Food cost percentage is the cost of food sold divided by food sales for the same period in one restaurant. Cost of food sold is opening inventory plus purchases minus closing inventory. No sourced universal benchmark exists, so an owner compares the figure against their own earlier periods and their own menu.
What food cost percentage measures — and what it quietly assumes
Food cost percentage — the cost of food sold in a period, divided by food sales in the same period, expressed as a percentage.
Two things make this ratio harder than it looks, and both live in the definitions rather than in the arithmetic.
The numerator is not what you paid your suppliers this month. It is what left your storeroom and your kitchen and ended up on plates that guests paid for. Those are different numbers whenever you buy ahead of a busy weekend, or run down a freezer before a stocktake. The bridge between them is inventory.
The denominator is not your till total either. It is food sales, net of VAT, net of discounts, net of voided bills. Every one of those exclusions moves the percentage, and each of them moves it in the direction that flatters you if you forget it.
So the honest version of this metric always carries its own boundary conditions: which period, which revenue lines, which inventory counts. A restaurant that changes any of those quietly between two months has not measured a change in food cost. It has measured a change in bookkeeping.
The formula, and the three inventory numbers behind it
The cost of food sold is not read off an invoice. It is reconstructed:
Cost of food sold = Opening inventory + Purchases − Closing inventory
Opening inventory— the valued food stock on hand at the start of the period, PLN;Purchases— food bought during the period at invoice value, PLN;Closing inventory— the valued food stock on hand at the end of the same period, PLN.
Worked through: opening PLN 18,400, purchases PLN 61,200, closing PLN 16,900 gives 18,400 + 61,200 − 16,900 = PLN 62,700 of food sold.
That figure is still slightly wrong in most kitchens, because food does not only move to guests. It moves to the bar, from the bar, and to your own staff. The corrected version:
Adjusted cost of food sold = Opening + Purchases − Closing − Transfers out + Transfers in − Staff meals
Transfers out— food that left the kitchen for another cost centre, most often the bar, PLN;Transfers in— goods that came the other way, such as juice or cream drawn from the bar, PLN;Staff meals— the valued cost of what your own team ate, PLN.
Continuing the same period: 62,700 − 2,850 + 640 − 1,730 = PLN 58,760. Without those three corrections the kitchen carries the bar's citrus and your team's lunch, and every ratio built on top is shifted.
Then, and only then:
Food cost % = Adjusted cost of food sold ÷ Food sales × 100
Food sales— revenue from food, net of VAT, discounts and voided bills, PLN.
Why that number, on its own, tells you nothing yet
You now have a percentage. What you do not have is a verdict, and this is where most articles on this subject go wrong. The next section explains why this page will not hand you one.
Why this page names no benchmark band
You will find a food cost range quoted almost everywhere: in software vendors' blogs, in consultancy checklists, in benchmark guides assembled from those same blogs. We went looking for where those ranges originally came from, and the trail does not lead anywhere you could cite.
| What is usually offered | Where the figure is published | What that publisher gives as its own source | Why it is not printed on this page |
|---|---|---|---|
| An "industry standard" range for food cost | Restaurant software vendors' marketing blogs | Nothing, or another vendor blog | A vendor's marketing page is not a measurement; nobody states the sample, the period or the definition used |
| A consolidated "benchmark guide" | A buying group's downloadable PDF, October 2024 | A list of vendor blogs — Restaurant365, TheForkCPAs, TouchBistro, ToastTab, Sculpture Hospitality and others | It aggregates the same unsourced figures; aggregation does not create provenance |
| The often-repeated "acceptable variance" threshold | Inventory software documentation | The vendor that supplied the definitions names no threshold at all | The threshold has no author |
| A genuinely surveyed operations report | The US national restaurant trade association, paid publication | Its own survey | Not publicly available, US-only, and its cost definitions are not the ones a Polish P&L uses |
Numbers in this table: none. It is about where figures come from, not about their values.
