Four events collapse into one line in a daily total: a discount, a void before the bill closes, an annulment after it, and a comp. They mean different things and need different denominators. Judging them by a single combined percentage produces either false alarms or blindness, and usually both in the same month.
Four events that arrive in the same line of the daily total
The end-of-day summary of almost any restaurant point-of-sale system carries a line called "reductions", "corrections" or simply "minus". Underneath that one number sit four completely different things that happened in the dining room during the shift.
A guest was given ten percent off because the table waited too long. A server tapped the wrong beer and removed it from an open bill three seconds later. A closed bill turned out to be wrong and had to be cancelled. A dessert went out to a regular free of charge.
The first cost you margin on a dish that was cooked and eaten. The second cost you nothing. The third is a bookkeeping event with a legal shape. The fourth cost you the whole food cost of a plate and returned no revenue. They are not degrees of one problem, and a venue that watches them as one number is watching noise.
This page is about the management side of that line: what each event does to revenue, to food cost and to trust in your own reports, and how to tell an unusual pattern from an accusation. The legal side of the till — what the register must record, when the daily report is issued, what is kept and for how long — belongs to the fiscal cash register and the close of the day and is not repeated here.
Reduction line — the figure an end-of-day summary prints for everything that lowered the collected amount relative to the menu price of what was entered. A display convention, not a measurement: nothing in it says which event produced it.
Discount, void, annulment, comp: what actually separates them
The four differ along one axis above all: whether the food left the kitchen, and whether money came back for it.
A discount is a sale, not a cancellation
A discount lowers the price of something produced and consumed. The dish was cooked, the ingredients are gone, part of the money arrived. This is a sale at a lower price: its whole effect lands on revenue and margin, never on the count of dishes produced.
Discount — a reduction of the price at the moment of sale. The item is produced, delivered and paid for in part; only what you collect changes.
A void happens before the bill exists
A void is the removal of an item from a bill not yet closed. Usually it means one thing: somebody's finger landed on the wrong tile. The kitchen may not have started the item, and no fiscal document exists, because a receipt is issued at the moment of sale, not of ordering.
Void before close — removal of an item from an open bill. Usually an entry error, and usually it says almost nothing about money. Its natural unit is the item, not the zloty.
An annulment concerns a sale that did not happen
Polish law names this event directly, and it is the one place where this page touches the statute. The Minister of Finance regulation on cash registers defines an "annulled fiscal receipt" as a fiscal document issued by the register during a sale, confirming that a started sale was not completed (Rozporządzenie Ministra Finansów w sprawie kas rejestrujących, Dz.U. 2025 poz. 845, § 2 pkt 18, opened 28 August 2026). The nominative form of the term, paragon fiskalny anulowany, sits in § 2 pkt 1, where the act lists the fiscal documents of both online registers and registers with an electronic or paper copy; the definition in § 2 pkt 18 carries it in the locative. Both spellings were taken from the act by searching the string, and everything else here is rendered in English.
Two management facts follow. An annulment is a named object with its own document — it does not vanish, and it is not a void. And because the regulation names it, the register can count it, and a report that counts it exists.
Annulment — a sale started and not completed, carrying its own fiscal document under the Polish regulation on cash registers. What the law requires around it — issuance, retention, the daily close — belongs to the fiscal cash register page.
A comp is the only one that eats product
A comped dish is produced and given away. No revenue arrives, and the whole ingredient cost has already left the store room. It is the only one of the four that shows up directly in the gap between what recipes say the food should have cost and what it did — the gap analysed on theoretical versus actual food cost.
Comp — a dish released without charge by a decision of the venue. Product consumed, revenue zero. The only one of the four that hits food cost directly.
What leaves a fiscal trace and what leaves only a management one
This decides where you look for each number, and it is why one line in the daily summary can never answer the question.
| Event | Where the trace lives | What the Polish regulation calls it | Where you read it |
|---|---|---|---|
| Discount | Point-of-sale sale record | No separate name — it is part of the sale | Sales report by item and by server |
| Void before close | Point-of-sale log only | No separate name — nothing was sold yet | Terminal or user log, not the fiscal report |
| Annulment | Fiscal document | paragon fiskalny anulowany, § 2 pkt 1; defined in § 2 pkt 18 | Daily fiscal report and the register's own counters |
| Return or accepted complaint | A separate register required by law | Returns of goods and accepted complaints of goods and services, § 3 ust. 2–3 | The separate register, not the sales register |
| Obvious entry error after the receipt | A separate register required by law | An obvious mistake, § 3 ust. 4–5 | The separate register of obvious errors |
| Comp | Kitchen and stock records | No name in the regulation at all | Food cost reconciliation, not the till |
Three rows deserve reading twice.
