It’s All About the Customer: A Better Way to Forecast Retail Demand

What if retailers could increase their full-margin sales by making informed, customer-centric predictions about what shoppers are likely to buy in the coming months? It's being done already by forward-looking retailers and brands that have invested in transforming how they forecast retail demand. It all starts with understanding why, not what they’re buying.

Due to unpredictable pandemic-driven consumer shopping patterns, our modern problems with supply chains, shipping logistics, and more come as no surprise. The world of e-commerce demand forecasting has become a bit discombobulated, yet it’s often key for retailers in order to plan out their inventory for the coming seasons and predict demand in a way that ensures the maximum amount of full-margin sales.

Forecasting demand for an item so accurately that you’re able to sell that item at full margin is crucial. Consider a situation where a blazer’s full listing price is $120, but due to it not selling for whatever reason (the style is no longer current, it missed the mark on what’s trending, etc.), it then has to be marked down to $99. Or perhaps the price has to be dropped even lower — $45.99 — in order for it to sell at a location such as an outlet mall.

Avoiding this is job No. 1 for the demand planning team. Their primary goal is to not take a hit on that margin. In order to do that successfully, most retailers do what anyone would do when faced with uncertainty about what to stock up on for the next season. They look at what sold well in the past and then they order more of it; and, if it’s no longer being manufactured, they’ll order items as close in style, color and fit to that item as possible.

Of course, there are elements of prediction and trend forecasting involved as well, which are highly qualitative processes that can be sharpened with relevant, customer-centric quantitative data. This data is the key to retailers being positioned to anticipate demand and thus eliminate the uncertainty that causes overstocking, or even understocking. Popular items tend to sell out quickly and if you don’t anticipate which items will be popular accurately then your store will certainly lose customers to “out of stock” frustrations.

Customer-Centric Demand Forecasting

The key to better demand prediction lies in forecasting demand using a language of customer-driven attributes — i.e., not merchant-driven attributes. Merchant-driven attributes are those that can be predictable and cookie-cutter, provided directly by the manufacturer, distributor or manually attributed by the end retailer itself. They often only capture part of the picture and miss so many of the ways people in the real world search for products.

Consider a "woman’s lace dress." Merchant-driven attributes will always tell you that it’s “women’s,” it’s a “dress,” and it has a “lace fabric.” They will likely provide the colors and the size of the dress as well, but that’s about it. Yet, customers in the real world will also sometimes want more on that dress. Maybe they want the lacing to be in a floral pattern or they’ll have a predetermined sense of their desired cut, sleeve length, dress length, and whether that lace dress is intended for a special occasion, such as a wedding.

If that lace dress isn’t attributed correctly right at item set-up, it won’t be easily found in online searches — neither via search engine optimization/search engine marketing nor on-site search. And it certainly won’t be recommended to a consumer who might otherwise love to add it to her online cart. It might not even be ordered for next season, despite its popularity, because merchandise planning and demand forecasting teams thought it was the “lace fabric” nature of it that drove the sales of it — not the fact that it was floral pattern + midi length + sleeveless that actually got customers buying it in stores and online.

How Customer-Centered Proxy Products Lead the Way

So, there lies the problem in a nutshell (or shopping cart). But what’s the solution? The solution is to expand the attributes assigned to any given product to encompass the language of customers, and not merely the language of merchants. By doing this, retailers can increase their ability to sell their customers what they’re actually looking for and decrease the need to mark that inventory down later.

One key facet to making this work is proxy products, which are products that most closely resemble those that sold well in the past. However, this time, they’re determined based on 20 to 30 customer-centric product attributes, and not just a mere five or so merchant-centric product attributes. Just as in the lace dress example above, a retail e-commerce brand may find available proxy products for something that sold well in the past. Ordering its counterpart can occur with confidence, instilled by knowing it was the combination of floral pattern + midi length + sleeveless that drove sales, thereby increasing the ability to sell the new item at full margin.

Artificial intelligence-powered computer vision gathers proxy products and assigns a similarity value to help retailers accurately forecast demand for brand new product lines. When fed back into the supply chain, this information helps replace wholesale pre-orders with a leaner, demand-led, made-to-order model that fuels product development with AI to launch lines that are guaranteed to sell out quickly.

The supply chain is only as robust as the data that feeds into it at the source. In retail, granular, customer-centered product attributes can help reduce mountains of unsold inventory in the warehouse, manage assortment complexity, clean up the product catalog, and use data analytics to identify the most profitable SKUs.

Purva Gupta is the co-founder and CEO of Lily AI, the first customer intent platform built to power the present and future of e-commerce.