Around the world, 820 million people don’t have access to the food they need, yet one third of the world’s food is wasted. Here in the U.S., the problems are just as stark — 30 to 40 percent of our food supply goes to waste, costing retailers $57 billion annually, while one in eight people in America don’t have enough to eat. This waste isn’t just bad for people and businesses, it’s terrible for our planet too. Enormous amounts of water, fuel, electricity and fertilizer, and human labor are poured into food that never gets eaten and the methane produced from rotting food in landfill is a potent and toxic greenhouse gas.
This was the complex set of challenges Project Delta, our early stage moonshot, set out to solve. Our team’s mission was to create a smarter food system — one that knows where the food is, what state it’s in, and where best to direct it to ensure it doesn’t end up in a landfill and instead goes to the people who need it most. After two and a half years of prototyping and testing a range of technologies to help reduce food waste and food insecurity, I’m pleased to share that some of our prototypes and team are moving to Google so we can scale up our work.
We kicked off our journey out in the field, talking with and learning from farmers, fisherfolk, grocers, food banks and more. One of the first things we discovered was that many distribution failures come down to data “silos”. Just as food sits in silos across the country, information about food also sits in “silos” in organizations. There’s no easy way for food suppliers to share information about their available food, or for food banks to register their needs. This means that a food bank in Texas might work with a grower in Florida for oranges when a grocery store 20 miles away could have donated exactly what they need. We also discovered there’s no standard way to communicate about food items, quantities or locations. In the data sets we looked at, there were 27(!) different ways to refer to Texas e.g. TX, Texas. Tx., NX, TXTX etc.
To learn more about these challenges we headed to Tucson, Arizona. We started working closely with Kroger, the largest traditional grocer in the United States, Feeding America™, the country’s largest domestic hunger-relief organization, and the Southwest Produce Cooperative, a team of food bankers from the Arizona Food Bank Network that help find outlets for donated food across the country. Tucson is about 60 miles north of the border town of Nogales, through which about three billion dollars worth of produce flows each year. It was in Tucson that we met Dana Yost, the COO of the Community Food Bank of Southern Arizona, one of Feeding America’s member food banks. Dana helped build the Southwest Produce Cooperative. Every day this group manually routes truckloads of food to various hunger relief organizations in the U.S. relying almost entirely on phone calls, site visits and years of experience as food bankers. Dana and his team inspired us — what if we could develop a software system that operates like Dana’s brain, but at a much bigger scale?
Members of the Community Food Bank of Southern Arizona team. From left to right: Efrain Trigueras, Dana Yost, Tomas Lopez and Ramiro Urtusuastegui
To make it easier for our food organization partners to understand each other’s data, we prototyped an intelligent food distribution system built on Google Cloud. We affectionately named it “dana-bot”, after our friend Dana. “Dana-bot” automatically uploaded information from the Southwest Produce Cooperative’s donated food dataset, categorized and standardized each entry, then matched the food with food banks and food pantries based on real-time needs in the Feeding America network.You can read more about how we developed our intelligent food distribution system on the Google Cloud blog — here and here.
As we developed dana-bot, we started working with Kroger to bring grocery surplus data into the system too. Our tools were able to offer Kroger and Feeding America insights into their processes and revealed opportunities they didn’t know were there. For example Kroger — who plans to eliminate waste in their stores by 2025 — used to recycle excess deli products due to food safety concerns. By giving the team greater visibility into the food being recycled, Kroger was able to change how they manage the deli surplus so they can now safely donate this food, opening up millions more meals to communities that need it. This is just one of many departments that this technology could be applied to across the store, and we’re excited to keep building on this work with Kroger.
In addition to our work with grocers and food banks, we explored the problem of food waste closer to home. To understand how waste is generated and tracked, and how new technologies might help reduce food waste in commercial kitchens, we spent time in Alphabet’s kitchens (prior to the pandemic). Here we cut vegetables, interviewed chefs and kitchen staff, and sorted through waste. After seeing how manual the process is to record food waste — taking 30–60 minutes per shift for kitchen teams to track by hand — we saw opportunities for technology to reduce waste and save the kitchen teams’ time.
Getting hands on experience measuring and sorting food waste in Alphabet’s kitchens
We developed a prototype food identification and categorization system that uses a mix of computer vision and machine learning tools to automatically capture imagery of what is being thrown out. After running a small pilot of 20 units in Alphabet’s cafes for six months, our system was able to automatically collect two times as much information about the kitchen’s food waste as the manual system. In the future, we see potential for this system to suggest smart, safe ways to donate or reuse food surplus, too. For example, if the system recognizes there is a lot of rice being thrown out, it could recommend turning it into a fried rice dish or a rice pudding based on other ingredients the system knows are in the kitchen or about to be thrown out.
Project Delta’s prototype food identification and categorization system uses machine learning to automatically identify different types of food. The different colors distinguish the grapefruits from the lettuces
The Delta team’s time at X was a learning adventure, where we met some of the extraordinary people and organizations who help feed people every day. Project Delta’s food identification and categorization system, and our intelligent food distribution system (aka “dana-bot”) were just two of the technologies we developed at the moonshot factory. Thanks to the progress we’ve seen with our partners, we’re ready for these prototypes to start tackling food waste and food insecurity on a larger scale. While some of the technologies my team and I developed will continue to be incubated at X, a subset of the Delta team is moving on from the rapid prototyping lab environment of X and over to Google.
Post-COVID when Googlers return to our offices, we plan to roll out our food identification and categorization system across more Alphabet kitchens. For now, our team will focus on making our intelligent food distribution system more robust so we can help more organizations reduce their food waste. If you’re an organization looking to find smarter ways to redistribute your surplus, or a food bank with unmet demands, please get in touch with our team here.
*To support the great work Feeding America and Kroger do to ensure food goes to people and not landfill considering donating to The Kroger Co. Zero Hunger | Zero Waste Foundation or Feeding America.