The regulation states plainly that returns of goods and accepted complaints are not entered into the sales register and go into a separate one, carrying seven fields including the date of sale, the gross amount, the sales document and a protocol signed by seller and buyer — or an official note in its place (§ 3 ust. 2 and ust. 3 pkt 1–7). What the act calls an obvious mistake is corrected the same way: its own separate register, plus a fresh entry at the correct amount (§ 3 ust. 4–5). The wordings here are rendered in English; the binding text is Polish and sits at the link above.
For registers with an electronic or paper copy, the regulation prescribes what the daily fiscal report must contain: among the required positions are the total number of annulled fiscal receipts and their value (§ 29 ust. 1 pkt 14 and pkt 15). That paragraph opens by naming its own scope — registers with an electronic or paper copy — so it is not a universal statement about every device in every venue, and is not presented as one here.
The management consequence is blunt. Two of the four events already have a legally shaped counter; two exist only in whatever your point-of-sale system chose to log. If it does not log voids by user and by hour, no amount of reading the fiscal report will produce that number. That is a systems question, and it is why integrations between tills, accounting and stock belong in the conversation before any threshold does.
Why the four cannot be watched as one combined number
Suppose a month ends with a "reductions" line of 8 340 PLN against 84 000 PLN of pre-discount revenue. That is 9.93 %, and it means nothing.
Split it and it becomes readable: 6 300 PLN of discounts, 1 260 PLN of voided items re-entered within seconds that cost nobody anything, and 780 PLN of comped food at cost. The three parts move for different reasons and have different denominators. Watched together they produce a number that rises when marketing runs a legitimate promotion and falls when a shift stops using the system properly — the wrong direction on both events.
The second reason is uglier. A combined percentage forces you to compare things measured in different units. Voids are naturally counted in items, comps in ingredient cost, discounts in revenue given up. Adding them into one bucket of money means quietly deciding that a voided beer re-entered a second later is the same size of problem as a free main course. It is not.
A denominator of its own for every event
Each of the four events answers a different question, and each has its own base.
Discount share, and the base that inflates it
Discount share = Sum of discounts ÷ Revenue before discounts
Sum of discounts— total value given away as price reductions in the period, PLN;Revenue before discounts— what the same sales would come to at menu price, PLN;- the result is a dimensionless share between 0 and 1.
Backwards: 84 000 × 0.075 = 6 300 PLN.
Now the trap. Most point-of-sale systems put revenue after discounts in front of you — 77 700 PLN here — because that is the money that arrived.
The share is overstated, and the overstatement grows with the discount:
| Discounts given, PLN | Correct share (÷ 84 000) | Share on the after-discount base | Difference, percentage points |
|---|---|---|---|
| 2 100 | 2.50 % | 2.56 % | 0.06 |
| 4 200 | 5.00 % | 5.26 % | 0.26 |
| 6 300 | 7.50 % | 8.11 % | 0.61 |
| 12 600 | 15.00 % | 17.65 % | 2.65 |
| 21 000 | 25.00 % | 33.33 % | 8.33 |
At a small discount level the bases barely differ and nobody notices which is in use. In a promotional month they diverge by eight percentage points. The instrument goes wrong precisely when you most need it right, and in the alarming direction — a promotional week reads as a leak.
A second base question sits underneath, and it is not ours: whether revenue is taken gross or net of tax. Two honest reports can disagree by the tax rate alone. That is settled once, for every ratio in the venue, on the gross-or-net denominator page; this page assumes you have settled and labelled it.
Void share, counted in items
Void share = Voided items ÷ Entered items
Voided items— item lines removed from open bills in the period, items;Entered items— all item lines entered in the same period, items;- the result is a dimensionless share.
Backwards: 4 812 × 0.0349 = 168 lines, and 4 812 − 168 = 4 644 lines sold.
Count the same event in money and the picture changes for no good reason.
The same shift, the same behaviour, two numbers differing by more than a factor of two. The question is "how often does an entry have to be undone", not "how much money was involved". An error on a 4 PLN espresso and one on a 90 PLN steak are the same action by the same person; counting them in money says the steak matters twenty times more, which describes the menu, not the work.
Comp share, counted at cost
Comp share of food cost = Cost of comped items ÷ Cost of all items released
Cost of comped items— ingredient cost of everything given away free, PLN;Cost of all items released— ingredient cost of everything that left the kitchen, sold or not, PLN;- the result is a dimensionless share.
This one is measured at cost rather than menu price because its purpose is to explain a gap in food cost.
The difference is 0.93 percentage points, and it is exactly the comps. If the gap between theoretical and actual food cost was two percentage points that month, just under half of it is now explained and needs no investigating. Measuring comps at menu price gives a bigger, more alarming and unusable number.