That is the whole reason this page's title says "formula" and not "benchmarks". Reproducing a band whose origin we cannot show would be the single most quotable thing on the page and the least defensible. If you have seen a range and want to keep using it, that is your call — but use it knowing it is a rule of thumb passed between marketing departments, not a measured norm.
What does exist, and what we can show you, is official statistics. They answer a different question, and they answer it properly.
What Polish official statistics actually say about food service costs
These are Polish figures, from Polish and European official collections, and they describe enterprises rather than kitchens — read the caveat beside each one before you use it.
sbs_ovw_act, NACE I56, reference year 2023, data updated 10.03.2026). It is not a food cost, and reading it as one would be a serious error: that line buys rent, energy, outsourced services, laundry and delivery commissions along with the food, which is exactly why it lands far above any plate cost you could build. The rest of that table, with the caveat each row carries, belongs to the prime cost page, where it stands in full. It is not reprinted here: one table kept in two places starts to disagree with itself.
The Polish share sits well above the EU-27 one for structural reasons — more of the sector's money leaves as purchases and less of it as payroll — and not because Polish kitchens waste more food.
Poland's own statistical office adds the piece that matters most for the denominator.
Nearly one PLN in eight of Polish gastronomy revenue is goods bought and resold rather than cooked. That is not a detail — it is the reason the drinks question below has a real answer instead of a preference.
Building your own comparison base
If there is no external norm, the comparison has to be internal. It is built from three things, and none of them require an outside number.
Your own past
The first base is the same venue in an earlier period. Weekly is usually the shortest useful step, because that is the shortest span over which an inventory count is worth the hours it costs. The comparison is valid only if the boundary conditions held: the same revenue lines in the denominator, the same transfer and staff-meal corrections, the same valuation method for stock.
Deviation from your own base (p.p.) = Current period food cost % − Base period food cost %
- both terms are percentages, so the result is in percentage points, not percent.
Recipe changes kill a base. If you reformulated a dish in week three, weeks one and two are no longer comparable to week four for that dish, and you should rebuild rather than pretend.
Your own segment
The second base is the venue's own format. A tasting-menu kitchen buying whole animals, a bar kitchen reheating prepped components and a delivery-only operation buying portioned goods are three different arithmetics. Comparing across them tells you which business model you are in, which you already knew, and nothing else.
Your own menu
The third base — and the one this page argues is the most neglected — is the current sales mix. The same restaurant, with the same suppliers, the same prices and the same recipes, produces a different blended food cost percentage in two consecutive weeks simply because guests ordered differently. The section on the mix works this through in full.
Plate cost: one dish costed from the spec sheet to the plate
Plate cost — the costed sum of every recipe component at yield-adjusted prices, for one portion exactly as it is served.
Each row below is a component of the row above
"Exactly as served" includes the sauce, the garnish, the side that is not charged for, and the bread that arrives before anyone orders. If it leaves the pass on the plate, it is in the plate cost.
Plate cost = Σ (Yield-adjusted unit cost × Portion quantity in usable units)
Yield-adjusted unit cost— price per unit of usable product, PLN per usable kg or litre;Portion quantity— the amount that goes on the plate, measured in the same usable units.
| Component | Purchase price per unit | Yield (fraction) | Cost per usable unit | Portion, usable units | Cost in the portion |
|---|---|---|---|---|---|
| Beef cheek | PLN 34.00/kg | 0.68 | PLN 50.00/kg | 0.180 kg | PLN 9.00 |
| Root vegetables | PLN 4.50/kg | 0.78 | PLN 5.77/kg | 0.120 kg | PLN 0.69 |
| Butter | PLN 32.00/kg | 1.00 | PLN 32.00/kg | 0.020 kg | PLN 0.64 |
| Stock for the sauce | PLN 6.20/l | 1.00 | PLN 6.20/l | 0.090 l | PLN 0.56 |
| Herbs for the garnish | PLN 48.00/kg | 0.55 | PLN 87.27/kg | 0.005 kg | PLN 0.44 |
| Bread, served uncharged | PLN 0.90/portion | 1.00 | PLN 0.90/portion | 1 portion | PLN 0.90 |
| Plate cost | PLN 12.23 |
Numbers in this table: a worked example built by the author to demonstrate the arithmetic. Every purchase price and every yield fraction here is an assumption of the example, not a published norm — measure your own, from your own deliveries and your own butchery, and replace all six rows.
