U.K. Retailer Embraces Fresh Thinking

Product demand, it’s often said, can run hot or cold. One U.K. retailer is taking that idea to its logical extreme by mapping daily product sales to weather changes. Their goal is to create real-time demand forecasting for thousands of products across more than 3,500 stores based on historical sales analysis and weather forecasting data.

It’s fresh thinking in more ways than one. By accurately predicting product demand, the retailer hopes to reduce the amount of food that is inevitably wasted—particularly fresh food such as fruits, meats and vegetables. Accurate stock replenishment is not only good for the environment, but good for their bottom line as well, because changes in the weather can drive increased demand for certain kinds of products.

Sometimes, Simple Logic Isn’t Enough

The retailer faced a typical big data challenge: they needed to analyze a lot of data in very little time. Their database consisted of 5,000 separate stock keeping units (SKUs) across thousands of stores with historical sales data from the last three years, amounting to trillions of bytes of data.

Initially, the retailer used their existing in-house analytics tool to develop their sales forecasting model. The trouble was, the massive size of their database required them to “chunk” the data across multiple analytics servers in order to break the problem down to a manageable size. Even then, results still took five days to compile, by which time the forecasts were no longer valid. So the retailer was left with little choice but to manage its stock levels based on crude approximations and educated guesswork.

Fuzzy Logix to the Rescue

The main challenge with predicting sales based on weather patterns is that weather changes quickly and often. The retailer was getting weather feeds every four hours into their database, but they weren’t able to understand its impact until four or five days later. What they needed was a better analytics solution; one that could analyze a mountain of data without moving that mountain back and forth between an army of servers. They found that solution with DB Lytix, the in-database analytics tool from Fuzzy Logix.

With Fuzzy Logix, the retailer was able to create a more robust forecasting model that took into account factors such as recent weather forecasts, store locations and SKUs. And what emerged was a clearer connection between product sales, stores and the weather. For example, the retailer was able with Fuzzy Logix to ascribe “hot” and “cold” values to its products and assign weather factors to each of them. Products that sold better on hot days such as ice cream or salad were assigned a weather factor; a 1.25 high-hot factor would mean sales of that product would be 25% higher during a warm day. Similarly, high-cold items such as soup or roasted chicken would have a correspondingly high value on cold days.

In addition to identifying hot and cold factors for its product mix, the retailer also sought to cluster its stores into groups based on their sensitivity to weather. At first, the retailer assumed that stores in the same geographic locations would appear in the same cluster, but analysis with Fuzzy Logix proved otherwise. What the retailer found was that geographic location was less of a factor in determining weather-related sales than the proximity of people and places. Stores in urban areas, for example, reacted to weather changes in one way, while stores located near public parks represented a different kind of cluster. By combining store clusters with the various weather factors of products, the retailer now had a more accurate forecasting model that changed as often as the weather.

The Importance of Time

The most critical factor in sales forecasting is one of time. It takes fresh answers to ensure fresh products in thousands of stores. With Fuzzy Logix, the retailer was able to accelerate the analytics process from five days to 46 minutes. Early analysis indicates that the retailer will be able to save millions of pounds with its new sales forecasting solution and significantly reduce food waste. That would allow the retailer to achieve break-even ROI within the first month that the new forecasting model goes live—a prospect that will likely drive their appetite for more Fuzzy Logix in the future.

Supply chain analytics