UMass Amherst uses AI to cut food waste and lower disposal costs
UMass Amherst is pairing two AI systems to attack waste from both ends: before food hits the plate and after it lands in the bin. University leaders estimate the approach could save $1 million+ as it scales across campus.
The driver is simple: contaminated waste dilutes recycling, raises fees, and turns good material into trash. Add overproduction in dining, and you get cost without value.
The core problem: contamination and overproduction
Contamination happens when trash, recycling, and compost mix. Much of what gets collected as recycling is so mixed it can't be processed. On a campus this size, that adds up fast.
UMass pays $126.79/ton for trash and $94.64/ton for recycling - a $32.15/ton gap. In the last reporting year, the university generated about 1,442 tons of recyclables and 3,073 tons of trash. Leaders say contamination rates have topped 50% in the past.
MetaFoodX: reduce overproduction before it starts
MetaFoodX uses AI scanners to weigh trays before and after meals. The system shows what students actually eat, by dish and by time, so dining teams produce the right amount at the right moments.
That data feeds procurement. As Dining's sustainability lead explained, once the system covers enough service, ordering can be tightened to match real demand. The department projects up to $1 million in savings if run at full capacity by making and buying less food.
UMass piloted the scanners in fall 2024 and has since implemented them at Harvest Market and all four Amherst campus dining halls. Procurement adjustments will build over time as patterns stabilize.
rStream: clean up the waste stream after the meal
rStream, founded by two UMass alumni, tackles the back end. Staff bring waste barrels to the Office of Waste Management, where a mobile sorting machine - hauled by trailer - uses a camera and AI vision to identify each item and sort it into recycling, compost, or trash. The model learns continuously from each run.
The impact is financial and operational. If a system sorted 500 tons of recyclables out of trash from several residence halls, UMass would save about $16,075 on tipping fees alone (500 x $32.15/ton). That's just one slice of campus waste.
The tech is still in the experimental phase. rStream currently tests on campus once or twice a semester, sorting material from multiple buildings. For the most recent visit, Dining and Facilities each paid $5,000. The founders expect "one or two loops around the sun" to gather enough data to engineer a permanent setup, and they're still working through how best to deploy across a sprawling campus.
Beyond cost avoidance: potential new revenue
With cleaner streams, the university could selectively capture materials with resale potential. Think sorted UMass-branded plastics repurposed into recycled clothing or other goods for the campus store. It's early, but cleaner inputs create options.
What managers can take from this
- Instrument demand at the point of consumption. Weigh, track, and forecast by dish and time to adjust production - then lock those insights into purchasing.
- Quantify the contamination tax. Know your tipping fee gap (here: $32.15/ton) and set explicit targets for contamination reduction by building or route.
- Pilot with clear KPIs. Start with a few high-volume locations, measure overproduction, contamination rate, and dollars saved per ton, then expand.
- Design for your logistics. Mobile sorting can bridge gaps while permanent installs are engineered - especially across distributed facilities.
- Explore secondary value. Cleaner, separated streams open doors for resale and branded upcycling if volumes are consistent.
- Budget for iteration. Expect phased rollouts, vendor learning curves, and operational tweaks before full-scale procurement changes stick.
Where to learn more
To understand why contamination drives cost and how to prevent it, see the U.S. EPA guidance on reducing recycling contamination: EPA resource.
If you're upskilling teams on AI for operations and facilities, browse role-specific training: AI courses by job.
The bottom line
MetaFoodX trims overproduction before food is made. rStream recovers value after it's tossed. Pairing both gives UMass Amherst a practical path to lower costs, cleaner waste streams, and future revenue options - with clear metrics at every step.
Your membership also unlocks:
