AI is cutting grocery waste by a quarter, and the fix isn't sexy tech—it's replacing spreadsheets with models that know when your strawberries are leaking water.
The Summary
- Afresh, an AI startup, just raised $34 million to expand software that cuts grocery waste by up to 25% by fixing fresh food ordering.
- Four million tons of food waste annually in U.S. grocers, worth $27 billion, because produce managers still use pen, paper, and guesswork.
- The underlying problem isn't complex algorithms—it's that grocers literally can't count what's in their stores accurately enough to order right.
The Signal
The grocery waste problem is a data collection problem disguised as a forecasting problem. Afresh's founders discovered this when they walked into stores a decade ago and found produce managers writing orders on spreadsheets. Not digital spreadsheets. Printed ones. While packaged goods had basic inventory software, fresh food—the high-margin, high-waste category—ran on educated guesses and muscle memory.
The invisible complexity shows up in the details. A strawberry container loses weight as the fruit dehydrates. A customer at self-checkout rings up organic apples as conventional. A package of salmon goes bad and gets tossed, but no one logs it accurately. These aren't edge cases. They're the baseline reality of fresh food retail.
"It was ultimately a pen and paper process."
Afresh's solution starts by building a real-time model of what's actually in the store, factoring in perishability variables most systems ignore. The software ingests hundreds of billions of transactions per grocer, tracking pricing, promotions, shipping origins, and product-specific decay rates. Then deep learning models forecast demand using signals like food stamp distribution timing and weather patterns. An optimization layer suggests order quantities.
The 25% waste reduction isn't theoretical. Chains test in 10 to 20 stores, compare performance against control groups, then roll out system-wide. The models learn continuously, getting sharper as they process more spoilage events, stockouts, and purchasing patterns. This is narrow AI doing exactly what it should: solving a specific, high-value problem that humans can't handle at scale because the data environment is too noisy and the variables too numerous.
Key operational wins:
- Real inventory visibility for products that literally change weight on the shelf
- Demand forecasting that accounts for external factors (weather, benefits timing) humans can't track manually
- Continuous learning from actual waste and sales data, not static rules
The $34 million round, co-led by Just Climate and High Sage Ventures, signals investor confidence that this isn't a one-trick play. Fresh food is the hardest category to automate because it's the most variable. If the models work here, they'll work everywhere grocers have similar opacity problems—prepared foods, bakery, deli. The broader pattern matters: AI wins when it solves data collection first, prediction second.
The Implication
This is what useful AI looks like in 2026. Not chatbots or generative hype, but agents that handle the grunt work humans hate and can't scale. Grocers get better margins and less waste. Customers get fresher food. The model doesn't replace the produce manager—it gives them a tool that actually works.
Watch for Afresh-style solutions in other industries where inventory meets perishability: pharmaceuticals, cut flowers, restaurant supply chains. The playbook is clear now. Find a category where humans are guessing because the data is bad. Fix the data layer. Build models on top. Scale from there.