How Can Substitutions Be Improved In Online Grocery?

A university study has found that personalizing alternatives notably raises acceptance rates when an online grocer order requires a substitution due to an out-of-stock.

An accompanying survey of 25 international retailers found the solid majority of consumers (70%) would allow substitutions if their first choice is not available. In responding to a substitute situation due to the out-of-stock:

• Half of retailers surveyed rely on data based on the popularity or sales of similar items.
• Almost a third (31%) employ “decision trees or category-based logic,” such as replacing one brand of yogurt with another brand, or one size of pasta with a different size.
• Less than a third (31%) of the retailers surveyed leverage customer purchase history as well as machine learning, or artificial intelligence models, for personalization.

Partnering with a European online grocery platform similar to Instacart, the study — led by researchers at Argentina’s Universidad Torcuato Di Tella School of Business — came up with a system to identify optimal substitutes via personalization before the picking process began.

Historical basket data for each customer was used to build an algorithm that ranks customer preferences and suggests substitutes when the preferred products are stocked out.

The suggestions then weighed “stable preference patterns,” such as the unlikelihood that another dairy-based substitute would be accepted by a customer who frequently buys lactose-free milk despite being in the same category. A “contextual filter” was also added to ensure that suggested substitutes were “relevant,” or within the same price range or subcategory.

The study found that compared to the retailer’s existing machine learning system, the personalized and context-filtered suggestions led to a 26–32% increase in acceptance rate. The researchers wrote, “The biggest gains came from customers with few past transactions—traditionally the hardest group to personalize for. Importantly, a simple relevancy filter—such as only suggesting items within ±50% of the original item’s price—improved success rates by more than 12 percentage points.”

The study found substitutions costing retailers, on average, $2.66 in labor costs and lost margin for every $1 in recovered profit margin. The costs include delivery refunds for unwanted substitutes, the time in-store pickers spend searching for the desired product in the backroom or elsewhere, and/or time spent attempting to contact a customer to ask for the best replacement. Some retailers will also substitute a more expensive replacement without extra charges.

Poor substitutions were also found to be a leading driver of churn, even over late delivery.

Online Grocery Research Produces New Findings on Customer Preferences

Research led by London’s Leeds Business School around substitution techniques found:

• In vertical differentiation (VD) categories where products can be ordered easily according to their levels of quality or performance (i.e., margarine, ketchup), shoppers are more inclined to consider the alternative satisfactory if it is from the same brand.
• In horizontal differentiation (HD) categories that that cannot be categorized objectively, as they are generally differentiated by attributes such as flavor (i.e., cereal, pizza, potato chips), shoppers are more inclined to consider the alternative satisfactory if it is from the same brand.
• Consumers are generally much more liable to accept products they have purchased previously.

The study also indicated automation may enable retailers to contact a customer directly to ask for the best replacement instantly when in post-purchase, “opening the door to more options to satisfy customers.”