Data Science in Brick & Mortar Retail: Turning Big Data into Sales
One of the main challenges of retail chains and their suppliers is the out-of-shelf problem. Customers looking for a specific product are unable to find it which results in decreased revenues for the retailer. The supplier suffers too as this situation increases the risk of the consumer buying a competitor’s product instead and thus potentially shifting their future preference.
Various estimates indicate that the out-of-shelf time amounts to 5-30%. And even the lowest estimate is not something negligible. For example, if a chain’s total turnover increases by just 2%, the net profit can grow much more significantly – by 7-12%. The reason is the peculiarity of the retail industry where the fixed costs constitute a very large portion of all expenses, e.g., running the office costs the same regardless of the sales numbers.
Naturally, both the retailer and supplier work hard to ensure that the goods are on the shelves at all times, but the stats show that their efforts are not entirely successful: according to CrowdSystems, the average goods availability in Russian retail chains is 79%. The inventory estimates themselves sometimes carry error, despite a complex data collection systems set up by the vendors.
The main challenge is to accurately determine whether a product is on the shelf and prevent its absence. There are a few methods to achieve that: forecasting, clustering, segmentation, classification, pattern detection, and so on. They enable a timely determination of what’s out-of-shelf and has to be replenished. The task at hand has many aspects and stages each of which offer several potential solutions to choose from. Whatever you choose though, an important part of the solution is evaluating the algorithms by a set of metrics specifically tailored to this particular problem.
In 2016, we piloted our own solution – the Optimal Shelf Availability Hybrid Platform (OSA HP). The introduction of the Platform into one of the largest retail chains in Russia for the months of August to November increased the aggregate turnover in 10 categories by 5.4%. Various brands saw 2.3% to 11.3% sales increase.
We created a novel approach at the intersection of machine learning and retail business processes and named it Data Sensing. We employ simulation modeling, deep learning, and image recognition. Our service is rapidly developing and we constantly expand our talent. We already have a proven record of solving the above described problem by applying machine learning to retail industry and we are just warming up. If you believe you can add value to our solution and are excited about developing your skills in the context of the omnipresent and enormous industry of retail – join us!