The same worksheet is what turns an event enquiry into a defensible price rather than a guess; we walked through that flow in pricing an event catering job.
Yield: a purchased kilogram is not a plated kilogram
Yield — usable weight after trimming, cooking and portioning, divided by purchased weight, expressed as a fraction between zero and one.
This is the single correction that changes plate cost the most, and the one most often applied backwards.
Yield-adjusted unit cost = Purchase price per unit ÷ Yield
You divide, and you divide by the fraction — not by the percentage. Beef cheek bought at PLN 34.00/kg with a yield of 0.68 costs 34.00 ÷ 0.68 = PLN 50.00 per usable kilogram. Dividing by 68 instead of 0.68 gives PLN 0.50, which is absurd enough to catch; multiplying by 0.68 gives PLN 23.12, which is not absurd enough to catch, and is wrong in the direction that makes every dish look profitable.
That the gap between purchased and plated weight is real, and large, is not a matter of opinion. Eurostat's food waste collection (env_wasfw, data updated 18.02.2026) records, for reporting year 2023, 293,062 tonnes of food waste from the "Restaurants and food services" activity group in Poland — 8 kg per inhabitant — against 6.5 million tonnes and 14 kg per inhabitant across the EU-27. That aggregate covers accommodation and other service activities alongside restaurants, and it measures collected waste rather than kitchen trim specifically, so it is not a yield figure. It is evidence that the difference between what you buy and what you serve is big enough for national statisticians to weigh.
Your own yields you have to measure. Weigh in, butcher, weigh out, once, per supplier and per cut. The number changes when the supplier changes, and it changes when the cook changes.
Operations that produce the same items every day are the easiest place to see this, and the most expensive place to get it wrong, because a yield assumption baked into a repeating spec sheet repeats with it — the mechanism we described for diet catering, where the same error comes back daily. Multi-site operators have the mirror-image problem: the same dish, different yields per kitchen, discussed in running several venues off one screen.
Pricing from plate cost: "multiply by three" is a hidden target, not a rule
Menu price, net of VAT = Plate cost ÷ Target food cost (as a decimal)
Target food cost— the share you intend the plate cost to take of the net price, as a decimal such as 0.30, not as 30.
Where does the target come from? From the previous sections: from your own measured base, not from an article.
That dish at PLN 12.23 of plate cost therefore prices at 12.23 ÷ 0.30 = PLN 40.77 net.
Implied target food cost = 1 ÷ Multiplier
- a multiplier of three implies exactly one third; the rule is a target in disguise.
That is the first problem: you are adopting a target chosen by whoever coined the phrase. The second is worse. The rule prices each dish in isolation, and the number you are actually judged on is not any single dish. It is the blend.
The mix moves the percentage without a single price changing
Sales mix (menu mix) is defined on our menu engineering page and quoted here unchanged: "Menu mix (MM) share — units sold of one dish, divided by the total units sold in the same menu category." The full method for using it lives there: see menu engineering. This page uses it for one purpose only — to show that the blend is not the average of the parts.
Blended food cost % = Σ (Plate cost × Units sold) ÷ Σ (Net menu price × Units sold) × 100
- both sums run over every dish sold in the period;
Net menu price— the menu price excluding VAT, PLN.
