Predictive analytics can help retailers to foresee their customers’ preferences and purchase choices. This will help store managers decide which product mix to put on sale and to manage their inventory more efficiently.

Retail: algorithms are helping to predict customer choices

The rise and rise of data collection in the era of Big Data presents retail businesses with a sizeable challenge. Customer data is proving extremely useful, and has today become a real strategic tool for retail players. However, sorting the huge volumes of information available requires advanced software systems. Celect, a Cambridge, Massachusetts startup working in this field, showcased its innovative offering at the recent DEMO Fall 2014 event in San Jose, California, where it won a ‘DEMO God’ award in the Smart Data category.

Celect was founded by three retail and machine learning specialists, among them Devavrat Shah, a professor at MIT whose work on developing predictive analysis algorithms L’Atelier reported on in an earlier article. His discoveries have called into question current consumer recommendation systems, based on absolute ranking of products and services on a fixed scale, and he argues that a system based on multiple paired comparisons is much more logical and accurate as customer choices are seldom about absolute judgement but choosing between a range of options. Moreover, if a customer has the choice between a bottle of champagne costing $40 and one at $20, there is a high probability that s/he will choose the $20 item. On the other hand, if the choice is between a bottle at $40 and one at $200, people tend to feel less guilty about spending $40 on the champagne. Shah decided to set up Celect and to apply his insights on consumer choice to the real retail world.

What to put where and when

One major concern for bricks-and-mortar stores is optimising the product mix and the underlying stock management. Retail businesses can obtain analytical data to manage their stock, using software packages such as those offered by JDA and Galleria, but that will not enable them to predict customer choice and adjust their product range and mix accordingly. This is where Celect comes in, providing a way of forecasting what the customer will want to buy from the choice of brands and items available, linked to a stock management programme which decides which product to place in which store for optimum results.

The Celect Choice Engine, which is based on a sophisticated combination of machine learning and Big Data, claims to be the first tool which models customer choice. The software draws together data from previous inventories, from CRM tools and from recorded customer preferences at specific local stores, and the system displays all this information in the form of dashboard. Store managers will be able to see the range of brands sold at the various outlets as larger or smaller boxes depending on the space and purchase budget allocated to them during previous sales periods. Retailers can use the Celect tool to view new brands they should consider putting on sale, and see how one brand compares with another price-wise. After inputting this data, the system will display new sizes for each brand’s box, indicating which purchase budget to increase or decrease. The Celect predictive software programme links directly to the retailer’s inventory system, providing an all-in-one stocking tool for retailers and merchandisers.

Anticipating customer choice: art or science?

All retail businesses are of course always trying to anticipate what consumers will want to buy and get it on display in order to improve the customer experience. Celect clams to “bring science to the art of retail”. The company website says that “optimized, hyper-local store assortments produced via Celect Choice Engine yield revenue increases of 7%-plus for the same spend″ and claims that the ROI on the Celect Choice Engine is “dramatic – much greater than 10 times – and visible within weeks″.  

Predictive Analytics technology certainly appears to have a great deal of growth potential in the field of commerce, whether of the bricks-and-mortar variety or online. Early this year Amazon announced it had filed a patent for an anticipatory shipping system. The e-commerce giant is hoping to be able to forecast what its customers want to buy before they type in their orders, based on their previous searches, their current average basket, or the time they spend looking at a product on screen. However, predicting future customer choices based on previous purchase history is far from being an exact science, especially for an individual customer, whose tastes and preferences may fluctuate and be subject to whim. It would seem more logical to use the predictive analytics algorithms approach across an entire group of people.

By Eliane HONG