AURA for restaurants

We are not building another site with a menu and a booking form. AURA is an external AI department for restaurants and hospitality venues in Warsaw: it sees what is happening in your venue today and tells you what to do about it — before it shows up in the results of the month.

Restaurant economics: the guide

The pillars of a small venue’s economics — one at a time, each on its own page. This is not a dated feed: these texts do not age and they link to each other.

• Restaurant Prime Cost: the Formula, and Why Your Own Number Is the BenchmarkPrime cost packs the two cost blocks a restaurant owner can still change into one percentage. This page gives the arithmetic, the definitions that decide the answer, and the reason it prints no target band: the widely quoted range has no traceable publisher.
• Restaurant Profit Margin: Where the Money GoesA restaurant has three profit margins, not one, and only the last answers what the owner actually kept. The full P&L waterfall from the till to the bank account, depreciation included, plus the decomposition that explains sales rising while profit falls.
• Restaurant Break-Even Point: How to Calculate It on Your Own NumbersOne division answers it: fixed costs divided by the contribution margin ratio. The hard part is sorting a restaurant’s own costs into the two buckets that division needs — and this page does that, in money, in covers and in days, on your figures rather than a borrowed benchmark.
• Food Cost Percentage: The Formula, and Why There Is No Universal BenchmarkThe full arithmetic of food cost percentage for a restaurant owner: the inventory formula with its transfer and staff-meal corrections, plate cost with yield, pricing from plate cost, and a decomposition that separates the price effect from the sales mix effect. No benchmark band is quoted, and the page explains why one cannot be sourced.
• Theoretical vs Actual Food Cost: How to Read the VarianceTwo numbers for the same period: what the recipes say the food should have cost, and what the counts say it did cost. This page reads the gap between them — the formula, the sign, the split into usage and price, and the five places where product leaves without a sale.
• Restaurant Labour Cost Percentage: The Formula and How to Build Your Own TargetThe formula for restaurant labour cost percentage, the six items that belong in total labour cost, and what a Polish employer really pays on top of the wage — with every statutory figure printed next to the document it comes from.
• Sales per Labour Hour: The Staffing Reality CheckA shift can close exactly on its sales-per-hour target and still have a quarter of its hours standing in the wrong place. This page gives the arithmetic, the hour-by-hour cut that finds those hours, and the reason it prints no target figure of its own.
• Restaurant KPIs: build a tree, not a listA flat list of restaurant metrics cannot tell you what to do when two numbers move in opposite directions on the same day. A tree can, because its spine is arithmetic all the way down to net profit at the root. This page is the map for the whole cluster: the root, the four levers, the daily leaves, and the rule that keeps the tree from growing back into a list.
• RevPASH: revenue per available seat hourRevPASH puts time into the denominator: dining-room revenue divided by seats multiplied by the hours those seats were open for sale. It merges occupancy and spend into a single number, and this page gives the formula, the decomposition that tells you which of the two factors failed, and the reason it prints a method instead of a benchmark.
• Menu Engineering: Which Dishes Actually EarnMenu engineering scores every dish on two axes — how often it sells and how much cash each sale contributes — and hands each of the four quadrants a different action. This page gives the full arithmetic: where both dividing lines come from, why the vertical axis is money rather than food cost percentage, and which of your own conventions quietly moves a dish between quadrants.
• Restaurant Inventory Turnover, Par Levels and the Reorder PointThree numbers, one calculation: how fast stock turns, how much of it to hold, and when the order has to go out. This page runs a single item through all three, shows that turnover is largely produced by your delivery calendar rather than by discipline, and prints no target band — because nobody publishes one.
• Table Turnover Rate and Seat Occupancy ExplainedTurnover counts parties on a table; seat occupancy is a share of the hours your chairs were open for sale. This page separates seat-hours, covers, seatings and turns, prices the gap between a party and the table it was given, and shows why a queue forms while chairs sit empty.
• Restaurant Demand Forecasting: Method and AccuracyA demand forecast is worth something only once you measure its error and only once it changes two things: the hours on the roster and the quantities on the order. Here is the arithmetic of both handovers, three error measures run on the same five days, and the limits the method does not cross.
• AI Restaurant Management System: What It Actually DoesFour different products are sold under this name and only one of them is a management layer. This page separates them, says plainly which parts of such a system use a model and which are ordinary arithmetic, prints the alert threshold in full, and lists the situations in which the honest recommendation is not to buy one.
• AI Receptionist for Restaurants: What It AnswersA voice assistant closes the calls whose answer is already written down and hands over price, contract, promise and conflict. Where the AI actually sits, what has to exist before the first booking, and the two rates that measure it.
