Mass customization and AI-powered supply chains: product variety and low inventories are not an oxymoron
Part I: Forecasting for long-tail products
By Kishore Jethanandani
The Amazon effect is writ large on the upheaval buffeting consumer markets and their supply chains as delivery time is shortened to a day. Close on its heels, the Shopify effect — mass customization combined with the speed of delivery — is reshaping consumer markets and their supply chains. Unprecedented demand volatility has rocked consumer markets as consumers crave novelty and personalization in not only products but also in their merchandising. Sales forecasting is prone to the risk of larger errors which can only be mitigated by innovation in tracking early indicators of demand for innovative products.
Personalization and customization are more than an afterthought for consumers — they materially influence their product choices. This is reflected in the fact that collaborative customization, including both function and form, is the leading type of customization. Transparent customization which affects functionality but not the form (packaging) is a close second while cosmetic customization of form alone is the least favored.
Ease of product discovery, aided by digital assistants, and exhilarating experiences offered by vendors are encouraging consumers to experiment. NorthFace, for example, learns about the individual needs of customers from their interactions with its conversational assistant and their responses to its recommendations. One-of-a-kind experiences in pop-up shops whet the appetite of customers — Glossier, for example, recreates the ambiance of Seattle’s landscape in its stores with its lush green foliage where its cosmetics nestle.
Direct-to-consumer start-ups led the drive for customization initially with niche products for customers. They are now scaling to grow at an accelerated rate by nurturing categories of lifestyle products around their debut launches. Casper, for example, started as a mattress company is evolving into a sleep company. Similarly, Harry’s which began as a shaving company is now extending into bathroom products. As a result, the growth rates of emerging direct-to-consumer brands are much higher than brick-and-mortar stores. For example, the US personal care and beauty product sales grew by a modest 4.5 percent in 2018 in contrast to 24 percent growth achieved by online sellers while direct-to-consumer brand SiO tripled sales in 2018.
The ripple effect of the direct-to-consumer segment is manifest in a billion-dollar investment Shopify is making to build a network of fulfillment centers tailored for the needs of direct-to-consumer vendors. It will be able to deliver products, in two days, across all but one percent of the USA which is competitive with Amazon. A whole new crop of companies are providing infrastructure for DTC companies for customer acquisition, supply chains, and machine learning.
As they scale, DTC brands are putting their stamp on brick-and-mortar stores. They are reigniting interest in these stores by redesigning them to bring new experiences to shopping.
Customization and inventory management for the long tail
Product variety creates forbidding challenges for inventory management of small lot sizes of products with a great deal of churn in the composition of the product portfolio. Forecasting demand is daunting for products that have a short history of sales. According to a study completed by E2Open in 2018, the number of active product items (net of those discontinued) increased by 36 percent, since 2010, while sales rose by only 15 percent corresponding to a decline in the market size of each item by 17 percent. The number of active and discontinued products or the total number of products newly launched or existing ones, increased by 263 percent in the same period.
While direct-to-consumer vendors often book demand online before they place orders, the survey showed that long-tail companies are responsible for much of the inventory. The top 11 percent of the fastest moving items account for 80 percent of the sales while 89 percent of the items in the long tail, with the slowest velocity of turnover, account for only 20 percent of the sales.
The long tail items are more prone to forecast error as the higher expectations of innovative products, at the outset, are not met. Their shorter history does not provide a large enough dataset for precision in forecasts. The extent of the error is muted by demand sensing or by reading the pulse of market demand at the moment. “Manufacturers still carry high levels of inventory because of the volatility in demand preferences and the safety stock they keep to secure themselves against the penalties they are required to pay in the event the precise specifications are not met,” Robert F. Byrne, Vice-President, E2Open told us.
Digital tools do provide visibility into the supply chain across multiple geographies. Artificial intelligence keeps track of inventories in excess supply in a region that can potentially be moved to another where it is running short. “Air transportation costs, the quickest way to move inventory, are disproportionate for low-value items. Trucks often have spare capacity, but it is hard to identify it in time to move goods fast enough with the current state-of-digital technology in logistics. However, upstarts in the logistics industry are investing large sums of money to digitize their processes which will make spare capacity visible,” Robert F. Byrne told us.
Advancements in forecasting techniques mitigate the risk of significant errors that could cause stock-outs or excessive inventory. “Machine intelligence paired with human knowledge significantly lowers forecasting error even in situations of product transitions,” Colin Kessinger, Ph.D., Managing Director of End-to-End Analytics, LLC surmised. “Forecasts for product categories are accurate despite product variation and innovation,” Colin Kessinger asserted. “Forecasts for entirely new product categories are error-prone due to an absence of any representative sales history,” Colin Kessinger agreed.
“The price structure of products within a category can be a reliable predictor of demand. Innovations in cars, for example, begin with higher-priced luxury models and spread to mid-level and lower price models,” Colin Kessinger explained. “The demand for these innovations in higher-priced luxury models is an early indicator for the market size of the category and customers of high-price luxury models tend to spend the same amount whatever they are,” Colin Kessingner added.
A variety of techniques for demand sensing provide early indicators of the demand for innovative products even before any of the products in a category are launched. Listening to comments on social media sites, with artificial intelligence tools for gauging sentiment, provides early indicators of preferences and purchase intentions. Demand generation can happen spontaneously when customers are encouraged to create their own products. A case in point is Adagio Teas which lets customers create their own tea mix and share it socially and they earn reward points when any of their friends or followers purchase it. Styku offers 3D models of products that make it easy to configure products for the precise body dimensions of individual customers.
Greater error in sales forecasting is unlikely to be a wrench in the growing trend towards mass customization. It will, however, require greater ingenuity in forecasting methods to prevent the expansion of the size of the errors. We can expect that demand sensing will become a growing art as more technology tools are brought to bear to feed early indicators to forecasters.