Artificial intelligence applied to marketing growth: the guide that shows up in the till

Step 1: set the ceiling on what you can pay per guest before touching any tool

Start with the number, not the software: a restaurant billing 180,000 dollars a year that puts 4% into marketing moves 7,200 dollars, meaning 600 a month, and that ceiling governs everything else. If your average ticket is 22 dollars and contribution margin sits near 65%, each new guest leaves you roughly 14 dollars gross, so paying more than 5 to acquire one strips your cushion for the second visit. The deliverable here is a single line written on one sheet: MAXIMUM CAC, with its figure. Verify it by dividing last month's spend by the identified guests who arrived through that spend, and if you cannot run that division because you have no identified guests, you already know the real problem, and it has nothing to do with artificial intelligence. The dish rules the campaign, and this reverses the order almost everyone works in. Pull out your menu, calculate contribution margin per dish —price minus ingredient cost, with no payroll or rent spread across it, since those live in your break-even— and sort from high to low.

Step 2: pick the dish you will move, not the message you will post

The top three are your candidates. Circana data for the quarter ending June 2025 shows value menus grew traffic 1% while total industry traffic fell 1%, which does not mean cutting prices: it means the guest hunts for a reason to choose, and you decide which one to hand over. This is done when three dishes sit on paper with their margin beside them, in dollars rather than percentages. Verification: if the dish you were going to promote out of habit is not among the three, switch it. Only now do you open the tool, and you point it at three concrete jobs: draft eight copy variants for the chosen dish, generate twenty captions tuned to the audience already buying, and classify the last ninety days of reviews to learn which word your satisfied customer repeats. No calendars. Diego F. Parra keeps insisting at Masterestaurant that AI brings nobody through the door: it multiplies what already works and speeds up discarding what does not.

Step 3: artificial intelligence enters here, and it enters as a production layer

The National Restaurant Association reported in its State of the Restaurant Industry 2025 that roughly 22% of operators planned to invest in AI tools that year, and hardly anyone measured return per dollar. The deliverable is eight finished pieces with an A and B variant, stored in a file you control. Verification: every piece names the dish from step 2. Every piece you publish has to end in a direct order of yours, and there is a cash reason to demand it: Paytronix measured in 2024 that guests order 35% more items per check on first-party ordering platforms than on third-party ones, a figure Lightspeed repeats across its 2025 online ordering statistics. On a 22-dollar ticket that gap is close to 8 dollars per transaction being given away today. Instagram will hand you reach —Restroworks measured 2.2% engagement against Facebook's 0.22%, ten times more— but reach is not ownership.

Step 4: send the traffic to a channel you own, never to a borrowed profile

The deliverable of this step is an ordering link of your own carrying a different source parameter per piece. It is verified when you open the order panel, filter by that parameter and see names, dates and tickets. No names means the data is not yours, it belongs to the platform. The list you built in the previous step is worth more than the whole campaign, because average restaurant retention hovers around 55% according to Restroworks, meaning almost half of everyone who tried your house never returns on their own. Email is not the channel to win them back: Mailchimp reports a 1.06% click and a 3.28% click-to-open for restaurants and cafés, among the lowest of any industry, while Omnisend measured a 45% SMS response rate in 2025 against email's 6%, and Tabular puts SMS clicks at 18%. With 600 dollars a month you cannot compete on reach, though you can absolutely write to two hundred people who already ate well.

Step 5: collect the second visit by SMS, where the cheap money sits

The deliverable is a sixty-character message carrying the dish from step 2 and a date. Verify it with redemptions counted at the register. First and costliest: switching on advertising with a broken funnel. When the menu hides its prices, when the photos date from 2019, or when delivery lands at 26 minutes, artificial intelligence will merely bring more people, faster, to a place that does not convert. Second: counting posts instead of guests, which measures effort rather than outcome. Third: leaving the customer list inside an agency account, because the day you walk away, you walk away without it. And the fourth, quietest of them, spending on discovery without knowing where your public discovers —Toast measured in 2026, across 1,466 U.S. adults, that TikTok drives 38% of restaurant discovery among Gen Z, a share that collapses if your guest is forty-five. Fix the funnel before paying a single dollar.

What happens if your CAC lands above the ceiling you set?

Say you close the month with 600 dollars spent and 60 identified guests: your CAC is 10 dollars against a 5-dollar ceiling, and the temptation will be raising the budget to reach scale.

Do not. With 14 dollars of contribution margin per guest you are earning 4 on that first visit, so only the second visit pays the business, and that forces you onto the retention lever before the acquisition one. Google Ads converts at 7.1% in the restaurant and food category according to WordStream's 2025 benchmarks, which tells you the platform is rarely the culprit. Double down on the owned list, measure frequency at ninety days and recalculate. Should CAC refuse to drop below 7 across two cycles, the chosen dish is wrong and you go back to step 2. Check five things and only five, with the paperwork in front of you. One: a maximum CAC exists in writing, in dollars, calculated on your real ticket and not on some internet average.

Closing checklist: how to know everything landed right

Two: the three chosen dishes are the ones with the highest contribution margin, and your latest published piece names one of them. Three: every link carries its source parameter and the order panel returns names with dates. Four: the customer list is exported into a file of yours, not sitting in someone else's account, and it grew over last month. Five: you can say, without opening a spreadsheet, how many guests the month's spend brought and what each one cost. Fail any of the five and the fix is not publishing more: it is returning to the step where the chain snapped. Start tomorrow with number one. The unit of measure. The traditional method counts published pieces; ours counts identified guests. A Bogotá restaurant went from 34 monthly posts down to 11, and attributed sales rose 22% that quarter, because those 11 came from the dishes with high contribution margin while the 34 came from a calendar.

Four differences that change the outcome

Where AI enters. At an agency it enters first, to fill the calendar. In the Masterestaurant method it enters at step three, once you already know which dish you want to move and how much you can pay per guest. That sequence is what stops artificial intelligence applied to marketing growth from becoming a factory of pretty noise. Who owns the data. If the customer list lives in the agency's account, you own no asset, you rent one. Each owned contact is worth 68 to 140 dollars of lifetime value depending on visit frequency, and that number is what carries retention and repeat purchase when a social algorithm changes its mood. How failure is handled. An agency paid by volume cannot afford to say a campaign failed, because it bills for producing. We kill campaigns at day 14 when CAC crosses 7 dollars, with no ceremony, and move the budget to the channel that actually delivered cash.