Food cost percentage: how to calculate it, what's normal for your concept, and how to actually move it

Food cost percentage is the most-quoted number in the restaurant business and one of the most casually miscalculated. Operators quote it from memory, compare it against benchmarks from a different concept, and compute it from purchases instead of usage — three small errors that compound into menu decisions made on fiction.

I spent years as the partner who closed the books at a four-location group, which means I've seen every version of food cost theater: the chef who "runs a 28" because he divides this week's produce invoice by this week's sales, the GM who hit budget by simply not counting the walk-in. This post is the version of the number that survives an audit — how to calculate it properly, what normal looks like for your concept in 2026, and which cost levers have actual evidence behind them.

The formula — and the mistake inside it

Food cost percentage is cost of goods sold divided by food sales:

Food cost % = COGS ÷ food sales

COGS = beginning inventory + purchases − ending inventory

The mistake lives in the COGS line. If you skip the inventory counts and just divide purchases by sales, you're measuring your buying, not your usage. Buy heavy for a holiday weekend and your "food cost" spikes; run the shelves down and it magically improves. The number becomes a purchasing-calendar artifact instead of an operating metric.

The discipline that fixes it is unglamorous: a real beginning count, a real ending count, the same period boundaries as your sales data. Weekly is the operating standard for a reason — a monthly count gives you twelve chances a year to notice a problem that compounds daily.

Separate food and beverage entirely. Bar programs run a completely different cost structure, and blending them produces a single number that flatters the kitchen and hides the bar.

What "normal" looks like by concept

Industry benchmarks put most restaurants between 28% and 35%, but the band for your concept is what matters — published ranges by type:

Two things the table should change about how you talk:

• "High" is not "bad." A steakhouse at 40% food cost with a $68 average check banks more gross-margin dollars per cover than a pizzeria at 22%. You deposit dollars, not percentages — the percentage is a control metric, not a scoreboard.
• Cross-concept comparison is noise. A fast-casual operator "beating" a fine-dining friend on food cost has learned nothing about either business.

Prime cost: the number above this number

Food cost only means something next to labor. Prime cost — COGS plus total labor — is the figure lenders and buyers look at first, and the published benchmark is 60–65% of sales (closer to 55–60% for quick service). Restaurants running above 65% have a structural problem, not a purchasing problem.

Prime cost is also where trade-offs get honest. Scratch-made dough lowers food cost and raises labor; pre-portioned proteins do the reverse. Chasing the food-cost number alone optimizes half an equation.

Theoretical vs. actual: where the money actually leaks

The single most useful upgrade to food-cost tracking is splitting it in two:

• Theoretical (ideal) food cost — what the period should have cost: every item sold, multiplied by its recipe cost at current ingredient prices, with perfect portions and zero waste.
• Actual food cost — what the inventory math says you really used.

The gap between them is the diagnosis. Theoretical too high means a menu problem — recipes, pricing, item mix. Actual far above theoretical means an execution problem: over-portioning, waste, spoilage, refires, unrecorded comps, or product walking out the back door. Industry guidance treats a variance under 2% as well-run and anything past 5% as a systemic leak. The dollar stakes scale brutally: published examples put a 4-point variance at a $1M-sales restaurant around $40,000 a year in vanished profit.

You can't compute theoretical cost on a legal pad — it requires recipe-level costing tied to item-level sales data, which is precisely the job of your POS and its analytics layer.

The 2026 backdrop you're costing against

Benchmarks are moving targets, and the current environment is the toughest costing market in years. USDA's May 2026 outlook forecasts overall food prices up 3.4% this year, with brutal dispersion underneath: beef and veal forecast up 12.1%, coffee-driven beverage costs up nearly 6%, while eggs are forecast down almost 30% after the avian-flu spikes finally unwound. Wholesale food prices overall stand roughly 35% above pre-pandemic levels per the National Restaurant Association's tracking, and its 2026 State of the Industry reports 82% of operators paying more for food, with roughly nine in ten full-service operators raising menu prices in response.

The practical translation: a static menu costed in 2024 is wrong in 2026 by double digits on some lines. If beef is 12% of your plate costs and it inflates 12%, that's a point and a half of food cost appearing from nowhere — without anyone in your kitchen doing a single thing differently.

The levers, ranked by the evidence

Everything moves food cost a little; these are the moves with the strongest published support, in rough order of evidence quality:

1. Measure waste formally. The strongest independent number in this category: a multi-country study of 114 restaurant sites found an average $7 saved per $1 invested in kitchen food-waste programs, with sites cutting waste 26% in the first year. The interventions were boring — weighing waste, training on storage and handling, menu redesign — which is exactly why they worked.
2. Track theoretical vs. actual weekly. The variance number converts "food cost is up" from an argument into a work order: it tells you whether to fix the menu or the line.
3. Engineer the menu on contribution margin. Star items, plow horses, and the dogs you finally cut — the framework is in our menu engineering guide. No honest headline stat exists for menu engineering's impact, but it's the lever that turns costing data into pricing decisions.
4. Re-bid suppliers on your top 20 items. Per the NRA's 2026 data, shopping suppliers was among the most common operator responses to cost pressure — your top 20 SKUs are usually 80% of spend, so bid those, not the whole order guide.
5. Tighten the inventory cadence with software. Inventory platforms claim a 2–5% food-cost reduction for their users — vendor-reported, so season to taste — but the mechanism is real: counts that take hours happen weekly; counts that take a day happen never. Our piece on AI-assisted inventory management covers what the current tooling automates.

What your POS should be doing about all this

Here's the uncomfortable test: if computing last week's actual and theoretical food cost takes your team more than a few minutes, your systems are the bottleneck — the data already exists. Item sales live in the POS; recipes and invoices live wherever you put them; the only question is whether they're connected. A reporting layer that holds recipe costs against item-level sales gives you theoretical cost as a byproduct of ringing tickets, and variance becomes a standing Monday report instead of a quarterly archaeology project.

Disclosure: I work at Katalyst, and the bias here is structural — we build the integrated version of this, so I think the integrated version is correct. But the math doesn't care which logo is on the terminal: pick the right benchmark band for your concept, count inventory like you mean it, split theoretical from actual, and put the variance in front of someone every week with the authority to act on it. Food cost is the most controllable big number on your P&L. It just refuses to be controlled from memory.

Related Katalyst products

• See Katalyst analytics