Short answer: spent grain is the biggest thing leaving a brewery that most breweries never properly cost. A brewery produces roughly 20 kg of wet spent grain per hectolitre of beer — about 85% of all its by-product by mass — and at ~80% moisture it spoils within days. The smart move isn’t one fixed answer; it’s using your own data to route each batch to its highest-value use (feed, food, or fuel) net of energy and haulage. AI helps with that allocation. It can’t invent a market that isn’t there.
The by-product nobody costs properly
Walk any brewhouse after mash-out and you’ll see the real output of the day: tonnes of warm, wet grain. For most breweries it’s handled as a chore — given away or sold cheaply to a local farmer as cattle feed, occasionally landfilled when no one collects it. That’s understandable. Wet brewers’ spent grain (BSG) is about 80% water, heavy to move, and microbially unstable: left a few days it heats, sours, and becomes a disposal problem rather than a feed.
But the largest by-product on site deserves more than a shrug. BSG is high in protein (around 20-25% on a dry basis) and fibre, which is why its uses run well beyond the farm gate. The question isn’t whether it has value — it’s whether your volumes, locations, and processing options make the higher-value routes pay.
Feed, food, or fuel — and the data that decides
There are three broad destinations, in rising order of value and effort:
- Feed — wet to a local farm (cheapest, lowest value, must move within days), or dried into shelf-stable animal feed (more value, but drying burns energy).
- Food — milled into food-grade flour or processed for protein and fibre, going into bread, snacks, and supplements. Highest value per tonne, but tightly regulated and demand-limited.
- Fuel — anaerobic digestion to biogas, or biochar. Modest value, but it closes a loop and can offset site energy.
This is an allocation problem, and it’s exactly the kind machine learning is good at. Forecast next week’s spent-grain volume and grain bill from the brew schedule, then optimise how to split it across routes to maximise value net of haulage distance and the spoilage clock.
But none of that works without the unglamorous first step — measure first, model second. Most breweries can’t tell you how many tonnes of grain left last month, at what moisture, or what the haulage cost. You can’t optimise what you don’t weigh. A scale on the spent-grain auger, a moisture reading, and an off-take log are worth more than any model until they exist.
Where generative AI fits — and where it oversells
The generative-AI angle here is reporting, not routing. Once you have a by-product ledger, an LLM copilot can draft the circular-economy narrative for your CSRD or GRI disclosure, answer plain-language questions (“how much grain went to feed versus landfill last quarter?”), and flag streams worth matching to a nearby off-taker. That’s genuinely useful — it turns a spreadsheet into a story your sustainability team can file.
The trap is letting it generate the story without the numbers behind it. A diverted-from-landfill claim has to be true and verifiable, which is why this belongs with verifying ESG claims rather than just generating them. Generative AI drafts; your meters substantiate.
Where the circular-economy story breaks
Three honest limits. First, perishability beats cleverness: an optimiser that routes grain to food-grade milling is useless if you can’t dry or stabilise it the same day — the two-to-five-day spoilage window is the real constraint. Second, drying can wipe out the carbon win: dewatering 80%-moisture grain is energy-intensive, and unless you use waste heat, the net footprint can be worse than wet feed. Measure net energy, not gross diversion — it ties straight back to honest carbon accounting. Third, demand is the binding constraint, not supply: food-grade markets are thin, regulated, and carry little history, so models predict steady local feed offtake well and novel food markets badly. Supply you have in abundance; buyers are the hard part.
The bottom line
Spent grain is a brewery’s largest by-product and its most overlooked line item. You don’t need AI to start — you need a scale, a moisture reading, and an honest off-take log. Once those exist, forecasting and allocation models turn a disposal cost into a routing decision, and generative AI helps you tell the story to regulators and customers. Just keep the optimiser inside the limits the biology sets: route fast, measure net energy, and never claim a diversion you can’t prove.
Frequently asked questions
How much spent grain does a brewery produce? Roughly 20 kg of wet spent grain for every hectolitre of beer, and it makes up around 85% of a brewery’s total by-product by mass. At about 80% moisture it’s heavy, perishable, and spoils within a few days — which is exactly why where it goes matters.
Is brewery spent grain actually valuable? It can be. Wet, sold as cattle feed, it earns very little. Dried and turned into shelf-stable feed, food-grade flour, or biogas, it earns more — but each route adds processing cost and energy, so the value depends on volumes, moisture, and how far you have to haul it.
Can AI help manage brewery by-products? Yes, for the routine decisions. AI can forecast weekly spent-grain volume from the brew schedule and allocate it across feed, food, and energy routes to maximise value net of haulage and spoilage. It can’t create a food-grade market that isn’t there, and the carbon case only holds if you measure net energy, not gross diversion.
Part of the ESG track.