From 4.8% to 10.9% EBITDA: how we stopped renting guests from delivery apps and got to ranking a restaurant on Google with the Restaurant Model Canvas and the Demand Radar

The trattoria billed USD 620,000 and kept 4.8% EBITDA

Fourteen months without a new photo on its Google listing, the wrong holiday hours, and the order button pointing to a third-party app: that was the starting point for a neighborhood trattoria with 48 seats, 11 payroll employees, and USD 620,000 in annual revenue. It was growing 8% year over year and EBITDA had tightened to 4.8%. The owner summed it up with the line that repeats in every audit, «we sell, but nothing is left», and he was right twice over, because 61% of transactions came through two platforms charging commission on the USD 34 delivery ticket, while the dining-room ticket stayed at 27. Nobody had calculated acquisition cost. Nobody knew guest LTV. Traffic was bought BLIND and the monthly P&L never showed it, because commissions arrived netted out. Ranking a restaurant on Google fails when the earned traffic is handed to a middleman who resells it the following month.

The problem was not visibility, it was customer ownership

That trattoria paid with product, with service, and with five-star reviews to show up first in its neighborhood, and then sent that click to an app that returned the same customer converted into a variable cost. The per-order arithmetic is brutal: USD 34 through an app left roughly USD 8.90 of contribution after the 24% commission and packaging, against USD 16.40 through the owned channel. With 1,900 monthly delivery orders, shifting twenty points of mix was worth USD 28,500 of extra annual contribution, more than any campaign the available budget could buy. Industry data points the same way: guests order 35% more items per check on first-party platforms than on third parties (Paytronix, 2024). Splitting campaigns by intent revealed that 38% of the budget went to buying brand clicks, people already typing the restaurant's name into the search bar, traffic the business had earned organically.

Ad spend was not mis-sized, it was mis-attributed

There was no growth there, only a toll. Reallocating that 38% toward category and local-intent searches held conversion in line with the sector, where restaurants and food average 7.1% conversion in Google Ads (WordStream, 2025), and cost per new guest fell because we stopped paying twice for the same customer. Diego F. Parra insists on a distinction agencies tend to lose: a brand ad does not acquire, it RECAPTURES, and recapturing what was already yours is the most expensive way to feel busy. The Google listing, meanwhile, generated 43% of inbound calls with zero investment. The Masterestaurant Acquisition Dashboard took the three pieces the trattoria treated as separate departments —the Google Business Profile listing, direct ordering, and audiovisual content— and measured them as ONE funnel with a named owner and a Monday review. The listing moved to weekly cadence: new plate photos shot under service light, every review answered inside 48 hours, holiday hours loaded a month ahead, and the order button pointing at the owned domain.

How the Masterestaurant Method was applied: one funnel, one owner, weekly cadence?

Short-form content fed discovery, a channel already accounting for 38% of restaurant discovery among Gen Z (Toast, 2026), where 51% of users say they have dined out because of what a restaurant posted (Restroworks, 2025).

Each action carried an assigned cost and was measured against the LTV of a returning guest, never against impressions. Channel mix went from 61% on apps to 38% on apps and 62% on owned channels by month nine, with total delivery volume growing 6% alongside, proof that demand was not cannibalized but margin recovered. EBITDA climbed from 4.8% to 9.1% without touching menu prices and with payroll untouched. Repeat purchase held the curve, which is where LTV is decided: the sector averages close to 55% customer retention (Restroworks, 2025), and the owned base made it possible to switch on SMS, a channel with a 45% response rate against 6% for email (Omnisend, 2025), a decision that looks less obvious once you see that email in restaurants and cafés barely reaches 1.06% clicks (Mailchimp, 2025).

The nine-month result: the mix moved and the margin showed up

I got this wrong for years recommending newsletters first; the response data wins. Pulling the apps overnight would have been a cash mistake, and that is worth saying before anyone reads this case as a crusade. Suppose the trattoria had shut both third-party channels in month one: it loses 61% of its transactions immediately, the kitchen is oversized for the remaining volume, break-even breaks, and the owner needs six months of cash he does not have to rebuild owned demand. The way out was treating the apps as a paid DISCOVERY channel, with their commission understood as an explicit acquisition cost, while the second purchase was pushed to the owned domain through a packaging insert and a measured incentive. You pay to meet the customer once, not to rent him every month. Coexistence works when there is a handover plan; without that plan, it is dependency wearing another name.

Transferable lessons

Under USD 500,000 a year: this week claim the Google Business Profile listing, fix the holiday hours, and upload six plate photos shot under service light; do not hire an agency yet. From USD 500,000 to 1 million, this case's band: calculate contribution margin per order in each channel and put a date on shifting twenty points of mix toward direct ordering. Above 1 million: split brand campaigns from category campaigns and measure acquisition cost against LTV, with a named owner for the funnel. Above 5 million: audit listing consistency location by location, because a single badly loaded holiday schedule poisons the brand average. Past 10 million, the celebrity-chef archetype running large-format venues faces the opposite problem, plenty of fame and a stunted direct-order channel, and its first step is migrating reservations and delivery to the owned domain before spending another dollar on reach.

Limits of this case

I would not expect this result in three contexts, and saying so avoids the survivorship bias that ruins any case reading. First, in a restaurant without a present owner: here there were nine years of operation and one person deciding every Monday, and without that cadence the dashboard becomes a report nobody opens. Second, in markets where owned delivery lacks a courier fleet or enough urban density, because shifting mix to a direct channel demands real logistics rather than a different button; this city had 700,000 inhabitants and dense competition, which means a market with riders available. Third, in a business with weak product or reviews below four stars, where lifting visibility only accelerates bad word of mouth. The listing amplifies whatever exists. If what exists cannot hold, fix the kitchen before the algorithm. The structural gap was never visibility, it was MARGIN PER ORDER. A 34 USD app order left roughly 8.90 USD of contribution after a 24% commission and packaging; the same order through the owned channel left 16.40.

Where the leak actually was, in numbers?

Across 1,900 monthly delivery orders, shifting 20 points of mix beat any campaign anyone could buy. Ad spend was not badly sized, it was badly attributed.

Splitting campaigns by intent revealed that brand terms —people already searching the restaurant by name— consumed 38% of the budget buying clicks the business had already earned organically. That Google listing was the most profitable asset and the worst tended. It produced 43% of inbound calls at zero cost, while the entire budget flowed to a paid channel whose restaurant conversion rate sits near 7.1% per WordStream (2025): healthy for the sector, nowhere near enough to justify abandoning organic. Instagram versus Facebook explained the wasted production: Restroworks (2025) measures 2.2% engagement on Instagram against 0.22% on Facebook, and the team kept publishing first on the lower-return network because «that is how it was always done». One figure closed the email debate: click rate for restaurants and cafés runs at 1.06%, with 3.28% click-to-open, among the lowest of any industry per Mailchimp (2025).

Where the leak actually was, in numbers — in practice

Email works as a cheap repeat-purchase reminder, never as an acquisition engine.