Whether you live in a country that celebrates Thanksgiving or not, we can all relate to having appreciation for what we have, particularly when it comes to an uninterrupted supply of food. Whether it’s on holidays or just during gathering with friends, I often find myself at the dinner table contemplating the incredible spread before me and thinking Wow, I’m very fortunate. But then, likely because I work for a company that enables the grocery industry to serve its customers as efficiently as possible, I find myself thinking about the journey our food takes from farm to fork (I’ve come to learn that it really is remarkable). Take the Thanksgiving turkey, for example. How did my local grocery store estimate how many would be purchased in my store, not to mention all its other stores across the region? Not just that, they’re also competing with so many other retailers in the region who demand the same product, at the same time. In retail, the discipline that wrestles with this problem is called demand forecasting. And, until recently, it was much more art than science.
Frozen vs. fresh – a tale of two poultries
Let’s revisit the noble turkey. When I started to consider this blog, I looked up turkey statistics and spoke to one of our experts on supply chain. When it comes to forecasting for turkey purchases for Thanksgiving in the U.S., it’s interesting to note that the majority of the multi-millions of turkeys purchased are harvested and frozen throughout the year, and in some cases, several years prior to sale. Because of this, a retailer may not worry so much if it overstocks the popular bird. Sure, overstocks take floor space, but since the turkeys are frozen, they can be sold at Christmas and kept in stock longer and then put on promotion as a lever to bring customers into the store. An out-of-stock would be more significant because of missed sales and disappointed customers.
The challenge for demand forecasting? Getting it right for the multiple millions of fresh turkeys consumed at Thanksgiving – turkeys that can’t be kept and sold later.
Demand forecasting for seasonal fresh items
When a retailer considers demand forecasting for fresh turkeys, it must place its order at least six months before Thanksgiving. This enables suppliers the minimum time to ensure there are enough eggs available in the spring to be carefully incubated, raised and shipped, resulting in enough fresh birds in store for Thanksgiving. And, the timing of stocking the fresh birds at precise levels is critical too, as retailers need to meet demand, but don’t want too many turkeys sitting around that may spoil. Retailers are often sharing the forecast with their traditional/larger suppliers and smaller local suppliers. This sharing of forecast hopefully leads to better service levels from suppliers, which, in turn, leads to customer satisfaction at the shelf. It’s all connected, and a beautiful thing – when it goes smoothly.
Getting the forecasting equation right is difficult with traditional systems
Here’s what I’ve learned recently through working with experts in supply chain and demand forecasting. Traditional methods of forecasting often miss the mark when it comes to fresh items. This is because of the many variables at play, to name a few: the complexity of paths to purchase these days, weather complications, or other “events” that interfere with the growth, manufacturing, shipping of goods. When forecasting professionals attempt to compensate for these factors, they must often manually intervene in systems that are not made to handle that intervention. As a result, expensive errors can occur.
For example: A retailer overstocks cantaloupe. As its shelf life diminishes, the store reduces the price to sell, and customers buy all of the remaining melons. In a traditional forecasting system that lacks the power of AI or machine learning, which can easily assess the impact of the markdown on future demand for the melons, the forecasting professional must go into the system and note that the sales were a result of a markdown. If this isn’t done correctly, or at all, the supply chain thinks it should order more – perpetuating the cycle of overage and waste.
AI-enabled demand forecasting systems make life easier
Artificial intelligence is a game-changer in the forecasting space. AI-based demand forecasting systems use machine learning, which learns exponentially faster and more broadly than a human could ever do. Machine learning algorithms rapidly detect patterns and anomalies that often elude teams of data scientists, giving users immediate insights from the data and suggesting next-best actions based on those insights, not on gut-level hunches too often enlisted by demand professionals. These systems are truly a sea change for demand forecasting practitioners. They free them from tedious manual work to focus on more strategic initiatives and, along with the technology, contribute to a smarter, more efficient supply chain. One that has vastly reduced errors, waste and increases customer satisfaction resulting from product availability.
It’s not just about turkeys
I don’t want to alienate my vegetarian friends here (even though that Tofurkey you served last year still gives me nightmares). It’s not just about turkey at Thanksgiving, it’s about forecasting for events that are highly complex due to a multitude of factors. Factors far too complicated for complete human understanding (at least in a short time). So, whether it’s fresh items for Thanksgiving like turnips or sweet potatoes, onions, cranberries, or items like seafood and ready-to-eat meals throughout the year, the job of accurately forecasting demand for Fresh categories is one that requires the assistance of AI and machine learning.
A happy Thanksgiving to all who celebrate it. But for all of us, wherever we are, I hope you share my sense of appreciation for the complexities of our food chain and the technological advancements that help get food where it needs to be with minimum waste and cost, and maximum health and enjoyment for customers.
Learn more – Ready our viewpoint paper Improving Forecasting Accuracy in an Age of Retail Disruption