Food Waste in Food Service: The Numbers and the Method

Food waste is counted officially, and that count has a cut of its own

There is an official European count of food waste, published by activity, and food service is one of the activities in it. That is the useful half. The other half is that the count was never built to describe one restaurant, and every mistake people make with it comes from quoting the first half without the second.

The dataset is env_wasfw, "Food waste and food waste prevention by NACE Rev. 2 activity - tonnes of fresh mass" (Eurostat, dataset env_wasfw, series last updated 18 February 2026, opened 28 August 2026). It is an annual series running from 2020 to 2023, so 2023 is the newest year available, not a stale pick on our side.

Three labels travel with every figure in it, and all three change what the figure means:

• the operation in the rows read here is waste collected — mass collected as waste, not mass a kitchen felt it had lost;
• the material is food waste — bio, household and similar waste, a waste-statistics category, not a menu category;
• the unit of mass is fresh mass, the weight as it is, water included. A kilogram of soup and a kilogram of dry bread are one kilogram each.

None of that is a criticism of the statistic — it is doing what a national account is supposed to do. It is a criticism of the sentence "the industry wastes X" when X was lifted out of this dataset without any of the three labels attached.

Two units in one row, and they do not move together

The same row is published twice: in tonnes and in kilograms per inhabitant. Put the EU27 food-service line side by side and something odd shows up. The tonnage is the same number in both years — 6 500 000 tonnes in 2022 and in 2023 — while the per-inhabitant figure moves from 15 to 14.

The explanation is not a contradiction in the source, it is resolution: the EU27 aggregate is rounded to the hundred thousand tonnes, the per-inhabitant figure is published as a whole kilogram, and between them there is not enough precision left to carry a small year-on-year move. The rule that follows is the first thing this page asks of you: read this row as a level, not as a trend. Anyone who draws an arrow through two annual points at this rounding is drawing it themselves.

What the food service cut covers, and what it does not

Food service cut — the code I55_I56_N-S_FOOD in the Eurostat dataset. It is an aggregate: it puts accommodation and food and beverage service together with a group of adjacent sections of the economy. It is not a code for "restaurants", and reading it as one changes the meaning of every share computed from it.

Eurostat's display label for this code is short: Restaurants and food services. The code string is longer than the label, and the code string is the honest description. Opened separately in the NACE Rev. 2 code list (Eurostat, NACE Rev. 2 code list, opened 28 August 2026), the parts read:

In the same code list I55_I56_N-S_FOOD is marked as an aggregate and carries no explanatory label at all — while some neighbouring aggregates in that very list do carry one, spelling out what they include and exclude. So the boundary of this cut cannot be read off the label, and there is no explanatory note to read it off either. What is left is the code string, and it names accommodation next to food and beverage service and then a run of whole sections from N to S.

What that means in a sentence you can repeat

The hotel breakfast buffet, the school canteen, the hospital kitchen, the staff canteen inside an office building and the restaurant on the corner all sit inside the same published number. The restaurant is in there. It is not the whole of it, and nothing in the source says what fraction of it the restaurant is.

Call this line "restaurants" and every ratio built on it is wrong in a direction nobody can estimate. That is worse than a wrong number: a wrong number can be corrected, an unknown-direction one cannot even be checked. The label on the box and the contents of the box are two different facts, and only one of them was measured.

Poland and the EU27 across two years: tonnes and kilograms per inhabitant

Here is the whole reading. Twenty-four values were requested and twenty-four came back — no gaps, nothing interpolated on our side (Eurostat, dataset env_wasfw, opened 28 August 2026).

One thing about precision, because it decides what you may do with the table: the Polish figures arrive to the single tonne, the EU27 figures already rounded to the hundred thousand on every line. Any percentage computed from the EU27 column inherits that rounding. And one label that is easy to skip: the source’s full name for the "Total" row is "Total (aggregate changing according to the context)" — it is the total of the count as assembled, and the denominator used below, not a promise that every activity is inside it.

Why the food service share of the whole count is smaller than the conversation suggests

Public conversation about wasted food tends to picture a restaurant bin. The count does not agree. The arithmetic is ours, done on the published figures, and shown so you can repeat it:

• Poland, 2023. 293 062 ÷ 4 651 137 = 0.0630, that is 6.30 % of the whole Polish count.
• Poland, 2022. 264 546 ÷ 4 545 178 = 0.0582, that is 5.82 %.
• EU27, 2023. 6 500 000 ÷ 57 900 000 = 0.1123, that is about 11.2 %, and this one carries the rounding of both figures.