People, hours, items: three cuts of the same month
One month of till data answers three different questions depending on how you slice it, and each slice has its own failure mode.
| Cut | What it shows | What it is easy to mistake for a problem | How to check |
|---|---|---|---|
| By person | Whether one server's rate stands apart | A trainee, or whoever works the busiest station | Compare against hours worked and items entered, not headcount |
| By hour | Whether the rate spikes at particular times | Shift change and the last hour before close, when corrections pile up | Look at the same hour across many days, not at one evening |
| By item | Whether particular dishes generate corrections | A new dish, or one with size or garnish variants | Compare against the age of the item and its number of variants |
The order matters. Start with the item cut: a menu problem explains away much of what otherwise looks like a people problem. Then hours, because a process problem tends to be tied to a moment in the day. Only then people, once the first two cuts have removed most of the innocent explanations. The other way round starts with the most accusatory cut and reasons backwards, which reliably produces the wrong conversation.
Every cut needs the data pulled the same way each month — a reporting job rather than an analytical one, which is what reporting that assembles itself and dashboards for the numbers you watch exist for; the numbers an owner actually looks at covers how a small set of them earns a place on a screen.
An even series means a procedure, not a coincidence
This is the measurement most venues never make, and the one that turns a suspicion into a fact.
Evenness = Standard deviation of the daily count ÷ Mean daily count
Standard deviation of the daily count— spread of the count across the days of the period, in the units of the count;Mean daily count— arithmetic mean of the same daily counts;- the result is dimensionless: both parts carry the same unit.
Take two servers over ten shifts. The first records 3, 1, 4, 0, 2, 5, 1, 3, 2, 4 voids a day — twenty-five in all, mean 2.5, standard deviation 1.5, evenness 0.60. The second records 2, 2, 2, 3, 2, 2, 2, 3, 2, 2 — twenty-two, mean 2.2, standard deviation 0.4, evenness 0.18.
Their monthly totals are almost identical, and a report that shows only totals shows nothing. But the two series do not describe the same thing. Guests are irregular, kitchens are irregular, and a number driven by guests bounces around — the first server's series is what that looks like. A number that barely moves is not being driven by guests; something is producing it the same way every day. That can be a workflow — a station where every order is entered and then adjusted — or a routine. The measurement does not say which; it says where to look, which is far more than a monthly total ever did.
The rule generalises beyond the till. Regularity in a number that ought to be irregular is information, and the information is always "a mechanism", never "bad luck".
A worked case where a high void rate was sound work
The numbers below are a worked example to check against your own data, not a report about any venue.
Split the 4 812 entered lines of that month by terminal.
Three checks come first.
First, the evenness of the bar's daily series. If it sits near 0.18 rather than near 0.60, the cause is a mechanism.
Second, the time between the void and the replacing entry. If nearly every voided line is followed within seconds by a similar line at the same table, nothing was removed from anybody's bill — an entry was corrected.
Third, the item cut. If the voids concentrate on drinks that come in two sizes, the mechanism has a name: the bartender enters the order while the guest is still choosing between small and large, and every change of mind becomes a void. The rate is high because the work is shaped that way, and it stays high until the ordering flow changes.
After those three checks you have either an explanation or a genuinely open question — and in the second case, a dated, countable observation to bring to the shift instead of a feeling.
An alert threshold that does not accuse in advance
There is no published industry norm for the share of discounts, voids or comps in a restaurant. We looked: neither the European statistical catalogue nor the Polish national statistical office publishes a series for any of the three. Any figure quoted as "the industry standard is X percent" arrives without a method — without the denominator, the period, or which of the four events it counts — and a percentage without a method is a rumour, not a benchmark.
So the threshold is built from your own series, in four steps.
- Measure twelve weeks of each event separately, each on its own denominator. Nothing before that is a baseline; it is an anecdote.
- Take the median week, not the mean. One promotional weekend drags a mean upward and quietly raises your alarm level for the rest of the year.
- Set the band from the spread you observed, not from a round number. If the weekly void share moved between 2.8 % and 4.1 % across twelve normal weeks, 4.1 % is not an alarm; it is Tuesday.
- Make the alarm about change, not level. A rate steady for three months that then moves is worth a look. A rate that has always been high and never moved describes your process, and changing a process is a different project from investigating a person.
A threshold built this way has a property a borrowed number never has: when it fires, it says something changed in your venue, and there is no honest way to argue with it. That makes it usable in a conversation. Turning the observation into a dated, assigned piece of work rather than a floating worry is what task tracking off the back of a measurement is for.
Five minutes in the morning, one hour a month
The daily and monthly views answer different questions and do not merge.
Every morning, five minutes. Look at yesterday's four numbers side by side, not summed, and ask one thing: did any of them move away from where it has been sitting? You are looking for a step, not a level. Discounts and comps are worth a glance in money, voids in count. A morning that produces no question is a successful morning, and most will.
