Not being able to deliver the items that customers demand when they demand them has been a long-time struggle for retailers. In online grocery retail, this issue is magnified. There is often a delay from the time an order is made until it is picked in the store, which increases the risk of not fulfilling the customer’s order.
Consequences? Customers commonly receive orders where products have been substituted, where products are only partially fulfilled, or where products simply have been removed from the order.
Obviously, partial fulfillment and substitutions help the retailer control lost sales at the moment of purchase, but also negatively impacts the customer’s future purchase behavior.
Using data from the online operation of a large grocery retailer, a group of researchers (including LTPlabs’ partner Pedro Amorim) evaluate the impact of failure on future purchases. The results show that while failure has a minimal impact on churn, it severely impacts both the time until the customer makes the next purchase and the basket value of that purchase.
For each failed item, customers wait a quarter of a day longer for their next purchase, and this purchase has a basket value that is 0.7% smaller.
The combined effect is that failures have a substantial impact on the annual revenues per customer. The simulation made by the researchers shows that if the retailer increased the average service level from 91.5% to 93%, the average annual revenue per customer would increase by 100 euros (4.6%).
The impact on future purchase behavior does, however, depend on the type of product that fails as well as on which customer experiences failure. For instance, customers seem to be more tolerant towards failures for perishable products. On the other hand, customers with a more extended purchase history with the retailer are less sensitive to failures than first-time shoppers.
Although avoiding the failure is impossible, the retailer may mitigate its impacts. Substitution has a larger negative impact on future purchase behavior than partial fulfillment does. However, this small difference is offset by the larger cost of lost sales associated with not substituting. These results support the idea that retailers should keep improving their substitution policies and processes while proactively avoiding them.
These insights can be translated to operations policies that differentiate service levels for specific products or customers. It’s clear that slight differences can have a substantial impact on total revenues.
While these policies require proper system support to be put in operation at scale, the change in inventory cost is minimal because the average service level is kept constant. Using algorithms to perform these operations represents a huge opportunity for online retailers.
By: Pedro Amorim