This article first appeared in the
Autumn 2017 issue of The Record.
Retailers worldwide sell thousands of products every year. Fashion retailer City Beach, for example, operates an ¬e-¬commerce outlet and 62 stores across Australia, adding 300-400 new lines to its 60,000-strong product portfolio each week. Meanwhile, US-based homeware and décor brand Pier 1 Imports has an online store, boutiques in Mexico and 1,000 physical stores across North America. But how do these retailers know which products will sell? And how do they choose the optimum product assortment and inventory level to ensure that they can always meet customer demand?
According to Microsoft’s Retail Industry lead for data and analytics ShiSh Shridhar, retailers must first ask themselves two key questions. Do they need a repetitive replenishment model, or do they need a model that enables them to regularly update their product assortments as seasons, trends and customer demands change?
“At one end of the spectrum, grocery retailers repeatedly replace the same products in a timely fashion to ensure they are never out of stock; at the other end, fashion retailers build a new product assortment each time they replenish stock, switching colours, styles and more,” he explains.
“Others, such as electronics retailers, take a hybrid approach by repeatedly replenishing the products that are selling fast, while continually monitoring trends to ensure they are replaced when customers demand the latest smartphone or TV, for example.”
Once they have identified their position on the spectrum, retailers must choose the optimal product assortment and then ensure they have the right inventory levels to prevent out-of-stock or over-stock situations. However, these processes can be challenging, particularly for brands that have multiple stores, cautions Shridhar.
“While retailers can use their aggregated sales data to create a network-wide replenishment schedule for core products, they can’t assume that customer demand for every single product will be exactly the same at all of their stores,” he explains.
“Everything from the climate to customers’ socio-economic status can cause demand for specific products to differ from one store location to another. For example, organic vegetables will likely sell better in stores in affluent towns, compared to those where customers have a lower income. Consequently, to prevent over-stock or out-of-stock situations, retailers must plan assortments at a hyper-local level.”
Tom Fuyala, CEO of Microsoft partner 11Ants Analytics, claims that planning assortments to match the specific demand at each store is an intractable problem for retailers.
“Operationally, the simplest scenario is for retailers to offer identical assortments across their entire network, but this means products don’t necessarily match actual customer demand, which directly impacts sales and gross profit,” he comments.
“The opposite extreme where companies create customised assortments on an individual store basis better matches products with customer demand, but is often operationally challenging. Consequently, many retailers have a company-wide core range augmented with localised assortments, or develop clusters of stores with each cluster having its own unique assortment. The more these can be optimised, the greater the impact on store profitability.”
Shridhar recommends that retailers harness advanced analytics, machine learning, business intelligence and other data mining and visualisation tools to identify what products will sell well at different stores.
“Retailers can use Microsoft’s Cortana Intelligence Suite to capture their historical sales data and analyse it alongside information from external sources – such as economic indicators, news sources and weather reports – to find patterns that will help them to accurately predict demand at a hyper-local level,” says Shridhar. “By applying the cognitive capabilities within artificial intelligence (AI) solutions and Microsoft Azure Machine Learning, retailers can even perform sentiment analysis on social media posts to pinpoint what colours, styles or fashions are most popular in specific locations. Power BI makes it easy for brands to visualise these insights.”
Microsoft partner Neal Analytics has used Azure Machine Learning to help beverage manufacturer and distributor Arca Continental improve the accuracy of its demand forecasting for Coca-Cola sales in Mexico. Neal Analytics’ predictive analytics solution identifies the factors that lead to people buying Coca-Cola, and runs the data against Arca Continental’s historic sales data so the company can deliver the right level of inventory to stores in various cities at different times of the year.
Machine learning can also help when it comes to overcoming another key demand forecasting challenge – setting the right price.
“If products are too expensive they won’t sell, but if they’re too cheap they’ll go out of stock quickly and the retailer will lose profit,” says Shridhar. “AI and machine learning allow retailers to analyse their sales data, competitors’ prices and socio-economic indicators to determine the optimum prices for their products.”
Microsoft partner Blue Yonder, for example, has used Azure Machine Learning to help multichannel retailer OTTO measure the connection between customer demand patterns and price changes. The Blue Yonder Price Optimization solution automatically determines sales- or profit-increasing prices throughout the entire lifecycle of every product for every season, which has boosted sales and revenue, and decreased product returns.
Retailers are also relying on advanced analytics to optimise shelf space, ensuring that stores are filled with both profitable products and core, but not necessarily profitable, products.
“Bread isn’t necessarily profitable, but if a grocery store stopped selling it then every customer that wants bread as part of their weekly shop will go elsewhere,” Shridhar says.
“By using advanced analytics, retailers can identify their core products and perhaps find more profitable versions of each one to ensure they satisfy customer demand, while maximising their profit margins.”
Shridhar expects that more retailers will soon adopt machine learning and advanced analytics to overcome the four key demand forecasting challenges – having the right assortment, ideal inventory level and best prices, and optimising shelf space. “Once they can better predict and meet customer demand, retailers will drive sales and revenue, and boost customer loyalty,” he says.