Paid advertising for restaurants: the numbers that actually decide whether you keep spending

The ceiling on what you can pay for a new guest

Not a cent more, because payroll, rent and utilities never load onto the plate: they live in the month's break-even, and mixing them collapses the math into a number that looks prudent and is simply wrong. Now bring in the other side of the scale. According to the National Restaurant Association (2026), average pre-tax net margin for an independent restaurant in mature markets sits near 5%, and every point of badly measured ad spend eats directly into that thin cushion. Badly measured advertising eats a real slice of that margin every year, and nobody notices until the fiscal close.

Why does platform ROAS lie without lying?

Platform ROAS measures correlation wearing a causation label, which is how a healthy-looking reported multiple coexists with zero additional dollars in the till.

Meta credits as a conversion any order placed within the seven-day window after a click, including the one from your Tuesday regular who was coming anyway. For example, a restaurant runs three months of reach campaigns, watches its follower count climb several times over, and bills the same as the previous quarter; the dashboard and the P&L are both telling the truth. The test that settles the argument takes fourteen days: switch the campaign off in a comparable postal code and measure the gap. Where operators have run it with discipline, a large share of reported ROAS evaporates, and what remains is your real number.

The ground you compete on: 99% of restaurants have a social profile

Almost the entire sector already sits where you want to advertise: 99% of restaurants keep at least one social media profile, according to Restroworks 2025. That doesn't make advertising pointless; it means presence stopped being an advantage years ago and now only buys the right to compete for the same attention. The operational consequence is awkward for anyone selling creative work. If most of your competitors post into the same feed, the differential comes from the asset behind the ad rather than the ad itself, and there Google rules: BrightLocal 2025 measured that businesses in the local pack top-3 hold 47 more reviews on average than positions 4 through 10. Forty-seven reviews cost less than a modest paid reach budget and they don't switch off when you stop paying.

Owned channels convert better than rented ones

Before raising the ad budget, look at what you already own: email marketing and influencer marketing you control typically return several times more than the paid ad that only reaches the guest who was coming anyway. A sixfold gap, at minimum. The consultant's reading isn't «stop advertising». It's that sequence matters: owned channel first, rented channel second, because a mailing list is still yours when the algorithm changes, and a Meta audience is not. Most restaurants already run some rewards program (Paytronix), so the contact data is sitting there, unexploited, in a large share of the operations that call me about paid advertising.

Retention: the multiplier that decides whether your CPA is expensive

A 12 USD CPA is ruinous if the guest comes once and cheap if they come six times, and that factor belongs to retention, not to the campaign. Paytronix measured in its Annual Loyalty Report 2024 a monthly member retention of 62% among the best QSR programs and 57.8% in full service. At 57.8% monthly, out of every 100 guests you acquire, 33 remain by month three and 19 by month six; lifetime value gets calculated on those, never on the 100 the dashboard shows. I apply it this way with my clients at Masterestaurant: paying 12 USD for a guest who leaves 16.32 USD of contribution per visit only works if the second visit exists. If your repeat rate isn't measured, you aren't buying customers, you're renting transactions at customer prices.

The delivery channel and the volume trap

Prepared food delivery in the United States moves a massive figure every year, and that scale is precisely why so many owners advertise inside the apps without doing the subtraction first. For example, on a given order, platform commission takes between 15% and 30% (Independent Restaurant Coalition), out of the same pocket that funds your CPA. Rerun the first passage's math with that bite included and your cost ceiling drops well below where it started. There sits the paradox worth resolving: the channel delivering the most volume leaves the least margin available to buy that volume. You resolve it with destination, not budget — advertise toward your own ordering channel and let delivery handle discovery instead of scale.

How to read these numbers in YOUR operation?

Translate the benchmarks into three scenarios before approving a single dollar.

For example, a small operation running on a modest margin keeps a limited slice of revenue as clean profit, a sensible ad budget stays modest, and the only metric worth demanding is cost per incremental order measured with a geographic switch-off. For example, a mid-size operation keeps a meaningful slice of revenue as profit, runs a proportionally larger monthly ad budget, and it's time to reserve part of spend for switch-off tests in two postal codes. Group of four locations or more: consolidate reviews before scaling ads, because the 47-review gap of the top-3 (BrightLocal 2025) gets earned once and pays across all four sites, while CPA must be paid every month in each one.

Where these benchmarks come from and what they don't tell you?

The figures in this document come from verifiable public sources:

the National Restaurant Association for operating margins, Restroworks 2025 for social penetration, BrightLocal 2025 for the local pack, Paytronix for loyalty retention, Stripo and Socially Powerful for return by channel, Statista for the size of US delivery. None is proprietary research and none has been adjusted. The limits deserve saying out loud. Almost all of that measurement is US market, and commissions, tickets and labor costs across Latin America or Spain move on another scale; retention averages come from operations with a mature loyalty program, which is not the average case; and platform ROAS doesn't appear here as a reference because it isn't comparable across accounts. Use these numbers as a decision frame and measure your own in fourteen days.

Where the arithmetic of paid advertising breaks?

Platform ROAS is not a measure of causation, it is labelled correlation. Meta credits as a conversion any order falling inside the seven-day window after a click, including the Tuesday regular who was coming regardless.

The way to size that is to switch the campaign off in a comparable postcode for fourteen days and measure the gap: where operators ran it, much of the reported ROAS disappeared. Contribution margin outranks the check. Payroll, rent and utilities do not load onto that order —they belong to the monthly break-even— and mixing them in drags the ceiling down to an unreal 4 or 5 USD that no campaign can ever meet. Guest lifetime value changes the scale of the problem. The most repeated error is optimising the cost of the first purchase while ignoring the second entirely. Delivery conversion is decided before the ad runs.

Where the arithmetic of paid advertising breaks — in practice?

For example, an aggregator order carrying a higher commission than the owned channel gives up part of its contribution, so your CPA ceiling drops;

the same ad that was profitable on the owned channel stops being profitable on the aggregator without a single pixel changing. Online reputation is a gate, not a parallel channel. The public score remains the most decisive filter for a guest who does not know the venue: according to BrightLocal (2024), 71% read Google reviews before choosing where to eat. Buying traffic into a listing with a weak star rating is paying for more people to see the problem. The decision calendar matters as much as the creative. Thirty days of waiting at a 26 USD CPA burns an extra 1.400 USD on a modest budget; a fourteen-day kill and reallocation recovers that money for the campaign that does convert, and algorithmic learning does not break if you hold the ad set and move budget between creatives.