Quick Answer: Food cost varies between locations of the same restaurant brand for three reasons: sales mix (some locations legitimately sell more of the highest-cost items), ordering behavior (some teams consistently bring in more product than demand requires, which feeds waste and shrinkage), and forecast quality (veteran managers guess demand well; new stores and new managers guess blind). The spread persists because a standard P&L shows all three causes as one number. The fix is a location-level expected number, what each store should have used given what it actually sold, so mix can be separated from mismanagement and each location is compared against its own benchmark instead of the brand average.
The owner of a fast casual group with a dozen locations, nine-plus years into the business, described his version of this to us recently, and his numbers will sound familiar to any multi-unit operator: food cost across his locations ranges from 26 percent to 36 percent. Same brand, same menu, same vendors. His best-running stores sit at 26 to 28. A few long-tenured locations hover stubbornly at 33 to 34. And his newest stores, opened within the last year, run the highest of all.
His explanation for the good stores was the most honest sentence we have heard about restaurant forecasting: his AI system, right now, is his individual managers. The veterans running 26 percent do it on an internal system they have built up over years of watching that store. That is the whole game in one line, and it points to both the cause of the spread and the way out of it.
How Much Do Locations of the Same Brand Normally Vary?
More than most operators expect, and restaurants are not unusual in this. Economists have documented for decades that performance dispersion inside the same industry is enormous: the canonical research finding is that the 90th percentile plant produces almost twice as much output from the same measured inputs as the 10th percentile plant, even within narrowly defined industries, and that these gaps persist year after year. A restaurant brand is that phenomenon in miniature: a dozen buildings running the same playbook with materially different results, and the differences do not fix themselves.
The useful reframe is that this spread is not only a problem; it is information. An eight-to-ten point food cost gap between your best and worst locations is the most actionable number on your P&L, precisely because it is internally benchmarked. Same recipes, same suppliers, same pricing. Whatever separates 26 from 36 is happening inside your own walls, which means it is diagnosable and fixable, if you can attribute it.
What Actually Causes the Spread?
Three suspects, and they hide behind the same number.
Sales mix. Some locations genuinely sell a richer mix. A store where the highest-cost proteins outsell the cheaper items will run a higher food cost percentage while being perfectly well managed, and possibly more profitable in dollars. This operator suspects exactly this at some of his 33 percent stores. Higher food cost from mix is not a failure; it may even be a strength.
Ordering behavior. His own words for the second suspect: are my guys bringing in too much? At a cook-to-order concept like his, visible waste stays low, so over-ordering does not show up in the trash. It shows up as product sitting too long, quiet spoilage, over-portioning from abundance, and shrinkage, all of which land in food cost with no label attached. Excess inventory has a way of disappearing, and the P&L cannot tell you it happened, only that the percentage is high.
Forecast quality. The third suspect explains why his newest locations run worst. A store that has not lived through its own seasons has no history to guess from: a hot week or a slow week swings the percentage wildly, and every order is placed against a blank page. Meanwhile the veteran manager at the 26 percent store carries years of pattern recognition, which is why the operator calls his managers his current AI. That intelligence is real, and it is also learned, slow to build, unevenly distributed, and it resigns when the manager does.
Here is the trap: on a standard P&L, all three suspects produce the identical symptom, a high food cost percentage. Mix looks like waste. Waste looks like mix. A blind new store looks like a badly run old one. Operators end up either accusing good managers or excusing bad ordering, because the one report everyone reads cannot tell the difference.
How Do You Tell Mix From Mismanagement?
You separate the suspects with one addition: an expected number per location. The method is straightforward. Take what each location actually sold, item by item, from the point of sale. Run those sales through the recipes. The result is what that store should have used, at its own sales mix, called theoretical usage. Now three comparisons become possible that the P&L alone can never make.
First, compare each store’s expected food cost to the brand average: that gap is pure sales mix, and it is legitimate. Second, compare each store’s actual usage to its own expected usage: that gap is execution, the over-ordering, over-portioning, and shrinkage that deserve management attention. Third, watch the execution gap over time: a store whose gap shrinks is improving even if its percentage stays high because of mix, and a 28 percent store with a widening gap is quietly getting worse behind a healthy-looking number.
The same machinery solves the new-store problem. A location with thin history should not be forecast from its own data alone; its demand patterns can be extrapolated from sibling stores with similar profiles until it accumulates seasons of its own. That is precisely how the veteran manager’s learned intelligence gets replaced for a store that has no veteran: the brand’s collective history becomes every location’s starting point. This is the layer we build at ClearCOGS, expected usage and forward-looking order and prep numbers per location, per item, per day, so the 26 percent manager’s judgment effectively gets installed in all twelve buildings, including the two that opened last spring.
What Should a Multi-Unit Operator Do This Week?
Start with attribution, not accusation. Pull each location’s item-level sales for last month and compute what each should have spent at its own mix. Rank locations by the gap between actual and expected rather than by raw food cost percentage. The ranking will surprise you: some high-percentage stores will prove innocent (mix), and some average-looking stores will reveal the largest execution gaps. Then point your attention, your training, and your forecasting investment at the gap, because the gap, unlike the mix, is entirely yours to close.
Frequently Asked Questions
What is a normal food cost variance between locations of the same brand?
Spreads of five to ten percentage points are common in multi-unit groups, especially those with new locations. The absolute spread matters less than knowing how much of it is sales mix versus execution, which requires comparing each location to its own expected usage.
Is a higher food cost location always badly managed?
No. A location selling a richer mix of high-cost items can run a higher percentage while being well managed and highly profitable in dollars. Mismanagement only becomes visible when actual usage exceeds what the location’s own sales justify.
Why do new restaurant locations have higher food costs?
They lack history: no seasons in the data, no learned demand patterns, and often newer managers. Every order is a blind guess, and blind guesses skew toward over-ordering. Extrapolating demand from established sibling locations closes the gap faster than waiting for experience.
Can you fix location variance without a full inventory system?
Largely, yes. The attribution step needs item-level sales (which the POS already records) and recipes. Full inventory platforms add the actual-usage record; the expected-usage side, which drives daily ordering and prep decisions, comes from forecasting rather than counting.
Sources
- Syverson, C. What Determines Productivity? NBER Working Paper 15712 (published in the Journal of Economic Literature, 2011). nber.org
