They say if you want to know the future, look to the past – but what if you don’t have a clear view of the past? And what, specifically, do you look at to inform the future?
This is the dilemma of ecommerce demand forecasting. There are so many channels where cash flow is generated – so many SKUs across multiple marketplaces – that it creates a situation where we have more data than we can handle.
This is where artificial intelligence and algorithms come in. Traditionally, what powers the algorithms are historical data signals, such as previous sales, number of customers and repeat purchases. In these instances, you’re looking at what you sold at the same time in the previous year.
The holiday season is a prime example of this forecasting dilemma. In Q4 of every year, sales explode across nearly every category – even ones you might not expect! To predict what your holiday sales will be this holiday season, brands typically look back at last year.
But as we all know, last year was an anomaly due to the global pandemic. It was a time of highly variable ups and downs, where there was a spike across all ecommerce categories throughout the course of the year, but extreme in some and medium in others, not to mention the location-level demand inconsistencies driven by geographic lockdowns.
The holiday season, too, had been marked by uncertainty around how much consumers would feel comfortable spending during a slow economic recovery. Many thought that despite the boost in ecommerce, retail spending would be down overall, yet when the 2020 numbers rolled in from the NRF, it was actually above expectations.
So if you only look at year-over-year, or seasonality, it doesn’t take into account the many events that have occurred over the past 365 days that have affected consumer behavior, which we now know (all too well) can shift quickly and either accelerate existing trends, or pivot in a moment.
To forecast accurately, you must also consider the more recent 30, 60 and 90 days. These signals will tell you more about where consumer behavior is going, as you can observe the trends leading up to the holiday season.
For instance, say in August you start to see an increased interest in a certain type of ingredient (e.g., sambucus, or elderberry) in products that boost your immune system. Typically you wouldn’t associate an immunity product as part of a holiday boost, even though it is the start of the winter cold season. However, with the recent spike in COVID-19 cases across the country, that product is starting to gain more traction, and if that trend continues, you’d need to anticipate higher sales in the coming months.
Another factor related to COVID cases is ecommerce vs. brick-and-mortar. Many retailers saw a slowdown in ecommerce sales in July as customers went back to shopping in stores. However, if cases rise and people go back to their laptops and mobile apps as their primary means for shopping, we’re going to see ecommerce demand rise again, and brands will need to ensure their products are available in the channels where consumers are shopping.
A lot of factors to consider, right? That’s why we use a method called ensemble forecasting. It’s what they use in weather forecasting, when there are always seasonality factors, like “this time last year,” but also recent data that shows current trends from the past few weeks. It generates multiple forecasts that produce a range of potential outcomes based on differences or perturbations applied to the data after it’s been incorporated into the model. In other words, each forecast compensates for a different set of uncertainties.
With this level of forecasting – 90% accuracy and higher – you can enter this holiday season feeling confident that you won’t run out of inventory.
Most digital commerce brands understand why that’s important at the most basic level. When you don’t have the product that consumers are searching for, you can’t sell it, right? So that’s a lost sale, and therefore lost revenue – not to mention a disappointed consumer!
But there are deeper implications than just that one lost sale.
When you’re out of stock, you lose the buy box on Amazon – that white box on the right side of the Amazon product detail page where customers can add items for purchase to their cart. Only businesses with excellent seller metrics stand a chance to win a share of that valuable real estate.
You’ve worked hard to win that box. When you’re out of stock, you lose the buy box, but you then also lose your search ranking. Amazon’s algorithm recognizes that your product is not available and will no longer put you on the first page, where 70% of customers will stay and never move past.
The downstream impact of out-of-stock is tremendous. If you’re out of stock just one week in, say, early December, it will impact the next week, the week after, and even well into January. To say it’s a sales momentum killer would be… more than 90% accurate.
So as we roll into August and you start making decisions about inventory, remember that it’s not just about creating accurate benchmarks to guide your marketing strategy and cash flow. Being in stock at critical moments, thanks to accurate demand forecasting, will ensure you maintain the search ranking you’ve worked so hard to achieve and continue the momentum you’ve built around that particular SKU, and your seller ranking, too.
And if you use Tradeswell, your CFO will thank you. With demand forecasting that integrates all data, including financial metrics, you get more than simply top-line sales predictions. When you know your margins at the SKU level across all digital marketplaces where that product is sold, you not only know what your Q4 will look like in terms of gross sales, but what your bottom line will look like, too.
Want to learn more about our platform and the Tradeswell advantage? Let’s set up a time to talk.