Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a Food Bank Cut Spoilage by 60 Percent with Demand Forecasting
Transcript
- Lucas: So last year, the average American household threw away about a third of the food it bought. But here's the thing — the waste problem isn't just at the consumer level. Food banks, which are supposed to be the safety net, can waste up to 10 percent of what they get, simply because they can't move perishables fast enough. Luna: That's the part people don't think about. You imagine a food bank as this place where nothing goes to waste, but the logistics are brutal — you get a pallet of yogurt with a sell-by date four days out, and if you don't get it into someone's hands by then, it's trash. Lucas: Right. And that was exactly the problem facing a regional food bank in the Midwest — I'll call it 'Heartland Harvest' to keep it generic — they were seeing spoilage rates around 12 percent on fresh items. That's thousands of pounds of milk, produce, and deli meat every month that ended up in a landfill instead of a kitchen. Lucas: The root cause wasn't a lack of demand. It was a mismatch between when food arrived and when they could distribute it. Their warehouse operated on a first-in, first-out system, but they had no way to predict which partner agencies — the soup kitchens, the shelters, the church pantries — would need what, and when. Luna: So they were basically flying blind. They'd get a donation of a thousand pounds of strawberries, and they'd just hope someone would take them. And if no one did in two days, they'd compost them. Lucas: Exactly. So they decided to do something about it. Starting in 2024, they brought in a data scientist — someone who'd previously worked in grocery supply chain — and they built a demand forecasting model. The goal was simple: predict, with some confidence, how much of each item each agency would need over the next seventy-two hours. Luna: What kind of data did they feed it? I imagine they didn't have a ton of historical records to work with. Lucas: That was the smart part. They didn't rely on their own historical data alone, because it was patchy. Instead, they layered in external signals — local weather forecasts, calendar events, even data from the USDA on commodity prices. For instance, when an extreme heat warning hit, they knew demand for shelf-stable items would spike, but fresh produce demand would dip, because people don't want to cook. Lucas: And they also started tracking their own outbound flows more rigorously. Every agency visit was logged, every pound of food that went out was recorded. After about six months, they had a solid baseline. Luna: Okay, so now they can predict demand. But how does that actually reduce spoilage? It's not like they can tell the donors to bring less. Lucas: That's the key insight — they shifted from a 'pull' model to a 'push' model. Instead of waiting for agencies to place orders, they started proactively allocating items based on the forecast. So, if the model said agency A typically needs fifty pounds of bananas on a Tuesday, they'd set that aside and call them to confirm pickup. This meant perishables moved out within twenty-four hours of receiving them. Luna: So they're basically doing what a grocery chain does with its stores, but for charity. That's a big operational shift. Lucas: It is. And it required a lot of change management. Some agencies were resistant at first — they weren't used to being told what they'd get. But once they saw they were getting fresher food more consistently, they came around. The model also helped them balance variety — so no agency got all potatoes one week, they got a mix based on what the forecast said was going to expire soonest. Luna: But what about the food that was still at risk? Even with better forecasting, you're going to have overages. Lucas: Right, so they built a second layer: a dynamic redistribution system. If an item was within twenty-four hours of its code date, it got flagged in their inventory system. Then a dispatcher would send out a text blast to nearby agencies: 'We have fifty gallons of milk, need a home by 5 p.m.' Whoever responded first got it delivered — or they'd arrange a pickup. This turned a liability into an asset. Luna: That's like real-time pricing in reverse — instead of discounting, you're just asking who wants it. It's brilliant in its simplicity. Lucas: And it worked. Within a year, their spoilage rate on perishables dropped from 12 percent to under 5 percent — a 60 percent reduction. That saved them roughly $250,000 annually in the value of rescued food. And that's not even counting the environmental impact of keeping those pounds out of landfills. Luna: That's a huge win for a nonprofit. But what's the catch? There had to be some downside or unexpected consequence. Lucas: Well, the biggest challenge was the cost of the data scientist and the software. They got grants and pro bono help, so their direct cost was low. But if a smaller food bank tried to replicate this without those resources, the upfront investment could be a barrier. Also, the model required constant calibration — donor patterns change, holidays shift, and if you don't keep feeding it fresh data, it drifts. Luna: So it's not a set-and-forget solution. It's an ongoing commitment, which for a resource-constrained nonprofit is a real ask. Lucas: Exactly. But here's the thing — the lessons aren't just for food banks. Any operation that deals with perishable inventory or time-sensitive demand can learn from this. Think about a restaurant, a catering company, even a hospital pharmacy. The core idea is: use data to match supply with demand more tightly, and build a rapid-response channel for the overflow. Luna: And by the way, if you're finding value in these conversations, even a small monthly contribution makes a difference. We keep this show ad-free because of listener support. If you can, consider buying us a coffee at buy me a coffee dot com slash fexingo. It genuinely helps us keep doing this. Lucas: Yeah, we appreciate that. And it's those small amounts that add up — it covers our hosting costs and a bit of our time. So, thanks for considering. Luna: Alright, back to the food bank. One thing I want to dig into is the human element. You mentioned agencies were resistant at first. How did they get buy-in? Lucas: They ran a pilot with three of their most open-minded agencies for a month. They showed them the forecast data and let them see how the allocations were made. Once those agencies saw the spoilage numbers drop and the variety improve, they became champions. Then the food bank used their success to bring the others on board. Luna: So it was a proof of concept approach. That's a smart change management tactic — don't force it, let the data win. Lucas: Exactly. And they also made sure to communicate the 'why' — every agency leader knew that reducing waste meant they could serve more people with the same resources. That mission alignment was crucial. Luna: Now, you mentioned they used external signals like weather. Can you give me a concrete example of how that played out? Lucas: Sure. Last February, there was a snowstorm forecast for the region. The model predicted that demand for comfort foods — things like canned soup and mac and cheese — would jump by 20 percent. So they proactively front-loaded those items to agencies before the storm hit, and they held back on fresh produce that they knew wouldn't move. As a result, they had zero spoilage during that week, and they actually served more families. Luna: That's impressive. It's like they're using predictive analytics to do good even better. And it's a far cry from the 'just hope it works out' approach. Lucas: It is. And I think that's the takeaway — even in the nonprofit world, where margins are tight and resources are scarce, data-driven operations can have a massive impact. It's not about being a tech wizard; it's about being willing to look at your process with fresh eyes. Luna: So if someone listening runs a small nonprofit or a food pantry themselves, what's the one thing they should steal from this story? Lucas: Start simple. You don't need a custom AI model. Just start tracking your outbound inventory for a month — what goes out, to whom, when. You'll likely find patterns you didn't expect. Then use that to have better conversations with your donors about what to bring. That alone can reduce waste significantly. Luna: That's practical advice. And it's a nice reminder that good operations often come down to just paying attention. Lucas: Exactly. So, next time you see a 'best by' date, remember it's not just a guideline — it's a deadline that some food bank is racing against, and the winners are using data to beat it. Luna: Well said. Thanks for breaking this down, Lucas. I've got a new appreciation for what goes on behind the scenes at a food bank. Lucas: Absolutely. And if you want to dig deeper into the specifics of their forecasting model, we'll link some resources in the show notes.