Restaurant Marketing ROI, Measured on Margin

Two quantities called by one word

Ask three people whether the campaign paid off and you get answers to three different questions. The agency answers with revenue: the platform reported sales several times the spend. The accountant answers with profit: the month closed lower. Nobody is lying — but the link between attributed revenue and profit is not an observation. It is a calculation, and it has one honest form.

ROAS (return on ad spend) — attributed revenue divided by advertising spend. It is a ratio of two money amounts, so it carries no unit and no sign of profit in it.

Return on margin — incremental contribution margin minus advertising spend, over advertising spend. It answers whether the campaign left money behind after paying for itself.

Read the table by its denominators. Rows one and three divide by the same number and still disagree, because their numerators are different substances: one is turnover, the other is what survives the variable costs of producing it. Which figures deserve a place on an owner's screen is covered in the numbers an owner actually looks at.

ROAS: what it counts and what is not in it

The definition is short enough to hide its own problem:

ROAS = Attributed revenue ÷ Advertising spend

• Attributed revenue — sales assigned to the campaign by a written attribution rule, over one stated period, in PLN;
• Advertising spend — everything paid to run the campaign over the same period, in PLN;
• the result is a pure ratio: PLN divided by PLN leaves no unit at all.

Two things follow, both broken every week. The first: ROAS is not a percentage. A ROAS of four means turnover was four times the spend, not four per cent of it; a per-cent sign turns a healthy campaign into a catastrophe.

The second, and this is the reason for the page: there is no profit anywhere in the formula. The numerator is turnover, and turnover pays suppliers, packaging, the card fee, the delivery platform and the hourly staff called in because the room filled up. What is left is the only part that can pay for the advertising — and the formula never asks how large it is.

Contribution margin ratio — the share of net sales left after variable costs; the full definition and the method for building it belong to the restaurant break-even point, and this page uses the ratio as an input rather than redefining it.

Break-even ROAS: the identity that ends the argument

Almost every argument about advertising is an argument about a threshold nobody wrote down. It exists, it takes two lines to derive, and it is not a matter of opinion.

Step one: the condition for exactly zero

The campaign pays for itself when the money it leaves behind equals the money it consumed. What it leaves behind is not the revenue it produced but the contribution margin on that revenue — the revenue multiplied by the contribution margin ratio. So the break-even condition is one line:

Incremental revenue × Contribution margin ratio = Advertising spend

Nothing has been assumed: this is the definition of a cost being covered, with the correct numerator.

Step two: the same condition, written in multiples

Divide both sides by the advertising spend, and then by the contribution margin ratio. The left side loses the ratio and gains a division by spend; the right side loses the spend and gains a division by the ratio:

Incremental revenue ÷ Advertising spend = 1 ÷ Contribution margin ratio

The left side is the definition of ROAS, so the threshold has a closed form:

Break-even ROAS = 1 ÷ Contribution margin ratio

• Contribution margin ratio — the restaurant's own ratio for the sales the campaign produced, a decimal between 0 and 1;
• the result is a pure ratio, like ROAS itself: one divided by a unitless share stays unitless.

Break-even ROAS — the ROAS at which advertising exactly covers itself: one divided by the contribution margin ratio.

This is an identity, not an estimate: nothing entered those two lines except the definition of a cost being covered, so the result carries no assumption about industry, city, channel or season. Anyone disputing the threshold is really disputing the contribution margin ratio. And because that ratio sits in the denominator, the threshold rises fast as the ratio falls: identical campaigns with identical spend and identical revenue can earn opposite verdicts, decided by a number that lives in the kitchen and the supplier contracts, not in the ad account.

There is a degenerate case, and it belongs in the text, not in a footnote. If the contribution margin ratio is zero or negative, the identity has no usable answer: division by zero is undefined, and a negative ratio makes the threshold negative, which no ROAS can reach. That is not a failure of the formula — it says the sales being advertised do not cover their own variable costs, and the decision is about the menu, the channel or the price. Advertising something that loses money on every unit enlarges the loss in exact proportion to how well the advertising works.

