Demand Forecasting: Using Data to Fight Food Waste

Overproduction is a major cause of food waste. However, a lot of kitchens, understandably, want to be as prepared as possible when higher demand is anticipated. To avoid running into disappointed customers, many will often order more inventory than is necessary. Our Internal Decision Science and UX Research teams are on a mission to prevent food waste while ensuring happy guests and operators.

Getting Started

In general, when a kitchen orders ingredients and estimates production quantities, these decisions are based on experience. With several considerations involved in predicting whether demand is going to be low or high on a particular day—including weather and city events—it is necessary to observe a variety of kitchens in different locations. Therefore, the team started off by monitoring 14 pilot cafés from the Business & Industry sector across North America.

Challenges & Considerations

As the forecasting model continues to evolve, there are three main features that are being focused on to ensure the project’s success:

  1. Accuracy: It’s important to ensure the reliability of the algorithm’s recommendations, and determine a margin of error that minimizes food waste while preventing underpredicting
  2. Flexibility: To best assist kitchens with planning, the model must be able to predict at least three days out 
  3. Scalability: In order to have the most positive impact on the business, the solution—which works to identify patterns and memorize external variables—needs to be adaptable across 40k+ cafés

What’s Next

The team has optimized the algorithm and applied their latest learnings to ensure the project’s success as it expands. So far, the team has:

  1. Shifted from single observation to predicting a demand range; this is due to particular menu items, such as soup, leading to more overproduction than others 
  2. Automated the model selection process through machine learning, which has included factoring in potential losses to avoid under-predicting
  3. Improved customization of the model to suit particular kitchen and customer needs

With demand forecasting and procurement planning, there is hope that kitchens will see a significant reduction in both costs and food waste; creating a lasting impact that could positively affect our business, our communities, and our environment.