Demand Planning Analytics
Every Forecast is a Bet That Tomorrow Resembles Yesterday
How a Food and Beverage Manufacturer Rebuilt Demand Planning After the Forecast Broke
Every demand forecast makes the same assumption: that next year will look enough like last year for the math to hold. For most companies, most of the time, it does.
A food and beverage manufacturer ran its planning this way for years. Historical sales data fed the model, the model produced forecasts, and the forecasts drove production and distribution. It worked well enough that nobody had a reason to question the assumption underneath it.
Then the assumption broke. COVID rearranged demand and supply at the same time. Buying patterns shifted. Distribution channels changed. The historical data the model relied on described a world that no longer existed, but the model kept producing confident numbers anyway. Forecasts still looked like forecasts. They just stopped being right.
In food and beverage, that confidence is expensive. The product has a shelf life, so every planning error lands in one of two places: a lost sale (too little on the shelf) or waste (too much, and it expires before it moves). Most industries get to be wrong in one direction. Food and beverage pays for both.
Millions in waste, with no line item to show for it
The manufacturer knew it had a forecasting problem. The piece it couldn’t see was where the consequences were landing.
Waste was running $20 to $30 million a year in expired product. The number was visible in aggregate and owned by no one. It lived in the cost of doing business, distributed across facilities and product lines, showing up in totals but never attached to specific planning decisions or the people making them. Nobody was accountable for it because nobody could see it at the level where accountability would mean something.
The stockout risk was just as real but even harder to track. A product that runs out before the next shipment arrives is a lost sale that never shows up in a report, because you can’t count what didn’t happen. Planners could see what was on the shelf today. They couldn’t see when it would be gone.
Two reports built around two questions
We started with the question planners were asking every day but couldn’t answer systematically: when do we run out?
The first report was a stockout forecast anchored on depletion rates. Take what’s on the shelf, measure how fast it’s moving, and calculate the date it hits zero. The math for any single product is straightforward. The hard part is running it across an entire catalog, keeping it current as sales velocity changes, and surfacing the results in a way that lets planners act before the shelf goes empty.
The second report addressed the other side of the equation. Knowing when you’ll stock out tells you where to send more product. It tells you nothing about what you’re throwing away. So we built a wastage report that tracked expired inventory by product, by location, and by the planning decisions that put it there. Expired product went from an accepted cost of business to a number with a name on it: which SKU, which facility, which planning cycle.
Together, the two reports gave planners a view of both risks in the same frame. They could see the cost of ordering too little and the cost of ordering too much, and adjust accordingly.
What became visible
When $20 to $30 million in expired product carries a line-item breakdown by SKU and facility, the conversation changes. Planners could weigh overstock risk against stockout risk with both costs on screen, rather than optimizing for one while treating the other as overhead.
The stockout forecast gave distribution teams lead time they’d never had. Rather than reacting to empty shelves, they could see depletion dates far enough ahead to reroute or reorder. The result was a system that told planners when the forecast was wrong before the shelf did.
The Lesson
Forecasting works until the pattern it’s trained on changes. When that happens, the model doesn’t send a warning. It keeps producing numbers with the same confidence, built on a past that no longer applies. You find out months later, in expired inventory or empty shelves.
The value here was visibility into the two questions that matter most in perishable goods: when do we run out, and what are we throwing away? Once those answers lived in a report instead of a gut feeling, planning decisions got better because the cost of getting them wrong was finally in the room.
Find the Drift Before It Becomes Waste
Last year’s patterns run most demand forecasts, and they hold right up until the market moves. When it does, the model keeps producing confident numbers off a past that no longer applies, and you find out on a shelf. We build the reports that put both sides of that risk, stockout and waste, in one view, so a drifting forecast shows up as a number instead of a surprise. And if your patterns are still holding and the model is fine, we’ll tell you that instead of selling you a rebuild. Reach out to us and book a free Data Health Check.
