Towards a Sustainable Food Supply Chain Powered by Artificial Intelligence
At Afresh, we are building the next-generation food supply chain powered by AI, reducing waste and making fresh, healthy food accessible to all. Every year, one-third of food produced goes to waste, globally. In the United States, 40 percent of all food waste occurs at the retail and consumer level, with the highest occurrence in fresh food departments. Our goal is to apply technology to optimize the supply chain and make fresh food more accessible.
In the United States, between 30–50% of all the fresh food grown at the farm ends up being wasted (Gunders, 2012). This includes up to 38% of all grain products, 50% of seafood products and 52% of fruits and vegetables, representing a $408B loss to the US economy. The worldwide economic impact of food waste is comparable to the gross domestic product of Turkey or Switzerland.
At the same time, the United Nations estimates, approximately one person in nine in the world suffers from hunger. Hunger is estimated to kill a greater number of people every day than AIDS, malaria and tuberculosis. From a moral perspective, we have an imperative to distribute food in a way that satisfies basic human needs.
Impact of Food Waste on the Environment and the Climate
Food waste is also an enormous problem for the environment, contributing to 21% of all freshwater use, 19% of all fertilizer use, consumes 18% of available cropland, and occupies 21% of total landfill volume .
This misuse of resources in turn drives massive amounts of greenhouse gas emissions, which have a significant impact on climate change. It is estimated that food waste contributes to up to 25% of all greenhouse gas emissions (Vermeulen et al. (2012)). According to a study by the Food and Agriculture Organization, food waste generates greenhouse gas emissions comparable to those of Russia.
The impact of reducing food waste would be immense. A 25% reduction to U.S. food waste would conserve 83B gallons of water per year, recover 90M meals per year, and eliminate 10M+ tons of greenhouse gas emissions per year.
Fighting Food Waste Using AI-Powered Supply Chain Automation
About 80% of US food losses occur at consumer-facing businesses — primarily in restaurants and supermarkets — and downstream in homes. Afresh develops technologies that significantly cut down this waste by automating supply chain decisions using artificial intelligence algorithms.
Our first product is a decision support system for supermarket operators that optimizes operational decisions and minimizes losses; the core of this system is a model-based engine for perishable inventory control. The crux of our approach is to construct a probabilistic model of the future (predicting demand, shipment times, etc.), and compute optimal decisions that minimize waste under this model.
Below are some of the real-world challenges involved in deploying such systems as well as machine learning techniques to address these challenges.
Massive Datasets + Deep Learning. A regional grocery chain has hundreds of stores, each generating daily time series for thousands of items over the course of years. This represents a dataset of 10–100M datapoints and requires algorithms that scale to this data. At Afresh, we use deep learning to effectively leverage such massive datasets and maximize predictive performance.
Rare Events + Multi-Task and Few-Shot Learning. Although each product may have years of historical data, certain rare events (such as holidays) are only seen once a year (hence, rarely). Handling these rare events requires joint multi-task learning across thousands of items and across hundreds of related time series.
Uncertainty Estimation + Probabilistic and Bayesian Methods. Accurate planning requires predicting not just a point forecast, but an entire distribution over model demand. In addition to enabling more accurate planning, probabilistic predictions are also a key component of interactive systems that can assess their confidence before making recommendations to a human operator.
Minimizing Waste and Stockouts + Planning Algorithms. Given probabilistic demand forecasts, we compute inventory decisions that balance minimizing waste and limiting out-of-stocks in a framework inspired by Model Predictive Control (MPC).
Keeping Humans in the Loop. Finally, in order to be useful, recommendations need to be surfaced to store operators along with confidence levels that correlate well with their performance.
Afresh in the Real World
As of 2021, our technology has been deployed across hundreds of supermarkets at multiple US grocery chains. By examining historical data, we can measure their historical levels of waste, and estimate the level of improvement offered by our system.
Across our engagements with grocery chains, we observed reductions in food waste of up to 50%, while also seeing sales increase due to reduced stockouts. These pilots were structured as A/B tests and in which our decision support system was measured over a number of stores for several months. We measured waste reduction numbers in pilot stores and compared to a baseline of all the stores in the chain during the same period.
As we continue to expand our customer base, we expect to have a significant impact on reducing food waste at the retail level.
We are growing our team rapidly and are looking for passionate, talented engineers to help us solve more problems like this. If you are interested in joining, take a look at our current openings.