So in Poland the food-service line is roughly one part in sixteen of the national count, and in the EU27 aggregate roughly one part in nine. That gap is itself a caution: two statistical systems assembling the same category can differ by that much, and it is not evidence that Polish kitchens are twice as careful as European ones.

That is not a trend and must not be read as one — a change in a statistical count can come from the way the count is assembled as easily as from the thing counted. If you want to know whether your own line is flat, the count will not tell you; you have to weigh. The same discipline applied to your own reporting is the subject of the numbers an owner actually looks at.

Households against venues: where the mass actually sits

The same dataset publishes a households line, and it settles the argument about proportion.

In Poland in 2023 households account for 2 561 260 tonnes.

Against the food-service line of 293 062 tonnes it is 8.7 times as much. Per inhabitant the two lines read 70 kilograms and 8 kilograms.

There are two wrong conclusions here and one right one. The first is relief — "so it is not us". The share of a national total says nothing about whether your walk-in cooler is throwing away money this week. The second is the mirror image: treating the household figure as a benchmark and telling yourself a venue should be near 8 kilograms of something. Per-inhabitant units do not apply to a venue at all, for the reason in the next section. The right conclusion is narrow: the national picture tells you where the public conversation is aimed, and nothing about your kitchen. For your kitchen there is only measurement.

The national figure cannot be carried onto your venue, and here is why

Three reasons, and each alone is enough.

Kilograms per capita — the unit KG_HAB, and that is the name the dataset itself gives it. It divides a national mass by a national population. It is built for comparing countries, and its denominator does not exist inside a venue: a restaurant has no population. Borrowing this unit is not an approximation, it is a category error.

The cut is an aggregate. The line puts your restaurant next to hotels, canteens, schools, hospitals and a run of service sections, and the source does not publish the split. No arithmetic gets from the box back to your kind of thing.

The operation counted is collection. Your four buckets describe something else — mass that stopped being sellable inside your walls, whatever happened to it afterwards. The two definitions overlap without being the same, and an overlap is not a conversion factor.

Where this page stops

Handing food waste over, and keeping a record of what was handed over, is a separate obligation under Polish waste law. This page is about management measurement, and it deliberately prints no article numbers and no waste codes: that side is held in full by the neighbouring page on the waste register and BDO for a venue.

Why two pages rather than one: they are two accounts of the same mass, and they are not obliged to agree. The legal record follows what physically left the premises and to whom. Your four buckets follow what stopped being sellable, including things that never left as waste at all — a case returned to a supplier at the door, or product re-purposed into staff meals. A manager who expects the two totals to reconcile will spend a week looking for an error that is not there.

Four buckets of your own measurement: receiving, kitchen, plate, display

Four buckets — a measurement of your own: receiving (what was not accepted, or was accepted and should not have been), kitchen (trim and errors), plate (what came back from the guest), display (what did not sell before its date). The split is not an official classification. It exists because the four are fixed by four different actions, and a single total tells you to fix nothing.

There is no official breakdown of this kind to borrow, and we checked before writing the section. A sweep of the full Eurostat catalogue in plain text returns exactly two food-waste products — the dataset used above and a derived indicator table on the same series — and nothing at any stage inside an establishment (Eurostat, full catalogue of datasets, opened 28 August 2026). The Polish statistical office's environment and energy section lists environmental protection yearbooks and municipal waste studies, and no food-loss breakdown by stage either (GUS, environment and energy thematic section, opened 28 August 2026).

So the four buckets are a method here, not a norm, and this page names no percentage that any of them "should" be. A foreign share whose measurement method is not described is not a number, it is a rumour.

Receiving

Everything that goes wrong before the goods are yours: short weight against the delivery note, a case accepted in a state you would not have accepted if anyone had opened it, a substitution nobody agreed to, a temperature failure noticed at the door. Small in mass, large in money, and the only bucket cheaper to fix than to measure — the fix is a conversation with a supplier, not a change in the kitchen. The price side of it is a different measurement, the gap between the price agreed and the price invoiced, and that belongs to purchase price variance and the three documents.

Kitchen

Trim, peel, bones, over-production, failed batches, portions started and abandoned. Part of this bucket is structural — a carrot has a peel — and part is error, and the whole value of weighing it is telling those two apart. Structural trim moves when the specification or the supplier format moves; error moves when the prep schedule moves.

Plate and display

The plate is what the guest left: the bucket everyone pictures and, in most kitchens that start weighing, not the one that pays for the exercise. It is also the slowest to interpret — a plate back half full can mean portion size, the dish, or three courses ordered. Weigh it, read it last. Display is everything on a counter, case or buffet that reached its date unsold; it is a forecasting result rather than a kitchen result, which is why it is tied most directly to demand and load forecasting.