Since the threshold depends only on that ratio, calculate it once and keep it visible. A dashboard showing it beside the live figure turns a monthly argument into a glance, and a calculator recomputes it the day the ratio moves — a supplier price rise, a change of delivery commission, a shift of the mix toward cheaper dishes.

Return on margin: the formula

The threshold answers yes or no. The next formula answers "by how much", and it is the one that belongs in a report:

Return on margin % = (Incremental contribution margin − Advertising spend) ÷ Advertising spend × 100

• Incremental contribution margin — the contribution margin of sales that would not have happened without the campaign, in PLN;
• Advertising spend — every cost of running the campaign over the same period, in PLN, including the work of the people who ran it;
• the subtraction leaves PLN, the division by PLN leaves a pure ratio, and the multiplication turns it into per cent.

Zero means the campaign paid for itself and left nothing behind. A positive figure is money added to the contribution available for rent and salaries; a negative one is money taken away. Unlike ROAS, this quantity has a sign, and the sign means what a reader assumes.

The same identity can also be written in money instead of multiples:

Required incremental revenue to break even = Advertising spend ÷ Contribution margin ratio

• Advertising spend — the full cost of the campaign in PLN;
• Contribution margin ratio — the same decimal share as above;
• PLN divided by a unitless share leaves PLN, which is the point: the answer arrives as a target in money.

These are not two rules. Multiply the money version by the ratio and it collapses back into step one; divide it by the spend and it becomes the threshold in multiples. A target in PLN is what a manager carries into a shift, a target in multiples is what a marketer carries into an ad account, and both are the same fact.

Why only the increment goes in the numerator

Most reported returns are wrong in the same direction: the numerator gets filled with the revenue the channel touched rather than the revenue it created.

Incremental contribution — the contribution margin of sales that would not have happened without the campaign. The definition of an increment and the method for proving one belong to the control-group page; this page takes the increment as an input.

Regulars who were coming on Friday anyway will click a paid link placed in front of them, and their spending then appears in the campaign report in full, as revenue the campaign produced. It did not produce it, it intercepted it. Money of that kind inflates the numerator without adding anything to the restaurant, and it does so exactly where the number is most persuasive: inside the channel that measures itself.

The error survives because a report full of intercepted revenue looks exactly like one full of created revenue, and the ad account cannot know who would have come anyway. Without a comparison group the honest position is to call the numerator an upper bound: at most this much, probably less.

Attribution: three rules and how each moves the answer

Before any of this arithmetic runs, someone decided which sales belong to the campaign. That decision is not a technical detail, and not the truth.

Attribution rule — the written rule that assigns a sale to a channel. Three common rules give three different answers from the same data, and none is the truth; the rule is a convention chosen in advance so that periods can be compared.

The rules disagree most on the sales a restaurant cares about most: the long ones. A guest who saw a post in spring and searched the name before booking in autumn is claimed by three channels under three rules, and every claim is defensible. Changing the rule therefore changes every return on this page without anything changing in the restaurant. Write it down, date it, and treat a change the way an accountant treats a change of policy — an event to disclose.

The calculation also rests on the point where a stranger becomes a contact. If the path from advertisement to enquiry leaks, no attribution rule can recover the sale, because there is nothing left to attribute — a different failure, covered in a neighbouring trade by why customers do not leave enquiries. A lead form recording the source at the moment of the enquiry removes the guessing, and analytics keeps one rule applied the same way month after month rather than reinvented per report.

Costs that get left out of the denominator

What counts as spend and what does not

The denominator looks like the easy part, because a platform sends an invoice with a number on it — but that invoice is rarely the whole spend. The test is not which budget line the cost came out of, but whether the cost would have existed if the campaign had not run. On that test, the following belong in the denominator and are routinely left out:

• the hours of whoever set the campaign up, watched it and turned it off — agency, employee or the owner at midnight;
• the production of what was advertised: photographs, video, the design of the offer;
• the discount or free item inside the offer, which belongs either in the spend or in the incremental contribution, and must appear in exactly one;
• the commission, platform fees and payment fees on the orders the campaign pushed into a channel that charges them;
• tools bought to run or measure the campaign, prorated over the period they served.