Once a month, one hour. Rebuild the three cuts — people, hours, items — recompute the evenness of every series with more than a handful of events, and re-derive the band from the last twelve weeks so it follows the venue rather than freezing at the quarter you first measured. Check too that the four events are still entered as four; the fastest way to lose this measurement is for a shift to start recording every comp as a hundred percent discount because it takes one tap fewer.
Once a quarter, with the fiscal side. The annulled receipt counters in the daily fiscal report and your management count of annulments should tell a consistent story. Where they do not, it is either a definition drifting or a device configured differently from the one beside it, and both are worth ten minutes.
For one site this fits in a spreadsheet. For several it does not, because the interesting comparison is between locations, and that is only honest when every site enters the four events the same way — the problem described in running several restaurants off one screen and, at the system level, in which layer to add next.
Bringing it to the shift without turning a measurement into an accusation
A number about a person is a loaded object, and whether it lands well is mostly a matter of the order of the sentences.
Open with the mechanism, not the person. "The bar shows seven and a half percent of lines voided and the dining room two — what is different about how orders get entered at the bar?" is a question about a process. The same figure as "you void four times as often as anyone else" is a verdict passed before any investigation, and it produces a defensive answer whether or not there is anything to defend.
Bring the three cuts, not the total. If you have checked the item and hour cuts, say so: it shows the person was not the first place you looked.
Say what you measured and how — the denominator, the period, the definition of the event. A shift that can see how the number was built can also correct it, and a correction from the floor ("half of those are the two-size drinks") beats a confession.
Never present the evenness figure as proof of anything about a person. It signals that a mechanism exists, and mechanisms include badly designed workflows, undertrained staff, a menu with too many variants and a broken printer — all commoner than the explanation people jump to.
And keep the measurement running afterwards. A number pulled only when something feels wrong reads as surveillance. A number on the screen every morning, visible to everyone, that mostly says nothing is simply part of how the venue is run — the principle described for operational numbers in restaurant automation across bookings, suppliers and reviews and in the inventory of what can be handed to a system. Where these four ratios sit against everything else you track is in the tree of restaurant indicators, and how a discount changes the numerator of the average bill is the subject of the average check against the average guest.
Frequently asked questions
What is the difference between a void and an annulment?
A void removes an item from a bill not yet closed, so no fiscal document exists and nothing has been sold. An annulment concerns a sale started and not completed, and under the Polish regulation on cash registers it produces its own fiscal document, the annulled receipt. One lives only in your point-of-sale log; the other has a legal shape and a counter.
Which of these events actually costs the venue product?
Only the comp. A comped dish is cooked and given away: the ingredients are gone and no revenue arrives, so the whole ingredient cost lands in the gap between theoretical and actual food cost. A discount costs margin but not product, and a void before the bill closes usually costs nothing, because the item was never made.
Why should discounts be divided by revenue before discounts?
Because the question is how much of the potential revenue you gave away, and the potential revenue is the pre-discount figure. Dividing by revenue after discounts overstates the share, and the overstatement grows with the discount: at 6 300 PLN of discounts on 84 000 PLN the bases read 7.50 % and 8.11 %, at 21 000 PLN they read 25.00 % and 33.33 %.
Why count voids in items rather than in zloty?
Because a void is an action, and every action counts as one regardless of the item price. In the worked month, 168 voided lines out of 4 812 is 3.49 % in items and 1.48 % in money — the same behaviour, two numbers. Counting in money makes a mistyped steak look twenty times more serious than a mistyped espresso, which describes your menu rather than your work.
Can a high void rate be perfectly normal?
Yes, and it usually is when a whole station shows it. A bar where orders are entered while the guest is still choosing between two sizes generates a void on every change of mind. Before treating a high rate as a problem, check whether the voided lines are re-entered within seconds, whether they cluster on items with variants, and whether the station was always like that.
What does an unusually even pattern in the till mean?
It means a mechanism rather than guests. Guest behaviour is irregular, so an event driven by guests has a spread comparable to its mean. When the daily counts barely move — an evenness near 0.18 against a more typical 0.60 — something produces the number the same way every day. That can be a workflow or a routine; the measurement tells you to look, not what you will find.
How do I set an alert threshold that does not accuse anyone?
Build it from your own twelve weeks rather than a borrowed figure, take the median week instead of the mean, set the band from the spread you observed, and make the alarm fire on change rather than level. There is no published industry norm for any of these four events, so a threshold quoted without its method and denominator is a rumour and will fire on sound work.
Take last month and split it into four events with four denominators of their own. Start with the one whose daily series is too even to have been produced by guests — evenness means a procedure, not a coincidence. The rest of the restaurant-economics pages sit together in the restaurant section. If you want the four numbers on a screen every morning without anybody assembling them by hand, that is a question of end-to-end analytics for the venue rather than of another spreadsheet.