• Restaurant Phone Capacity: The Peak-Hour ArithmeticPhone-line capacity is measured in person-minutes, not in calls. Two multiplications, a subtraction and a division on your own telephony export give the lower bound of what cannot be answered at peak — and show why the real loss is always larger.
• Cost per Restaurant Booking, Channel by ChannelFour channels bring the same table and not one of them is free: the phone costs person-minutes even when nobody invoices it. Three divisions in order — cost per booking, cost per seated cover, net contribution per booking — and every one of them reorders the list.
• Guest Retention and Repeat Rate in a RestaurantRepeat rate measures only the guests you are able to recognise a second time. This page gives four formulas, the identification rate that sits under all of them, the window-length trap, and the boundary beyond which the number sees nothing at all.
• Guest Lifetime Value for a RestaurantGuest lifetime value is counted in contribution margin, not revenue, and only on guests the restaurant can identify. This page carries two formulas, the identity "active years = 1 ÷ annual churn" as a built-in check, the identified-guest bias, and the ceiling the figure sets on acquisition spend.
• Restaurant Loyalty Program: What It Really CostsA redeemed reward costs its plate cost plus the variable cost of serving it, not the menu price. This page counts earning, redemption and breakage on one closed cohort, sets the effective discount rate against member sales rather than house sales, and expresses break-even in incremental visits rather than in months.
• Restaurant Rating Math: How Many Reviews It TakesMoving an average rating is division, not persuasion. This page gives the formula for how many reviews a target needs, the condition that makes some targets unreachable at any volume, the weight of a single review at four different bases, and the ladder that shows why one bad rating costs seven fives at 4.5 and nineteen at 4.8.
• Did the Promotion Work? Holdout Groups ExplainedA jump in sales after a promotion measures the week, not the campaign: the weather, the calendar, the neighbourhood and the competition all moved with it. A holdout group is the part of the send you simply do not make — and the only way to end the quarter with one sentence that survives an argument about whether the promotion did anything at all.
• The Restaurant Facts Assistants Read About YouAn assistant repeats what somebody already published about a venue, so the measurable object is the publication, not the answer. This page gives three formulas — consistency, coverage and a plain count of contradictions — and states in the first section what it does not claim.
• Delivery Margin After Commission: The Real NumberThe advertised commission is one line of a settlement statement, not the whole deduction. A formula for the effective take rate measured from your own statement, the contribution margin of a single delivery order, and the two denominators that give two different rates for the same order.
• Delivery Menu Prices: The Uplift That Holds MarginA platform commission is a share of the price, so it grows with every uplift you make. This page builds the delivery price by division rather than addition, shows why the required uplift is not the take rate over one minus the take rate, and gives the commission at which an item stops contributing at all.
• Restaurant Customer Acquisition Cost, Honestly CountedAcquisition cost exists in two versions with different denominators, and this page keeps them apart: blended over all newly identified guests, paid over the guests a written rule attributes to paid channels. Payback is counted in visits, and the ceiling comes from lifetime value.
• Restaurant Marketing ROI, Measured on MarginROAS divides revenue by spend, so it counts money the restaurant never keeps. The break-even ROAS is one divided by the contribution margin ratio — an identity derived here in two lines, not a benchmark. The measure worth acting on is return on margin, built only from incremental sales.
• Restaurant Discounts: The Uplift They Must EarnA discount is taken from the price but paid entirely out of the contribution, so the volume it must add is always larger than the discount itself. The required uplift formula, the condition that makes it answerable, the ceiling beyond which no volume helps, and what a free dish really costs as a discount.
• The Cost of Staff Turnover in a RestaurantA departure never arrives as an invoice. Four blocks — separation, hiring, onboarding and the productivity gap — with the formulas, the Polish statutory items and a worked example in which three quarters of the money sits where nobody is looking.
• Gross or Net Sales: The Denominator That Changes EverythingThe same venue shows two different food cost numbers because two people divide by different things. This page gives the statutory rates, the formula for a check’s effective rate and the single multiplication that converts a ratio from a gross base to a net one.
• Polish Restaurant Economics in Official DataOfficial statistics publish revenue, outlet counts, employment and price indices for food service — and publish no food cost, no payroll share and no margin by outlet type. This page marks the line between the two, shows how to read a GUS or Eurostat table without adding to it, and what replaces a benchmark nobody issues.
• Booking Deposits: How Many Refusals They Are WorthA deposit trades no-shows for refusals. This page gives the break-even refusal rate formula, shows how little a deposit really recovers, and when that threshold drops to zero.
