This is just some announcement which people will want to pay attention to! Learn More
This is just some announcement which people will want to pay attention to! Learn More

A Japanese department store increased revenues by over 3% at a 5% greater margin, with a 15% lower cost of discounting

Discount strategies are complex

Getting your store discounting strategy right, especially in the fresh foods and fast-moving consumer goods categories, can mean an addition to your bottom line of 2-4%. That can be huge in an industry that averages just 2-5% earnings after tax. Unfortunately, most retailers still rely on packaged discounting solutions that make statistical estimates based on historical transactional data and apply the same calculations across their stores without considering demographic, competitor, and consumer behavior variations at each store. As a result, a lot of the value that could be extracted by developing more precise price prediction models is left unrealized.

For example, slight product variations like net weight, an additional ingredient, or an alternative flavor can create significant differences in sensitivity to price changes. Most existing demand forecasting solutions and pricing solutions can’t account for these variations, and therefore continue to recommend sub-optimal strategies. Most retailers also still rely on the expertise of store managers, and make intuitive changes based on their experience. These strategies work only if those intuitions are correct, and even when correct they are never consistent.

An innovative retailer tries AI solutions

A medium-sized Japanese retailer known for everyday low prices, similar to the Wal-Mart model in the United States, was keen to embrace AI solutions to improve their discount strategy. Known for their innovative approaches, the retailer wanted to improve operational efficiency and achieve more profitable growth through the application of AI.

They were looking for a partner company that could analyze the current state of their digital assets and help them determine their immediate readiness for AI, identify potential use cases, provide a roadmap for AI development, and begin by implementing one or two proposed solutions that would provide the client with the quickest ROI.

Our approach

CrowdANALYTIX first held a half-day workshop with the client’s group CIO, President, and the head of their merchandising unit. Potential solutions including price discounting, assortment optimization, dynamic mark-downs, logistics optimization, and more were discussed. Each option was analyzed by CrowdANALYTIX based on the availability of data, Price discounting was picked as the first solution to be built. Because the client had a template for what they wanted to achieve with the help of the algorithms, the problem as well as the business metrics the solution would impact were well-defined for this solution.

The first step was to add meta-tags to the retailer’s product data, create more intuitive features from their point of sale (PoS) data, and normalize their historical discounting and campaigns data. Once the data extraction and restructuring algorithms had been built, they were deployed and utilized to process historical and current data.

This processed data was then used to build three types of models: 1) CrowdANALYTIX first clustered and ranked products based on their elasticity and ability to influence the sale of other SKUs, which we called the Linked SKUs. 2) Clusters were created based on the size of transactions so that the model could focus on optimizing revenue from the more regular customers, keeping very large and very small customers out of the calculations so that the models did not provide skewed results. 3) Finally, the recommendation engine was deployed, providing the client with recommendations on which SKUs to discount by how much in each store every week.

Impact

Early field tests were very encouraging and resulted in over 3.5% increase in revenues for the selected categories and stores. The ROI on the investment was estimated to be between 480% to 630%, which was much higher than expectations. Encouraged by the results, the client is planning on deploying the solution first in 50 stores. By the middle of next year, the solution will be rolled out to all stores in Japan.

Data used: Historical sales data, Product catalog

Share This Post

Share on facebook
Share on linkedin
Share on twitter
Share on email

More To Explore