How to weigh so it takes five minutes a shift, not half a day

Measurement that costs half a day gets done for two weeks and then quietly stops, and a series that stops is worth less than one that never started: it leaves a false impression that the question was settled. Everything below keeps the cost per shift small enough that the habit survives.

• One scale per bucket, where the mass appears. A scale that requires a walk gets skipped.
• Tare the containers once and write the tare on them. Subtracting a bin weight from memory is where the first errors arrive.
• Record four fields and no more: product group, mass, bucket, shift. Dish names, reasons and staff names turn a five-minute task into a form nobody fills in.
• Weigh by product group, not by dish. Vegetables, meat, fish, dairy, bakery, prepared — six groups are enough to act on and few enough to remember.
• Decide in advance what you will not weigh and write that decision down. Liquids poured away, ice, coffee grounds — pick one answer and keep it for the whole period.
• Run two full weeks before reading anything. One week contains one of every weekday and none of the variation between weeks.

The recording belongs in the same place as every other shift task, so a missed weighing is visible the same evening rather than at the end of the month — that is what shift tasks and checklists are for, and what is worth handing to a system at all is the subject of what can realistically be handed over.

Once the numbers exist, they need to arrive somewhere without anyone retyping them. A weighing sheet copied into a spreadsheet by hand on Monday morning will be copied for three Mondays. Sending it straight into the venue's analytics and onto the screens the manager already looks at is not a refinement, it is the condition under which the measurement lasts past the first month.

Losses in kilograms and losses in zloty: two pictures of one thing

Mass and money rank the buckets differently, and that difference is the whole point of computing both. Vegetables are heavy and cheap; fish is light and expensive. Rank by mass and you spend the month redesigning vegetable prep; rank by money and you may find the same month's money sitting in two crates that should never have been accepted.

Loss share by mass = Mass of losses over the period ÷ Mass purchased over the period

• Mass of losses over the period — the sum of all four buckets for one product group, in kilograms, over one period;
• Mass purchased over the period — receipts of the same product group over the same period, in kilograms, taken from the goods-received records.

Both sides are kilograms, so the result is a dimensionless share.

Reverse check: 640 × 0.1219 = 78.0 kg.

Compute this per product group, never across the whole purchase. In a single total, vegetable trim disappears into the weight of drinks and flour, and the figure you get back is too blunt to point at anything. The relationship between purchases, stock and what actually moves through the store is held by inventory turnover in a restaurant.

Cost of losses = Σ (Mass of the bucket for an item × Purchase price of that item)

• Mass of the bucket for an item — kilograms of one item in one bucket over the period;
• Purchase price of that item — the price you paid per kilogram, from the invoice, not the menu price.

Kilograms multiplied by PLN per kilogram gives PLN. Worked on the same example inputs: 78 kg of vegetables at 4.20 PLN per kilogram is 327.60 PLN, and 26 kg of meat at 31.50 PLN per kilogram is 819.00 PLN, giving 1 146.60 PLN for the period.

Why the menu price is the wrong price here

Valuing a loss at the menu price produces a large, satisfying and misleading figure. The lost product was never sold, so its margin was never earned and was never lost either — it simply did not happen. Booking it inflates the number by the whole mark-up, and inflates the buckets unevenly, so the ranking sends you to fix the wrong thing. The purchase price is the money that actually left the account. The gap between what the food should have cost and what it did cost is a separate, complementary measurement, held by theoretical versus actual food cost.

What gets fixed first: usually not the plate, but the order to the supplier

Almost every venue that starts weighing expects the plate bucket to be the story, and in most first months it is not. The plate is the most visible bucket and the least controllable: it is downstream of portion size, of the dish, of what the guest ordered and of the mood of the table.

The receiving bucket is the opposite — small, unglamorous, and it responds to a single conversation. A supplier who knows that every delivery is opened and weighed behaves differently from one who knows deliveries are signed for at the door. Nothing in the kitchen has to change for that bucket to fall. Where the ordering and reception side is already systematised, the same records feed that conversation — the shape of it is in restaurant automation: reservations, suppliers and reviews.

Display is second, and it responds to arithmetic rather than discipline: it is a function of how much was put out against how many guests came, so two weeks of buckets read next to two weeks of covers say more about it than any amount of staff instruction. Kitchen is third because it splits — the structural half moves only when the specification or the supplier format moves, the error half when the schedule does. Plate is last, not because it does not matter, but because you can only read it correctly after the other three have stopped moving.