The fourth item turns a positive return negative most often, because it scales with success: a campaign that works pulls more orders into a commission-charging channel.

Internal hours are the item most often waved away, on the grounds that the salary was going to be paid anyway — reasoning that would remove every internal cost from every calculation ever made. The question is whether those hours could have gone somewhere else, and in a restaurant they always could; the cost brackets for automating a process treats that half directly. Where the campaign's own invoice sits in the profit and loss statement is settled by the restaurant profit margin page.

Advertising that pays back on the second visit, not the first

Some campaigns cannot break even on the visit they produced and are still correct decisions. This is not a loophole for justifying losses; it is a different arithmetic, written down in advance.

A first visit arriving at a discount, or into a delivery channel with a commission, can carry a contribution margin far below the restaurant's usual ratio, and measured on that visit alone the campaign is under water. But a guest who returns unprompted arrives with no acquisition cost and the normal margin, so the question becomes how many visits are needed before the accumulated contribution passes the spend.

That quantity has an owner: payback expressed in visits belongs to the cost of acquiring a guest. Two rules separate a real second-visit case from an excuse. First, the expected return rate has to come from measurement: if the restaurant cannot yet tell a returning guest from a new one, the second visit is an assumption and the campaign is judged on the first. Second, the promise has to be stated before the campaign runs — "it will pay back later", decided after a disappointing report, is not an argument but a way of never being wrong.

The same passage applies to a discount, a campaign whose cost is embedded in the price rather than paid to a platform. How much extra volume a discount has to add sits on the discount break-even page; the identity behind both is derived here, once.

When a negative return is a deliberate decision

Not every campaign is bought for its own return, and pretending otherwise produces a second kind of dishonesty — defending a decision with a number that was never the reason for it. There are three cases where a negative return on margin is defensible, and all three share one property: the benefit is real but lands outside the campaign's own denominator.

The first is an opening. A new location has no regulars and the first weeks are bought rather than earned; the job is to fill a room whose fixed costs run regardless, and what matters is the contribution against those fixed costs over a season, not the return on a fortnight.

The second is defending a position that would otherwise be taken. Bidding on the restaurant's own name is the clearest example: the return looks appalling, because most of those people were coming anyway. The real question is what happens if a competitor holds that position instead, and the campaign's own ROAS cannot answer it.

The third is buying information — a campaign run to find out whether a new area, format or occasion responds at all is bought as a measurement, and compares against the cost of not knowing it.

In all three the discipline is the same: name the reason before the money is spent, and calculate the return anyway. The number is not the verdict here but the price of the decision — and a price nobody calculates is a price nobody controls.

One campaign against the whole channel: two different denominators

A threshold calculated for a campaign does not transfer to a channel, and the reverse fails just as badly. The formulas are identical; the denominators are not.

A single campaign has a denominator that is easy to bound: this budget, this period, this creative. Its ratio belongs to the specific things it sold — a set lunch and a tasting menu work on different margins, and one threshold applied to both guarantees a wrong verdict for at least one.

A whole channel over a year has a denominator that includes the campaigns switched off, the periods with no spend, the subscription fees and the tools. Its mixed ratio is weighted by everything the channel sold, and drifts whenever the sales mix moves. For a group of locations the problem compounds: one threshold across the group is too low for the location with thin margins and too high for the one with thick margins, in both cases by an amount the group figure hides. Running three locations from one screen is a reporting problem in general, and this is a sharp edge of it — the group's average threshold belongs to no location in it.

Two neighbouring quantities are worth keeping separate. The cost of a booking from each channel has its own denominator and lives on the cost per booking page; the cost of acquiring a guest divides by guests rather than bookings. Neither answers whether the spend paid for itself — they answer what it bought. A report assembling all three without saying which denominator each uses is worse than one with a single figure, and an automated report helps only if the labels travel with the numbers.