Take two dishes and change nothing except how many of each were ordered.
| Plate cost | Net price | Dish food cost | Units, period A | Units, period B | |
|---|---|---|---|---|---|
| Dish A | PLN 12.23 | PLN 40.77 | 30.00 % | 300 | 420 |
| Dish B | PLN 6.40 | PLN 28.00 | 22.86 % | 300 | 180 |
| Total units | 600 | 600 |
Numbers in this table and the two lines under it: the author's worked example, continuing the plate cost calculated above; the unit counts are chosen to hold the total constant so that only the mix moves.
Same recipes. Same suppliers. Same prices. Same number of guests served. The blended food cost moved 1.28 percentage points, because guests shifted toward the more expensive-to-make dish. No invoice explains it and no stocktake reveals it, which is why owners so often go looking for theft when the answer is on the till roll.
Separating the cost effect from the mix effect
When your number does move, you are usually looking at both causes at once. They can be separated with two more runs of the same formula, holding one thing constant each time.
Cost effect (p.p.) = Blended % at new plate costs and old mix − Blended % at old plate costs and old mix
Mix effect (p.p.) = Blended % at old plate costs and new mix − Blended % at old plate costs and old mix
Interaction (p.p.) = Total change − Cost effect − Mix effect
Suppose beef moved and dish A's plate cost rose from PLN 12.23 to PLN 13.10, while the mix shifted as in the table above.
| Scenario | Blended food cost % | Effect |
|---|---|---|
| Old costs, old mix (the base) | 27.09 % | — |
| New costs, old mix | 28.36 % | cost effect +1.27 p.p. |
| Old costs, new mix | 28.37 % | mix effect +1.28 p.p. |
| New costs, new mix (what you observe) | 30.02 % | total +2.93 p.p. |
| interaction +0.38 p.p. |
Numbers in this table: the author's worked example, computed from the plate costs, prices and unit counts printed above; the interaction term is the residual that closes the arithmetic, 2.93 − 1.27 − 1.28.
The interaction term is not a rounding error and should not be hidden. It is the extra cost of selling more of the dish that also got dearer — real money, attributable to neither cause alone. Notice too that either effect on its own would have looked modest, and that a venue tracking only supplier prices would have found 1.27 of the 2.93 points and stopped.
This decomposition is where price and menu belong. What happens inside the kitchen — the gap between what your recipes say you should have used and what the count says you actually used, and its operational causes — is a different reading with a different formula, and it has its own page: theoretical vs actual food cost.
The denominator: drinks, Polish VAT, and what counts as food sales
Two questions decide your denominator, and in Poland the second one has a statutory answer.
Should drinks be in it? They may be, or may not be, but the numerator and the denominator must agree. Beverage cost belongs with beverage sales; food cost belongs with food sales. Mixing one side and not the other produces a ratio that drifts with your drinks attachment rate and answers nothing. The GUS split quoted above — production versus goods for resale — is the same boundary drawn at national scale.
Which VAT rate comes out? In Poland, one bill routinely carries two.
So a PLN 100 gross bill of PLN 70 food and PLN 30 drinks is 70 ÷ 1.08 + 30 ÷ 1.23 = 64.81 + 24.39 = PLN 89.20 net. Dividing the whole bill by 1.08 instead gives PLN 92.59, overstating net sales by PLN 3.39.
And because the error tracks the drinks share, it moves with the mix too: the same shortcut flatters you more in a bar-heavy week than in a lunch-heavy one.
Numbers in this paragraph: the two VAT rates are from the statute cited in the sentence before; the bill arithmetic is the author's worked example.
