Latest / Restaurant Business with Fexingo: Food Service, Hospitality, and Independent Operators / How Restaurants Are Using Data to Predict What Youll Order
Transcript
- Lucas: So there's this pizza chain in the Midwest — twelve locations, family-owned, been around since the late eighties — and they've been running an experiment that most people would assume requires a data science team and a six-figure software budget. Luna: Let me guess: they're predicting what customers are going to order before they order it. Lucas: Exactly that. And they're doing it with stuff most restaurants already have: point of sale history, a weather API that costs fifty bucks a month, and a shared Google Sheet. Lucas: Look, if these conversations have moved your work forward in some small way — maybe you've looked at your own inventory differently, or thought twice about a vendor contract — listener support is what keeps this show ad-free and independent. If that's something you value, you can chip in at buy me a coffee dot com slash fexingo. No pressure, ever. Just a way to keep the conversation going. Luna: Yeah, it makes a real difference. And we don't run a single ad, so it's entirely listener-powered. Lucas: Alright, back to the pizza chain. Their COO told me they were tired of guessing how many large pepperonis to prep on a Tuesday. So they started tracking sales against three variables: day of week, whether there was a local high school football game, and the forecasted high temperature. Luna: So basically, what any experienced manager might do intuitively, but they made it systematic. Lucas: Right. And the results are pretty striking. After six months, they cut food waste by twenty-two percent. Their variance between forecast and actual orders dropped to under eight percent. And they started scheduling one fewer prep cook on low-demand nights, which added up to about forty thousand dollars in labor savings over the year. Luna: Forty thousand for a twelve-unit chain is real money. But is this something a small independent can actually replicate without a tech team? Lucas: That's the interesting part. The COO said they didn't write a single line of code. They used the forecasting function in their POS system — most modern POS platforms have some basic predictive analytics built in, but operators rarely turn them on. Then they layered in weather data from a free government API and manually entered local event dates into a spreadsheet. Luna: So the barrier is less technical and more behavioral. You have to actually want to look at the data. Lucas: Exactly. And that's where I think the opportunity is. Big chains like Domino's and Pizza Hut have been doing this for years with sophisticated machine learning models. But the gap between what a large chain can do and what a twelve-unit independent can do is shrinking, because the tools have gotten cheaper and easier. Luna: But there's a flip side, right? If you're predicting what someone's going to order, you're collecting data about them. Where do you draw the line between smart inventory management and surveillance? Lucas: That's a fair question. In this case, they're not tracking individual customer data — no names, no phone numbers, no purchase histories tied to a person. They're looking at aggregate order counts by hour and by item. So it's more like weather forecasting for your restaurant than it is tracking a specific guest. Luna: Still, if a customer comes in every Tuesday and the system learns that, and then the manager says 'hey, you always order the same thing' — that can feel invasive. Lucas: It can. And I think operators need to be thoughtful about that. The COO I spoke with said they made a conscious decision not to store any personally identifiable information. They only keep transaction-level data — what was sold, when, and at what price — without linking it to a specific person. That's a good rule of thumb. Luna: I want to go back to the labor savings. Forty thousand dollars — that's roughly the salary of one full-time employee. So effectively, they're not hiring a person they would have needed before. Lucas: That's one way to look at it. But the COO framed it differently. He said they're not cutting hours across the board — they're reallocating them. Instead of having three prep cooks on a slow Tuesday, they have two, and the third person works a double on Friday when it's actually busy. So the same total labor hours, just better distributed. Luna: That's a healthier way to think about it. Nobody wants to work a shift where you're standing around. Lucas: Exactly. And the employees actually prefer it because they make more in tips on the busy shifts. So it's not just a cost play — it's an employee satisfaction play too. Lucas: One thing I found surprising: the biggest variable wasn't weather or day of week. It was the local high school football schedule. On game nights, demand for large pizzas jumped forty percent, and they would sell out of wings by eight o'clock if they didn't prep extra. Luna: That makes total sense. But a lot of operators wouldn't think to track that, or they'd rely on the manager's memory rather than a system. Lucas: Right. And memory is fallible. The COO told me they used to have a manager who was great at guessing demand — he'd been there fifteen years, knew the community. When he left, his replacement struggled. The spreadsheet system made that knowledge institutional, not personal. Luna: So it's a way to de-risk your operation against turnover. That's a big deal in an industry where manager turnover is over thirty percent annually. Lucas: Exactly. And the cost of implementing this was essentially zero. They already had the POS system. The weather API is free for basic usage. The spreadsheet cost nothing. The only real investment was the time to set it up and the discipline to enter data consistently. Luna: I wonder if there's a downside. What happens when the model is wrong? Like, it predicts a big night and the game gets cancelled due to weather. Lucas: That happened once during their trial. They'd prepped for a hundred large pizzas based on the schedule, and the game got rained out. They ended up with about forty extra pizzas. But because they had the data, they knew it was a rare event — not a systemic error. So they adjusted, and now they check the weather forecast for game day before finalizing prep. Luna: So the model improved because they paid attention to the exceptions. That's exactly how these systems should work — not as black boxes, but as tools that managers interact with. Lucas: That's the key insight, I think. Predictive ordering isn't about replacing the manager's judgment. It's about giving them better information so their judgment is sharper. The COO said his managers still override the forecast maybe twenty percent of the time — for things the model can't capture, like a funeral reception that books a large party, or a road construction project that might deter customers. Luna: Twenty percent override rate sounds about right. That means the model is useful eighty percent of the time, which is huge. Lucas: And the managers feel empowered, not replaced. They know they have a tool that handles the routine predictions, so they can focus on the exceptions — the things that actually require human intuition. Luna: What about other types of restaurants? Could a fine dining place use this approach, or is it mostly suited for high-volume, limited-menu concepts like pizza? Lucas: It works best for operations with a limited number of SKUs and predictable demand patterns — pizza, fast casual, quick service. For fine dining, where the menu changes frequently and each dish is more complex, the data set is thinner and the variables are harder to model. But even there, you can use it for basics like how many steaks to order for the week, or how much bread to bake. Luna: So it's about starting small. Pick one category — maybe your most popular protein — and build from there. Lucas: Exactly. The pizza chain started with just large pepperoni. Once they got comfortable, they expanded to wings, then to all their top ten items. Now they're forecasting for everything. But they took it one step at a time. Luna: I love that this is a case where the independent operator has an advantage over the big chain. The chain has to get buy-in from corporate, install new software across hundreds of locations, deal with franchisee pushback. A twelve-unit operator can just decide to try it and see if it works. Lucas: That's the beauty of being small and nimble. And I think it's a reminder that not every innovation has to be expensive or complex. Sometimes the best tool is a spreadsheet and the willingness to ask a simple question: what actually drives our sales?