What these numbers will not show: what would have happened with no advertising at all

None of these quantities compares the world with the campaign against the world without it. They compare revenue attributed to a campaign against what the campaign cost. That sounds academic until a busy month arrives: a warm week, a public holiday, a competitor closing. Each of those raises attributed revenue without the campaign having done anything. The gap has one honest closure, and it is not better tracking — it is a comparison against something that did not receive the campaign: a held-back group, a matched period, a location left out. That design belongs to the control-group page.

The second boundary concerns what is doing the work. Every formula here is division and subtraction: no model, nothing predicted, nothing that requires a machine to be intelligent. What a system genuinely does is keep the inputs from rotting — one attribution rule applied consistently, spend pulled from more than one invoice, the ratio kept current as prices move, the threshold beside the live figure instead of in a monthly file. That is bookkeeping done reliably, and this arithmetic gains nothing from being called artificial intelligence.

The third boundary concerns the industry's favourite sentence. There is no official figure for what a restaurant should spend on advertising or get back, and that is not an omission of ours but what the statistical catalogues contain. In the European statistical catalogue the entries matching "advertising" are about job advertisements, about the media sector's own turnover from selling advertising space, and about whether a firm buys internet advertising at all; the entries matching "marketing" are all about marketing innovation in past innovation surveys. In the structural business statistics indicator list the only code containing the word "advertising" is the purchase of advertising space for resale — an agency's business, not a restaurant's cost; a restaurant's advertising spend sits undivided inside total purchases of goods and services. The Polish statistical office's trade section, where gastronomy turnover is published, lists no publication about advertising or marketing spend at all. The ranges that circulate therefore have no publisher behind them: they are not norms, they cannot be checked, and this page will not print them as though they were.

Frequently asked questions

What ROAS does a restaurant need to break even?

One divided by its own contribution margin ratio, for the sales the campaign actually produced. That is an identity rather than a benchmark: set the contribution on the incremental revenue equal to the spend, then divide both sides by the spend and the ratio.

Why is return on ad spend measured on revenue misleading?

Because revenue is not money the restaurant keeps. Food, packaging, card fees, platform commission and the extra staff called in for a full room come out of it first, and ROAS never asks how much of the numerator survives those deductions.

What goes into the numerator of return on margin?

The contribution margin of sales that would not have happened without the campaign — nothing else. Not the channel's whole revenue, and not guests who were coming anyway and clicked a paid link on the way in. Where the increment cannot be established against a group that did not see the campaign, call the figure an upper bound.

How does the attribution rule change the answer?

It changes which sales enter the numerator, and therefore every quantity built on it, without anything changing in the restaurant. Last touch flatters channels that catch people already deciding, first touch the ones that create awareness, an even split understates the decisive step. The rule is a convention: write it down and date it.

Which costs belong in advertising spend besides the platform bill?

Everything that would not have existed if the campaign had not run: the hours spent building and watching it, the photography and design behind it, the tools bought to measure it, and the commission and fees on the orders it brought. Internal hours count too.

When is a negative return on a campaign still the right decision?

When the benefit is real but lands outside the campaign's own denominator: filling a new location whose fixed costs already run, defending a position a competitor would otherwise take, or buying information about a market nobody has measured. Name the reason before spending and calculate the return anyway.

Is a ROAS of four good?

Only against the threshold set by the restaurant's own contribution margin ratio. The same figure is comfortably profitable for a business that keeps a large share of each sale and a clear loss for one that keeps a small share.

Does the same threshold apply to delivery and to the dining room?

No, and applying one to both is a reliable way to reach a confident wrong answer. The ratio on a delivery order is reduced by the platform commission and the packaging first, so the threshold for delivery sits higher than for the same food served in the room. Two channels, two ratios, two thresholds.

Calculate your own threshold once, from your own contribution margin ratio, and keep it visible beside the live figure in the ad account. After that the argument about whether a number is good is settled by comparison rather than conviction, and the conversation moves on to which sales the campaign created rather than intercepted. The rest of the numbers a restaurant owner is expected to defend live on the restaurant guides hub, and the plumbing that keeps these inputs current is covered in restaurant automation.