By Matt Wampler, CEO of ClearCOGS
Let’s say you know tomorrow is going to be a $7,300 day.
That’s useful. It confirms the trend is up, tells you to staff appropriately, and gives you a target. But if you’re running a production operation where everything has to be made before 5am and each product comes in batches of twelve, that dollar figure doesn’t actually tell you what to make.
How many of the 64 varieties do you produce? Which ones can you run out of, and which ones absolutely cannot be 86’d before noon? How do you split production across six different dough bases when you won’t know the exact mix until the day is underway?
This is the forecasting problem that high-variety food operations deal with every day. It is a fundamentally different problem than the one most forecasting tools are designed to solve.
Quick Answer: A total sales forecast tells you how busy you’ll be. It does not tell you what to make. Item-level forecasting predicts demand for each specific product at each location, accounting for batch sizes, shelf life, and local demand signals. For high-variety operations with hard production cutoffs, that distinction is where margin is won or lost.
What Is the Difference Between Total Sales Forecasting and Item-Level Forecasting?
Most restaurant forecasting starts from the top. Estimate overall revenue. Apply historical product mix percentages. Get a rough count of how many units of each item to expect. For most restaurant formats, that approach is good enough.
For operations with large catalogs, batch constraints, and hard production cutoffs, “good enough” is where the waste and the stockouts live.
The challenge is not predicting total demand. Modern forecasting tools handle that reasonably well. The challenge is predicting individual item demand accurately enough to make production decisions by SKU, accounting for batch size, shelf life, sell-through priorities, and the reality that some items cannot run out while others can absorb a little slack.
Predict total sales at 95% accuracy and you have still got a production planning problem if you are wrong on which specific items drove that revenue.
Why Does Item-Level Accuracy Matter More for High-Variety Operations?
In a narrow product catalog, most items behave similarly. Weather affects them the same way. Day-of-week patterns track together. It is relatively straightforward to apply a general demand lift and see it distribute across the menu.
Add 60-plus product varieties and the picture gets more complicated. Some items are more sensitive to weather. Some spike on weekends, some hold steady across the week. Some sell through quickly when a specific event is nearby and barely move when it isn’t. Some are category anchors that customers expect to find, and running out of them damages something beyond just that day’s revenue.
A top-down average doesn’t capture any of that. It assumes every item in the catalog moves in proportion to total demand, which is rarely true once you look at the actual data.
The right approach builds a separate model for each item, at each location, trained on its own historical patterns and its own responses to local factors: school calendars, hotel occupancy, conventions, weather. That kind of model doesn’t just tell you how much you’ll sell. It tells you the mix, and the mix is what drives production planning.
How Do Batch Constraints Make the Forecasting Problem Harder?
Production in batch-based operations adds another layer of constraint that pure forecasting doesn’t account for on its own.
If the model says you need 17 of a particular variety, but that variety is only produced in trays of 12, your real decision is whether you make one tray or two. Make one, and you risk running out by early afternoon. Make two, and you’ve added 7 units of potential waste on a one-day product.
That math plays out across dozens of items, multiple times a day, with different shelf life windows, different waste tolerance levels, and different sell-through priorities for each. Some items need buffer. Some items should run out on schedule. The production guidance your team receives needs to reflect all of that, expressed in operational language, not ounces or percentages.
According to WorldBakers, a significant portion of food waste in commercial bakery and production operations comes from overproduction, with items produced in batches based on estimated demand that don’t match what actually moves through the day.
What Does Item-Level Forecasting Account For That Top-Down Models Miss?
The variables that make high-variety operations hard to predict are not the ones that average out across a catalog. They are the ones that show up differently for each item, at each location, on specific days.
A give-back night at a local school drives three times the normal Tuesday volume at one store and has no effect on another ten miles away. College campus schedules shift demand dramatically during exam weeks and spring breaks, by district, by year. Hotel occupancy patterns create demand spikes on certain days at certain stores that look like noise in the aggregate but are entirely predictable at the location level.
Item-level models trained on location-specific data learn these patterns. They treat each location as its own system rather than a scaled-down version of the whole brand, and they apply those local signals to each product separately rather than as a uniform lift across the menu.
What Changes When Item-Level Forecasting Drives Production?
When item-level forecasting is accurate enough to drive production guidance, the whole operation shifts.
The production team stops making conservative batch decisions to avoid stockouts and starts making precise ones because the number they’re given is reliable. Leadership can set real parameters: these five items must always have inventory at close, these ten can run down to zero by 8pm. Those parameters get built into the production output, not estimated on the floor.
The difference between knowing you’ll do $7,300 tomorrow and knowing exactly how that $7,300 breaks across 64 products, accounting for batch constraints and sell-through priorities, is the difference between a useful number and an operational plan.
For operations where production happens before the doors open and every item has a shelf life, that gap is where margin is won or lost.
Frequently Asked Questions
What is item-level forecasting for restaurants?
Item-level forecasting predicts demand for each specific product at each location, rather than estimating total sales and distributing that number across a menu. For high-variety operations, it builds a separate model for each SKU trained on its own historical patterns, local demand signals, and production constraints. The result is a production number per item, not just a revenue target.
Why doesn’t a total sales forecast work for batch production?
A total sales forecast tells you how much revenue to expect, but not which specific items drove it. For batch operations where every product is made before the doors open in fixed tray or unit quantities, the item-level mix is the actual production decision. Getting total sales right at 95% accuracy still leaves a production planning problem if the item mix is wrong.
How does item-level forecasting reduce food waste in high-variety operations?
Item-level forecasting reduces waste by replacing conservative batch decisions with precise ones. When the production team trusts the per-item number, they stop defaulting to over-production as a stockout buffer. Leadership can set explicit parameters for which items must stay in stock and which can sell out, and those parameters get built into production guidance rather than estimated on the floor.
What variables should a high-variety forecasting model account for?
A strong item-level model accounts for local events, school calendars by district, hotel occupancy patterns, weather, and day-of-week behavior, applied individually at each location and for each SKU. Items in the same catalog often respond differently to the same external signals, which is why a top-down model that assumes uniform lift across all products underperforms for complex menus.
Can item-level forecasting work for operations with 60 or more products?
Yes, and it is specifically designed for them. The value of item-level forecasting increases with catalog size because top-down averaging becomes less reliable as variety grows. Each product gets its own model trained on its own history, so the forecast reflects how that specific item actually behaves at that specific location, not how the average product behaves across the whole menu.
If your team is still translating a total sales forecast into production decisions by hand, we’d be glad to show you what item-level forecasting looks like in practice.
Sources
- WorldBakers. Transforming Food Waste. worldbakers.com
