It is probably for this reason that systems such as that used in the Amazon Go stores make use of shelves that can measure the weight of products to garner much more precise information on stock availability and product movements. The video technology in these stores is much more likely being used to track consumers and associate them with data points triggered by the shelves.

The Monoprix case study shed light on the value of this type of data integration but in their case the shelf-based data was coming from Electronic Labels that were used by the camera system to understand which products were out of stock/or not. This is an important learning – most technologies in retailing have limitations – Radio Frequency tags do not like being near metal and viscous fluids, video cameras find it hard to ‘look’ around corners, Electronic Article Surveillance tags are compromised when put inside aluminium. However, by combining the data from various technologies the frailties of one can be diminished by the capabilities of another – in effect making the data capture net larger and finer grained. But, this can come at a price and what was interesting about the Walgreen example was their efforts to try and repurpose existing technology to meet a specific need to help make the Return on Investment case more attractive. They freely admitted though that while the pilot project had been ‘successful’, there was some way to go before it would be rolled out across the retail estate.

One of the key findings from the ECR report on video analytics was that while they are beginning to offer real value across a range of retail environments and use cases, they are not a plug-in and forget technology; there needs to be recognition of the complexity of the environment in which they will have to operate and clarity of purpose in what they are tasked to deliver. If not, then they can become a terrible distraction, potentially hindering the application of other approaches that may be more capable of solving the problem. No doubt they will have a role to play in achieving greater inventory accuracy in the future, but what can and cannot be delivered now needs to be carefully understood when building a viable business case.

Leave a Reply

Your email address will not be published. Required fields are marked *

Post comment