What food cost percentage cannot tell you
It is a ratio, and ratios are silent about size. A percentage that improved while covers collapsed is not good news. It also cannot tell you why it moved — this page gives you price and mix, and there the boundary sits deliberately.
| Question | Where it is answered |
|---|---|
| Where the gap between recipe and stocktake comes from, and what causes it operationally | Theoretical vs actual food cost |
| Food and labour read as one number | Restaurant prime cost |
| Which dishes to protect, re-cost, reposition or remove | Menu engineering |
| How to put any of this on a screen you look at daily, and at what frequency | The restaurant KPI tree page of this series covers screens, thresholds and cadence |
Getting the inputs to arrive reliably is its own problem, and mostly a plumbing one: supplier documents, delivery notes and till exports that agree with each other. That is what our integrations and data work and automated reporting exist for, and the finance layer is where cost of food sold, sales and inventory value end up in the same place. If you want the same numbers projected forward rather than only backward, that is forecasting and scenario modelling; if you want the system to say which of the two effects above is moving, that is the decision engine. The practical background on supplier and delivery flows is in our article on restaurant automation, and on which figures an owner genuinely reads each day in reporting automation. Everything we do for food service sits together on Aura for restaurants.
Frequently asked questions
How do you calculate food cost percentage?
Take opening inventory, add purchases, subtract closing inventory, then correct for transfers to and from the bar and for staff meals. Divide that adjusted cost of food sold by food sales for the same period — net of VAT, discounts and voided bills — and multiply by one hundred. The three inventory numbers must be valued the same way at both ends of the period, or the result measures your valuation method rather than your kitchen.
What is a good food cost percentage for a restaurant?
There is no benchmark on this page because we could not find a citable source for one. The ranges circulating in the industry trace back to software vendors' marketing material and to guides assembled from that material, with no stated sample, period or definition; the one genuinely surveyed report is paid, American, and built on cost definitions a Polish P&L does not use. The comparison that does work is against your own earlier periods, in your own format, on your own current menu.
Should drinks be included in food cost?
Either choice can be defended, but the numerator and denominator have to match. If beverage cost is in the top, beverage sales must be in the bottom. Keeping them separate is usually more informative, because the two behave differently — Polish official statistics separate them too, splitting gastronomy revenue between own production and goods bought for resale.
What is the difference between food cost percentage and plate cost?
Plate cost is one dish, in PLN: the yield-adjusted sum of everything that reaches the plate, including sauce, garnish and anything served uncharged. Food cost percentage is a whole period, as a ratio. You can hold every plate cost constant and still watch the percentage move, because the percentage is a blend weighted by how many of each dish sold.
How does yield change the real cost of a dish?
Yield is usable weight divided by purchased weight, and it converts a supplier's price into the price of what actually reaches the plate. You divide the purchase price by the yield expressed as a fraction: a price of PLN 34.00 per kilogram at a yield of 0.68 is PLN 50.00 per usable kilogram. Multiplying instead of dividing, or using 68 in place of 0.68, understates every dish built on that ingredient.
Is the "multiply by three" pricing rule reliable?
It is a target dressed as a rule. Multiplying plate cost by three always implies a food cost of exactly one third, whatever the dish, because the multiplier is the reciprocal of the target. If you have chosen that target deliberately from your own measured base, the shortcut is harmless; if you have not, you have adopted somebody else's assumption. Either way it prices dishes one at a time and ignores the blend, which is what you are actually measured on.
Can my food cost percentage change if I have not changed a single price?
Yes, and this is the most common surprise in the metric. The blended percentage is weighted by units sold, so a shift in what guests order moves it while every recipe, supplier and menu price stays exactly where it was. In the worked example on this page the same six hundred covers, the same two dishes and the same prices produce a swing of more than a percentage point purely from the mix.
What should I compare my food cost percentage against?
Your own previous periods first, holding the definitions constant; then the same period a year earlier, which absorbs seasonality; then, separately, the same figure recalculated on last period's mix, which tells you how much of the change was menu rather than money. External ranges are not a comparison base, because none of the published ones state what they measured or on whom.
Work out the two numbers from your last complete period, run the mix decomposition on them, and you will know whether to call your supplier or read your till roll. Before you put the result away, record this period's mix alongside it — how many units each dish sold — because without it the next decomposition has nothing to stand on, and you are left again with a blended percentage that does not say what happened.