It’s a toy story.
The holiday season is a massively challenging time for retailers. Determining what the customers will want and what they will buy is a difficult task year-round. Still, it becomes even more complicated when the volumes of products are much higher while their lifecycle is shorter and when a large part of their revenue is at stake.
For most retailers, the holiday planning process starts as the new year begins. As the traditional markdowns pop up in the stores, the financial forecasts are locked in at the headquarters, and the marketing and procurement teams dive into their numbers to understand which categories of products will sell more or less next season.
The ultimate question is to understand what the shoppers will buy to please the impossible-to-buy-for niece, what video game will satisfy the younger cousin, and what partners will buy for each other. It means knowing which movie or tv show will be viral with the kids: Will dolls sell more than Barbies? Will Anna eventually be overtaken by the Little Mermaid? Will Paw Patrol merchandise remain a big hit?
‘Tis the season, and these are the questions!
Demand forecasting consists of determining what your customer will want to buy. The first step to estimating this is what most retailers already do, such as historical sales. This approach is definitely a solid baseline for establishing a forecast. However, by definition, POS data and other sales information are the results of what the customers bought, not what they wanted. Did a shortage of one product push them to another one? Or perhaps, did a competitor have an online promotion?
On top of these ‘standard’ events that impact sales for retailers, the last couple of years clearly demonstrated that external factors have a tremendous impact on traffic and consumer behaviors. The 2021 holiday sales were up 16.1% versus the same period in 2020, and as everyone knows, 2020 is no reference. This is why now more than ever, retailers need to enhance their forecasting process by leveraging internal and external data, historical and predictive, and machine learning (ML) algorithms to better predict demand.
An algorithm is a method, or an automated instruction, that uses input data to predict an output. ML algorithms, part of Artificial Intelligence (AI), use data to predict an output. Still, they also use the data to self-train, ‘learn’ from situations and improve their performances.
In a world where data is everywhere, retailers face two main issues. The diversity of data sources and the inconsistency of its quality based on the different systems make it difficult to ingest it into a unique platform in an organized and understandable manner. The second issue is that analyzing the impact of each data set on the sales for each category, each SKU, and each channel makes it an impossible task.
Enter ML algorithms. These game-changers can ingest data from various systems and software in a different format and standardize and cluster it. Using a process called ‘Feature Engineering,’ they also diversify the data by creating new variables to better understand what exactly impacts the volume of product sold. For instance, they make a new set of values such as temperature deviation from previous years, in percentage, and absolute number from the temperature data. By matching this information with the sales, they can understand what element in the weather impacts the sales and measure this impact. The measure of the effects coupled with forward-looking internal (price, promotion calendar, weather, etc.) and external data (movie release, inflation, social media, etc.) is run through the algorithm to generate the forecast.
Forecasting is only the first step in an intricate planning process. During the holiday season, the complexity of the planning process for the holiday peaks is only equaled by the volatility of consumer preference, and the 2022 season will be no exception. Dominated by inflation and an unstable political environment, retailers must be more agile than ever to ensure that the shelves will be full and the prices low as consumers are expected to prioritize availability over retailer loyalty.