• Build your own purchase price index: what your basket actually costsNational inflation measures the guest price, not your invoice. This page builds the series nobody publishes for you: a basket chosen by spend, weights that sum to one, a frozen base period, and the arithmetic that turns a basket rise into a plate cost rise — and stops short of turning it into a menu price.
• Guest Segmentation for Restaurants: RFM in PracticeThree numbers about one guest — how recently, how often, how much — turn a guest list into groups that deserve different actions. This page sets the boundaries as quantiles of your own base, shows why a sum of RFM scores is not a quantity, and computes the lapsing threshold from the guest’s own rhythm rather than from the calendar.
• Revenue Below Forecast: Finding the Real CauseThe dashboard said “−4.4% to forecast”, and that is a size, not a cause. Here is the arithmetic that splits 860 PLN into guests and average check, assigns it to a channel through share times own change, tests the rain by comparison rather than argument, and puts your own budget edit first — because it is the only suspect you can undo.
• Overnight Messages: What Can Be Booked Without a HumanSeven messages arrived overnight and they are not one thing. This page draws the line between a guest message that can be answered and booked without a person and one where any reply would be an invention — as four gates you can run with a pen, on your own last twenty messages, before you automate anything at all.
• Winning Guests Back: Measure Contribution, Not MessagesA base of 14 000 guests earns nothing until a campaign can be taken down to contribution. Here are 183 lapsed guests split into four segments — one of which receives nothing at all — and the arithmetic that runs from 71 conversations to 27 visits, 8 460 PLN of revenue and 3 900 PLN of contribution, every formula with its reverse check.
• The Delivery Is Late: Calculate Stop-List Risk, Don't CallA supplier writes that one item will arrive two hours late. The decision is a comparison of two clock times: the hour the item runs out against the hour the van arrives. Ours came out 14:25 against 13:40 — a forty-five-minute gap, stop-list risk low, nothing to do. The page shows the whole arithmetic and the margin within which the verdict holds.
• The Dish Still Sells but Earns Less: Three Options PricedA dish can hold third place by sales and still be the position quietly losing money. Contribution per portion fell from 18.40 to 13.70 PLN over five weeks while the sales ranking did not move at all. This page works out what the 4.70 PLN gap costs per month, and what a price rise, a gramming change and a supplier change are each worth.
• Closing Runs Late: Look for the Stage, Not the CulpritClosing ran 24 minutes over the norm on 7 of the last 11 shifts with one staff combination, and evening load did not explain it. Here is how to turn that into a stage rather than a culprit, how to test the fix over five shifts without fooling yourself, and what to write down so the same problem is not solved again next year.
• Evening Forecast: From “It'll Be Busy” to a Priced DecisionAt 17:00 four things were known: 74 bookings, good weather, an event in town and historically raised walk-in after 21:00. None of them is a decision. This page shows the conversion from a forecast into a request that can be accepted or refused — extending one shift by 90 minutes for 54 PLN: where the hourly rate falls out of it, what the money has to earn back, and why the two ways of being wrong do not cost the same.
• Three Levels of Autonomy: What Runs Without Asking YouThe evening report fits into nine lines, and two of them do not mean what the reader assumes. Here is the reverse check on both, the five signs that assign a case to one of three levels of autonomy, the price of a boundary drawn too low and too high, and a one-page register you can keep on paper.
• From Complaint to Cause: One Is Never the Whole StoryA guest waiting 35 minutes for the mains is one event, and it is answered tonight. Three delays inside the same group of dishes is a cause, and it is fixed the next morning. Here is the line between the two: the reply, the compensation somebody has to confirm, and the arithmetic of repeatability.
• Ten Locations: How 100,000 Events Become Three DecisionsAcross ten venues an owner does not scale attention — he scales layers of management, and every layer is not only a salary but a delay and a re-framed signal. This page takes apart the filter that turns 100 000 events a month into three owner decisions: what a rule is made of, why the largest bucket is a remainder rather than a measurement, and why the nine rows of the effect table must never be added into one figure.
• Opening a Restaurant in Poland: Approval, HACCP and the Records a Shift KeepsApproval attaches to a room at an address and to the activity declared inside it, and HACCP is the hazard analysis behind it rather than the binder. What the act actually requires, in the order the work happens.
• Allergens on the Menu: What Polish Rules RequireFood served without packaging still carries the allergen duty. The Polish regulation does not list the allergens — it refers to the EU regulation whose annex holds them.
• The Fiscal Till and the Daily Close in PolandThe daily fiscal report closes the day as a duty with a deadline, the receipt is handed over at the moment of payment, and an annulled receipt survives as its own record. That is why the till total and the management report rarely agree by accident.
