Forecasting Personas for Retail

Forecasting in retail and supply chain is rarely an end goal. Instead, it is a critical input to various downstream decision-making processes that support a retailer's supply chain operations. However, the entities to forecast on may significantly change based on the different usages. Two such scenarios are elaborated: 1) labor and resource planning for delivery operations and 2) inventory stocking and fulfillment decisions for omnichannel operations.

Labor and resource planning for delivery operations is critical for a retailer to deliver timely customer packages, including as fast as same- or next-day deliveries. These involve the distribution centers packing and dispatching the cartons, middle-mile carriers moving the cartons from the distribution centers to the regional delivery hubs, and finally, last-mile carriers delivering the cartons door to door. Each distribution center manager needs to complete labor planning for at least the upcoming four to six weeks. If a holiday or peak season is coming up, they must act even earlier to ensure the distribution centers can handle the volume surges. Each middle-mile operations manager needs to ensure that an appropriate number of long-haul trucks are available daily to move the cartons from a handful of distribution centers to the hundreds of regional delivery hubs. Finally, each regional delivery manager needs to plan for enough drivers daily to cater to all the ZIP codes in that region, ensuring the timely delivery of the packages. These logistics operations are supported by outbound carton forecasting, which forecasts the daily carton volume that will flow from each distribution center to each regional delivery hub, and each delivery hub to each ZIP code, as well as the volume that will flow through each middle-mile and last-mile delivery carrier. The data needs are historical daily time series of carton volumes across various distribution centers, delivery hubs and delivery carriers.

Inventory stocking and fulfillment decisions for omnichannel operations are critical for a retailer to meet customer demand within a reasonable time and minimal cost. These involve deciding the inventory quantity of each item to stock at each store and fulfillment center at any given time. In the omnichannel retail world, a customer can pay online and pick up at a store, or the same customer can pay online and request home delivery. It might be cost-efficient for the retailer to have the item in stock at the store from which the customer wants to pick it up, whereas for home delivery, the item can be delivered directly from any fulfillment center or store. The inventory managers need to consider various factors and costs to make these decisions. Some of those are inventory holding costs at each fulfillment center and store, delivery time and delivery costs from each fulfillment center and store to each ZIP code, and flexibility of the delivery network to fulfill customer orders in a timely manner if the order delivery mode is different than expected. These inventory stocking and fulfillment decisions are supported by fulfillment forecasting, which forecasts the daily demand of each item originating from each ZIP code based on customer locations broken down by payment channel (online payment, store payment) and delivery mode (home delivery, store pickup, store purchase). The data needs are historical daily time series of each order’s list of items, payment channel, delivery mode and customer ZIP code.

Understanding the entities and data needs of the specific forecasting problem to properly serve the downstream usage is at least as important as choosing the right forecasting algorithm. Forecasting on the appropriate entities at the necessary granularity level is already half the job done for the related decision-making processes.