March 16, 2023

Meal kit company saves $1M/year by upgrading to Stratus Data’s demand forecasting algorithm

The Client is a leading meal kit company. A subsidiary of a Fortune 500 food retailer, they deliver to thousands of locations across America.

In 2020, the Client was predicting future demand by calculating simple averages of sales from recent weeks. In search of higher accuracy, Chief Product Officer (CPO) Scott P. hired Stratus Data to build a more accurate algorithm that was optimized at the store and product level. The upgrade resulted in $1,000,000 of savings in the first year alone.

 

Challenge

High-stakes supply decisions meet highly variable demand conditions

Meal kits present a special challenge for supply chain management. A product only has two possibilities: sell or waste. Extra food can’t be stored (like hardware) in a warehouse, or sold (like clothing) at a discount over time. Undersupplying is costly too: customers become frustrated and switch to competing brands when they can’t find what they need.

In addition to high stakes, supply chain leaders face uncertain conditions. Consumer demand is highly variable. Today’s sales provide limited hints as to what the store will need in a week. Some stores stock their coolers using LIFO (Last In, First Out), while others use FIFO (First In, First Out). The Client sought a data science team to uncover the relationship between current demand and future supply. They selected Stratus Data for excellence in performance, organization, and written communication.

“In data consulting, analytical know-how is table stakes. […] Storytelling is a blind spot for many specialists: taking something analytically complex and making it digestible.”
– Shane A., Sr. Manager of New Product Development

Solution

Brainstorming sessions result in creation of customized forecasting algorithm

Stratus Data began by forming a comprehensive understanding of the Client’s operations and unique business problems. Then, Stratus teammates led a series of brainstorms with the Client’s leadership. “We always started at a high level,” recalls Sr. Manager of New Product Development Shane A. “We would double-click deeper and deeper into a problem. It was easy to follow, and we built a fact pack that has become like company lore.”

From these brainstorms, Stratus developed an approach to the special forecasting challenge. They conducted a hypothesis-driven analysis, testing several factors that may or may not need to be captured by the algorithm. One such factor was ramp-up time, the time it takes for demand to ramp up when the Client introduces a new product to the market. Another was cannibalization, where a new product steals sales away from an existing product. There was also the question of whether a comparison between recent and historical sales could speak to future demand. All analysis was documented with care and continues to be referenced today.

The final algorithm had to find a middle ground between simplicity and complexity. The more complex, the more accurate—but also the slower the runtime, and less transparency to stakeholders. Stratus’ engineers brought the level of optimization ability required to strike the perfect balance. They began with a simple algorithm and added complexity as needed. To fine-tune the model, they tested it on historical data across all stores and SKUs.

Ultimately, the algorithm was integrated, in a joint effort between Stratus teammates and the Client’s engineers, into the Client’s existing code and workflow. Years later, it continues to be run regularly to inform supply decisions for upcoming weeks.

Results

Money saved, time saved, legacy forged

The Client cites a million dollars of savings a year: $500K from better shrink (food waste) management, $500K from improved revenue. The 5-minute runtime of Stratus’ algorithm, down from the 1-hour runtime of the Client’s existing algorithm, means employees no longer have to stay late on a Friday evening to wait for reports to generate. Leadership can use the forecasts to answer questions and plan more accurately. Such forecasts benefit common downstream supply chain challenges like:

  • From whom and how much raw materials should we source?
  • How should we manage labor to meet demand?
  • When should we make and ship the product so they arrive at stores on time?
  • How much inventory should we hold to minimize cost?
  • Where can we optimize distribution?

Newly acquired by the Fortune 500 food retailer at the time of the project, the Client was able to demonstrate confidence and build trust in the relationship. Today, the company has its own in-house data team—“but it’s still great,” says Shane A., “to have an asset like Stratus that knows our products, challenges, and main business objectives. We can tap into them when we hit roadblocks we don’t have internal capability to surpass.”

–

Send us a message to learn how your company can forecast future demand to support important supply decisions.