• Alcohol Licence Fees in Poland: The Actual ArithmeticA Polish venue holds a separate alcohol permit for each of three drink groups and pays a separate annual fee for each. Below the sales threshold the fee is 525, 525 or 2100 PLN; above it the fee becomes 1.4% or 2.7% of last year’s sales of that group. Every amount here is read out of the act itself, together with the one figure the act does not contain.
• What One Hour of Work Really Costs in PolandThe rate in the contract is where the calculation starts, not where it ends. This page prints the statutory contribution rates with their article numbers, shows why the accident rate is a range and not a number, and gives a formula you run on your own payroll instead of a ready multiplier that would be wrong for almost everybody.
• Overtime, Night Work and Time Records in PolandA daily norm of eight hours, eleven hours of uninterrupted rest, a night supplement counted from the minimum wage, and a working time record that is not the roster. What each of those rules does to a restaurant roster — with the Labour Code article numbers and worked through in numbers.
• Takeaway Packaging in Poland: Fee and DepositTwo separate charges can land on one takeaway order, and only one of them ever comes back. The single-use plastic fee is 0.20 PLN per drinks cup and 0.25 PLN per food container, taken from the guest and passed on. The deposit is taken at the sale of a drink in covered packaging and returns when the empty packaging returns.
• Waste Records and BDO for a Restaurant in PolandThe waste record is kept as waste moves, not assembled before an inspection, and the hour on the transfer card is what makes the difference visible. What the record consists of, where BDO sits, and the three measurements that tell you which of the two you have.
• Restaurant Reviews and the Law: What Is AllowedShowing guest reviews is a market practice with its own statute. What the act forbids, what it obliges a venue to say, and which of the two measurable numbers a restaurateur can actually produce.
• Consent to Contact and Call Recording in PolandTwo different permissions are involved when a restaurant contacts a guest. Direct marketing by phone or electronic means needs prior consent under Polish electronic communications law; recording the call is a separate act with its own basis and its own notice. A booking confirmation is neither, and mixing all three into one checkbox is the usual mistake.
• Accessibility of a Restaurant Website and OrderingElectronic commerce services are covered by the Polish accessibility act, in force since 28 June 2025, and a restaurant ordering flow is one of them. What is covered, who is exempt, which five places break first, and how to check your own ordering path with a keyboard in twenty minutes.
• AI in Guest Service: What Polish Law Now SaysPoland now has its own act on artificial intelligence systems, and it points at EU regulation 2024/1689. What the act actually contains, why a venue is the party applying a system rather than supplying one, and what to settle before the voice line goes live.
• Supplier Payment Terms and the Cash PlanPolish law caps the payment term a contract may set and attaches a fixed recovery amount to late payment, owed without a reminder. Where those two numbers sit in the act, and how a thirteen-week cash plan is built on top of them.
• Cash and Price Display: Two Duties at the CounterA venue in Poland may not make the sale conditional on cashless payment, and it must display the price before the guest orders. Both duties are quoted from the acts, with the exceptions named and with the fine that only one of them carries.
• Restaurant Energy Cost: From the Bill to One CoverEnergy is the last large line a restaurant never divides by guests. Eurostat publishes half-year prices by consumption band and excluding taxes, which gives an outside anchor; your own kilowatt-hour price comes from your own invoice. Dividing energy spend by covers served turns both into a number a manager can move.
• Food Waste in Food Service: The Numbers and the MethodEurope's official food-waste count gives food service a line of its own — but that line is an aggregate wider than restaurants, and it cannot be moved onto one venue. What the numbers say, what they do not, and how to weigh your own losses in four buckets.
• Average Check or Average Guest: Two DenominatorsTwo ratios are called the average check out loud, and they are divided by different things: bills in one case, seated guests in the other. This page prints the identity that links them, shows a fortnight where the check rose 5.0% while spend per guest fell 6.25%, and takes apart the denominator trap inside the attachment rate.
• Voids, Discounts and Comps: Reading the Till HonestlyA discount, a void, an annulment and a comp arrive in one line of the daily total and mean four different things. This page separates them, gives each its own denominator with the arithmetic worked out, and shows why an unusually even daily series means a procedure rather than guests.
• How Many People a Shift Needs: The ArithmeticRosters get copied from last week, and the copy is right about the daily total and wrong about every hour. Here is the arithmetic that derives one from covers per hour and measured capacity, the legal frame it has to pass, and why the price of understaffing is an estimate while the price of overstaffing is a measurement.
• Purchase Price Variance and the Three-Way MatchThe order, the delivery note and the invoice describe one delivery, and the total on the invoice cannot say which of two losses happened. Price variance is negotiated with the supplier; quantity variance is returned at the back door, and only until the signature.

