How AI Can Help Retailers and CPG Firms Create Magic Moments on Easter
Easter and magic moments go hand in hand. Moments overflowing with chocolate bunnies and hard-boiled eggs painted vividly with bright colors. Moments that create memories. But creating those moments isn’t always such a magical experience for the retailers and suppliers that need to collaborate to get those eggs, chocolate bunnies, cake pops, buttercreams, and other goodies to families across the United States. Why? Because Easter is a moving target.
Easter 2022 falls on April 17. Easter 2021 fell on April 4. Easter 2020? That happened on April 12. The only constant for retailers and CPGs? They know the date will change. And a changing date creates challenges that can sabotage those magic moments if handled wrong – challenges ranging from demand forecasting to pricing.
We believe that the answer is to apply artificial intelligence (AI) to crunch all the variables that affect demand for Easter goodies, and, by consequence, decisions about pricing and stocking. That’s exactly what we’ve done for Easter, as well as other holidays.
We’ve developed a forecasting model using Meerkat. For context: Meerkat is our all-in-one demand, pricing, and promotions planning cloud platform, Meerkat gives retailers and CPG firms the ability to deliver the right product, at the right moment, and at the right price while simultaneously delivering financial value to their stakeholders.
Now let us show you how we went about it:
Step 1: Getting the Right Data
Data is the fuel to any composite AI engine for CPGers and retailers (including our own AI engine). There are three main ingredients to consider:
- Historical point of-sale (PoS) sell-out data, including competitors if possible.
- Promotional data: past, present, and future (typically from trade promotion management platforms).
- Logistics data (demand, in-stocks).
From a granularity standpoint, AI needs to help retailers and CPGers get down to at least an SKU level for all the information on a weekly basis.
From a historical standpoint, how far in the past we model is based on domain knowledge and some preliminary analysis. Some products will be discontinued over time, and new flavors or products will also be launched. Using AI to forecast demand means taking the data only relevant to one’s present items so that the demand forecast will be accurate.
Our own AI model for Easter means that after we catalog a brand’s existing datasets, we will have a baseline forecast and performance assessment. We will then look into adding extra information (which Meerkat has built-in), such as:
- Decoration sales.
- Church attendance.
- Online searches.
- Traffic distributions.
- Bureau of Labor statistics such as inflation and the Consumer Consumer Price Index.
- Stringency Index, COVID dummy variables.
- Stimulus packages such as SNAP benefits.
- Gas price and spending patterns.
- Hotel bookings.
- Flight ticket bookings.
Step 2: Understanding the Holidays and Power Windows (Univariate/Multivariate Analysis)
Now that we have the data in place, we take a look at holidays, and in this instance the curious case of Easter.
We first label and check all occurrences of holidays and special events, both in the past (back as far as the historical data go) and in the future (out as far as the forecast is being made).
We can include columns lower_window and upper_window which extend the holiday effect to [lower_window, upper_window] days/weeks around the date.
The prepared seasonal calendar file or dataset will then be passed to the model, a one-hot encoding of holidays/special events for each week. All the holiday dates can be fetched from python packages directly.
Sample block of seasonality Calendar dataset.
However, in some instances, seasonality may depend on other factors, such as a seasonal pattern that is different during the summer than during the rest of the year or a daily seasonal pattern that is different on weekends vs. weekdays. These types of seasonalities can be modeled using conditional seasonality.
The default weekly seasonality assumes that the pattern of weekly seasonality is the same throughout the year. Still, we’d expect the pattern of weekly seasonality to be different during the on-season and the off-season. We can use conditional seasonality to construct separate on-season and off-season weekly seasonality.
Example of conditional seasonal features that can be created so that sales effect will be observed clearly.
Step 3: Modeling Easter
Many simple time series models only use the historical values of the time series. Suppose Easter falls in April in two consecutive years. In that case, the models expect the next peak again in April because the model has only learned the seasonal effect, leading to these Easter forecasting challenges.
There will be an increase in chocolate sales leading up to Easter, which is inherent in the underlying holiday. This type of event is synonymous with the due date of a pregnancy, and the key indicators we examine to discern level of demand for Easter merchandise. Since both Easter and a pregnancy fall around a given date +/- a few weeks, the model will learn to pick up on variables such core medical milestones, an increase in chocolate sales, or any feature that explains a similar phenomenon.
Also, the moving date of Easter matters because this holiday has a powerful influence on many leisure activities. Many people use the feast of Easter to take a spring holiday, which has a substantial impact on the hotel, restaurant, and transportation industries. Which will have a cross-selling effect on our product; however, we will not know which features are most important until prototyping. Some features we can use for building the model include:
- Stimulus packages, gas spending, and other custom datasets will get this information into the model.
- Church attendance polls.
- Governmental data and emerging markets.
- We can also look at the future and past Easter calendars, including Paschal full moon information.
Step 4: Future-Proofing the Models
Meerkat’s Composite AI Engine’s Automatic model selection checks for a brand whether the time series can be forecast with a simple model or whether a more robust model should be used (Meta-learning). Individual data is analyzed to select the most appropriate model from a model library and fit it to a brand’s data.
The forecast considers input variables like the number of workdays, Saturdays, or Sundays, and past sales (If forecast at the day level). That is especially relevant for power windows and holidays.
This is where the right resource can help you. To overcome forecasting and investing challenges, businesses are turning to platforms that manage AI, first-party data, and third-party data cost-effectively. That’s why we created Meerkat.
Meerkat is an all-in-one demand, pricing, and promotions planning platform from Pactera EDGE that gives users the ability to deliver the right product, at the right moment, at the right price while simultaneously delivering financial value for your stakeholders.
Meerkat leverages state-of-the-art, pre-trained AI models that are fueled by your own company data, alongside a broad range from our extensive data partner ecosystem, to solve challenges such as forecasting and demand planning, pricing, and promotion planning.
Our solution allows organizations to jumpstart better decision-making by seamlessly integrating with existing enterprise platforms. Not only does this allow you to leverage and maximize ROI on past investments, but you are also able to scale rapidly, regardless of your organization's readiness level.
Want to create magic at Easter – every year? To learn more, contact Pactera EDGE.