Food Waste No More: Machine Learning Helps Grocers Reinvent the Supply Chain

Scott Brady
Innovation Endeavors
4 min readSep 16, 2019


Announcing our Investment in Afresh

By: Scott Brady and Andy Triedman

Afresh Founding Team: Matt Schwartz, Nathan Fenner, and Volodymyr Kuleshov

In 2014, Innovation Endeavors started Farm2050, an ecosystem of industry leaders throughout the food and agriculture value chain, to tackle the immense and existential problem of feeding the world’s population of 10 billion by 2050. How can we drastically increase food production when we’ve already got all the land this earth will offer us (in fact, the world has lost about 33% of its arable land over the past 40 years)? In emerging markets, where there is a relatively lower adoption of modern agricultural technologies, we can try to boost productivity with fertilizers, irrigation systems, and crop protection products. While certain crops could see additional productivity gains from being grown vertically indoors, as is the vision of Innovation Endeavors portfolio company Plenty, others are likely near their full potential production.

We might not actually need to grow much more food to feed a burgeoning population. Instead, we might simply reduce waste, which currently makes up about 40% of all the food we grow. At which point in the supply chain this waste is created varies by geography. In emerging markets, waste is more likely to occur on the field, during harvest, or in transportation, but in countries like the United States, over 60% of food waste occurs at the end consumer.

What causes so much food waste? Sure, our restaurants should serve smaller portions and you should eat that last bite of pizza crust (…or should you?). But perhaps a more significant driver of waste, particularly for fresh food, comes from the supply chain. Supply chains for fruits, vegetables, meats, and dairy are often long, slow, and inefficient; a tray of berries might have had only a week of edible life when they hit store shelves leaving only a day or two by the time they land in your refrigerator. With such little time to work with, it’s no wonder that we throw so much food away.

Before you go rushing to your nearest grocery store with a pitchfork, know that they want to solve this problem too, maybe even more than you do. A typical supermarket operates with razor-thin margins — on average around 2.5% in net profit, one of the ten least profitable industries in the US — and the value of the food they have to throw away is often equivalent to or greater than their entire profits. Unfortunately, it’s not an easy fix. Predicting supply and demand is already a challenging task, and in fresh food, the forecasting problem is orders of magnitude harder. Fresh food often doesn’t have bar codes so it is frequently mis-scanned, is measured by weight but loses mass due to water evaporating, and might get tossed without record if it falls onto the ground (or into a devious shopper’s mouth). Every morning, in every grocery store in the country, a store employee walks the aisles with a clipboard and pen and struggles with these problems — eyeballing, approximating, and guessing how much of each item to order.

When Matt, Nathan, and Volodymyr met at the Stanford Graduate School of Business, they hoped that modern machine learning tools might offer a better solution to this wasteful and costly problem. They set out to research the opportunity in an independent study project sponsored by Scott and, after finding success, formed Afresh. Innovation Endeavors participated in the company’s Seed round and today we are thrilled to publicly announce that we have doubled down to lead their Series A fundraise and accelerate the company into broader deployment.

Afresh plugs into supermarkets’ inventory and POS systems, as well as external data sources, to assemble a detailed understanding of fresh food moving into and out of each store. They then leverage cutting-edge, reinforcement learning-based machine learning algorithms, factoring in supply chain and logistical constraints, to recommend the precise quantity of each item a store should order each day. Store clerks follow their same workflow as always, but with a tablet instead of a clipboard, and order boxes that magically autofill with suggestions (that clerks can override). The results? For the clerk, less time spent ticking boxes and more time on the floor with customers. For the store, less waste (a 25–45% reduction for initial customers) and fewer stock-outs, meaning a massive impact on the bottom line. And for consumers, fresher food that will taste better and last longer on the shelf or in the fridge. With such a meaningful effect on both supermarket economics and environmental impact, Afresh has seen rapid traction, currently deploying in three major grocers across the country.

In the longer term, Afresh has its sights set not just on grocery store shelves, but on all the players that handle food — distributors, producers, restaurants, and more. By deploying its advanced machine learning platform broadly, Afresh aims to help the entire value chain be less wasteful and more profitable while providing healthier and fresher food to all.

We’re thrilled to partner with Afresh to solve some of the most costly and wasteful problems that face our food system. If you’re another team leveraging technology to drive transformational efficiencies in supply chains, we’d love to hear from you!