What we see in most venues

• Empty Tuesday, a queue on Friday. Demand in hospitality is uneven, and the staff schedule, supplier orders and campaigns rarely keep up with its rhythm. Demand forecasting and external signals — weather, events in the city, competitors — show this in advance, not after the fact.
• Bookings that never show up. A no-show costs you a table someone else could have taken. Automatic follow-ups and reminders work whether or not the hostess happens to remember them.
• Google reviews nobody answers in time. A bad review from a week ago left without a reply does more harm than the situation that caused it. AURA tracks guest reviews and Google continuously, not once a month.
• A phone nobody can pick up at peak time. Most calls come in exactly when the room is full and the staff has no time to answer. AI voice handles the inbound reception and runs outgoing calls instead of a person.
• A dish cost that grows quietly. Product prices change faster than the menu, and kitchen write-offs rarely end up in one report. The finance overview shows this as it happens, not only at the end of the month.
• Turnover in the kitchen and on the floor. A hospitality team changes faster than in other industries, and the knowledge of how the venue works leaves together with the people. Labor intelligence and a schedule with recommendations stay in the system, not in one person’s head.
• Suppliers and prices you have to watch by hand. Deeper integration with the systems of the venue connects supplier data with the rest of operations instead of keeping it in a separate spreadsheet.
• Season, weather and what you cannot predict by eye. A long weekend, a heatwave, rain on a Saturday evening — each of them changes the traffic in the venue. External intelligence takes this into account instead of guessing along with you.
• Events in the city you find out about too late. A concert or a match next to the venue changes traffic more than many a campaign. External signals catch it earlier than you will see an empty or an overflowing car park.

What “external AI department” means and where the AI actually is

It is a management layer on top of what you already run: bookings, the till, the staff schedule, reviews and the phone. It does not replace your POS or your booking calendar — it collects their numbers in one place and says what follows from them. And it keeps apart two things the industry likes to blend.

• A language model works where free text and speech have to be understood: answering the phone, reading a guest message, drafting a reply to a review.
• Plain arithmetic and statistics work where numbers are counted: occupancy forecast from your own history, dish cost, write-offs, shift planning.
• Rules and thresholds work where the decision repeats: when to send a reminder, when to raise an alarm about a review left unanswered.

Calling the second and the third “artificial intelligence” sounds better in an advert and worse in a conversation with your accountant. So we say it plainly.