If the weekly reading has to be assembled by hand from four sheets, it will be assembled twice. Letting the reports build themselves is what turns a two-week experiment into a permanent series, and which system layer should own that series is worked through in CRM or ERP: which layer to add next.

How to be sure the measurement did not improve just because you started it

The effect of the measurement itself — the trap in this whole method: as soon as losses are weighed, they fall, without a single change to any process. People behave differently while being counted. This is not fraud and not an illusion, but it is also not an improvement you can bank.

The consequence is a rule about comparisons. Do not compare your first two weeks with anything except each other:

• Not with a foreign figure. There is no published norm to compare against, as the absence check above shows, and a number found in an article without its measurement method attached cannot be compared with anything.
• Not with your own later periods, calling the difference a result. The first fall is bought by attention, and attention is not a process change.
• Do compare week one with week two inside the same period, with the same rules and the same list of what is not weighed. That comparison is the honest one.
• If you change the method, the series starts again. Adding a bucket, changing what counts as a group, or moving the scale changes the number without anything real changing. Write the date of the change next to the series.

The one figure that survives comparison between your own sites

Loss per cover = Cost of losses ÷ Covers seated

• Cost of losses — the result of the previous formula, in PLN, for one period;
• Covers seated — guests seated over the same period, from the same period's records.

PLN divided by covers gives PLN per cover. On the worked example above: 1 146.60 ÷ 1 820 covers = 0.63 PLN per cover; reverse check, 0.63 × 1 820 = 1 146.60 PLN.

Mass and money both scale with the size of the venue, which makes them useless for comparing a forty-seat dining room with a hundred-and-twenty-seat one. Per cover they become comparable — between your own sites, in your own periods, on your own method. This is deliberately the same denominator used for energy on the neighbouring page, so two very different cost lines can be read on one scale: see what energy costs a venue.

And it is comparable because the method behind it is yours and identical on both sides. Hand it to someone whose four buckets are defined differently and it stops meaning anything — which is precisely what happened to every industry benchmark this page refuses to quote. If you are building the first version of that internal discipline, the four thresholds where small businesses usually start are set out in automation in a small business: four thresholds.

Frequently asked questions

How much food does food service waste in Poland?

The official count puts the food-service line at 293 062 tonnes in 2023 and 264 546 tonnes in 2022, against a whole-country count of 4 651 137 and 4 545 178 tonnes — that is 6.30 % and 5.82 %. The figures come from the Eurostat dataset env_wasfw, and they describe the aggregate activity cut, not restaurants alone.

What does the official food service cut actually cover?

The code is I55_I56_N-S_FOOD. Eurostat's display label reads "Restaurants and food services", but the code string names accommodation and food and beverage service together with whole sections from N to S — administrative services, public administration, education, health and social work, arts and recreation, other services. So hotel breakfasts, school canteens and hospital kitchens sit in the same figure as restaurants, in proportions the source does not publish.

Do households or restaurants waste more food?

Households, by a wide margin. In Poland in 2023 households account for 2 561 260 tonnes — 55.07 % of the national count and 8.7 times the food-service line. In the EU27 aggregate the ratio is about 4.8 times. This says nothing about whether an individual venue is losing money on waste this week.

Can I apply the national figure to my own venue?

No, for three reasons. The per-inhabitant unit divides by a national population, and a venue has no inhabitants. The activity cut is an aggregate that mixes your kind of establishment with several others without publishing the split. And the count describes waste collected, while your four buckets describe mass that stopped being sellable.

What are the four buckets a venue should weigh?

Receiving, kitchen, plate and display. Receiving holds what was not accepted or should not have been. Kitchen holds trim, failed batches and over-production. Plate holds what came back from the guest. Display holds what reached its date unsold. They are separated because each is fixed by a different action.

Should waste be valued at purchase price or menu price?

Purchase price. The lost product was never sold, so its margin was never earned and was never lost — pricing the loss at the menu price inflates it by the whole mark-up, and inflates unevenly: high-margin buckets look larger than they are, and the ranking sends you to fix the wrong bucket first.

Why do losses fall as soon as you start weighing them?

Because the measurement changes behaviour before it changes anything else. The first fall is bought by attention, not by a process change, so it must not be banked as a result. Compare week one with week two inside the same period, with the same rules, and treat any change of method as the start of a new series.

Two weeks of weighing across four buckets, then the first week read against the second — not against anyone else's number, because there is no published one to read against. For the rest of the restaurant economics this measurement sits inside, there is the restaurant section.