Work out yourself what an empty table costs you

Before anyone sells you a system, it is worth knowing your own number. The formula is simple and uses what you already have — the bills and the schedule:

• Lost revenue for an hour = free seats × average bill per person × table turns in that hour
• Lost margin = lost revenue − the ingredient cost that would have arisen − the commissions that would have arisen
• Cost of a week = the sum of lost margin across every hour when the room stood empty and the staff was on shift

Take the average bill from your own system for the last full month, not from memory. Table turns are not “how many tables” but how many times the same table managed to empty and fill again. If that number comes out low, no system will change it — the problem is in the traffic, not in managing it.

How to work out the cost of one no-show

A no-show is not “one booking less”. It is a table that could take nobody else during that time because it was held. That is why it is counted differently from an ordinary empty table:

• Cost of one no-show = average bill of a booking that size × the share of hours when you have someone to give that table to
• Work out that share from your own calendar: how often in a week you turn guests away for lack of seats
• Cost of a month = cost of one no-show × the number of bookings that never arrived

The second factor is the decisive one and the one most often skipped: on a Tuesday at five a no-show costs you close to nothing, on a Friday at eight it costs a full bill. One averaged number for the whole week overstates or understates the result so badly that no decision can rest on it.

How this differs from a booking system and from an agency

An owner usually compares three things, and compares them not by features but by what is left after a month:

• A booking system — keeps the calendar and sends reminders. It knows about bookings and knows nothing about costs, reviews or calls nobody picked up.
• A marketing agency — brings traffic and reports on campaigns. It does not see what happens to the guest after the door, and it answers neither for the schedule nor for the dish cost.
• A management layer — joins the two and shows the result in money. It replaces neither the calendar nor the campaigns; without them there is nothing to join.

Which means the management layer is a third thing, not a cheaper substitute for the first two. If someone sells it to you as a substitute — work out what will stop working.

When this does not pay off — said plainly

There are venues this layer will give nothing to, and it is better to hear that now than after the rollout. Do not start if any of the following is true:

• The room is full every day of the week and you turn guests away — then the problem is the capacity of the kitchen and the floor, not the visibility of numbers.
• You have no history in any system: bookings in a notebook, bills without line items. A forecast built on nothing is guessing with a nice chart.
• The venue is seasonal and works a few weeks a year — the fixed cost of a management layer then spreads over too short a period.
• Nobody on your side has time to read a report and act on it. A report nobody opens is a cost, not a tool.

What you need to have on your side

A rollout starts not with us but with access to your own numbers. Without these four things there is nothing to count from:

• Access to the sales system with line items on the bills, not only daily totals.
• A booking calendar in a form a machine can read — any system, as long as it is not a notebook.
• Access to the venue’s Google Business Profile, if you want to work with reviews.
• One person on your side who makes decisions: the owner or the manager. They do not need to understand data — they need to be able to say yes and no.

How to check after a month whether anything changed

Agree on the measures BEFORE you start, and pick ones you can count without us. Otherwise the conversation after a month turns into an argument about impressions:

• Number of unanswered calls at peak hours — from the phone records, not from memory.
• Share of bookings that actually arrived — from the calendar, counting the same weekday as before the change.
• Time from a review appearing to a reply — measured in hours.
• Margin on the dishes in the menu, worked out with the same formula as a month earlier.

Compare like with like: Friday with Friday, December with December. Comparing November with May will tell you about the weather, not about the system.

What AURA does not do

A management layer does not cook, does not serve the room and does not replace the owner’s decision. Specifically: it does not set your menu prices, does not hire or fire people, does not negotiate with suppliers, does not answer for the quality of the food and guarantees no financial result. It shows the numbers, proposes an action and carries out what you agreed to. The decision stays with a human, and that is not a lawyer’s caution — a system that changes menu prices on its own is more dangerous than an empty Tuesday.

Questions restaurant owners ask

What it costs

The three tiers differ in how much AURA takes on: from watching, through understanding, to running the venue.

• AURA SEESee your restaurant.
• AURA THINKUnderstand your restaurant.
• AURA ACTRun your restaurant with AURA.

AURA's scope is defined after analyzing the business. The system isn't assembled from a public price list of modules. Scope depends on processes, the number of locations, integrations, and the responsibility AURA is meant to take on.

AURA is not ordinary SaaS: we connect the POS, CRM, marketing, reviews and external data and build the venue’s Digital Twin. That is an implementation, not signing up for an account. Unlimited users inside the location.