+5.1 EBITDA points in seven months: how we rebuilt a restaurant sales growth plan that was burning cash on ads, using the Restaurant Model Canvas and the Masterestaurant Demand Radar

The diagnosis: USD 3,180 a month to paper over an incomplete profile

This trattoria's sales growth plan started by cancelling ad spend, not buying it. Sixteen tables, 42 seats, 19 employees, an average check of USD 21.40 in the dining room and USD 17.90 on delivery, seven years of operation and an opening line I know by heart: we're selling more than ever and every month closes tighter. Both halves were true. Gross sales had climbed 14% year over year while EBITDA sank to 4.2%, and the whole explanation lived in the marketing line: USD 3,180 a month split across Rappi, Uber Eats and Meta in geotargeted ads, without a single report separating incremental sales from cannibalized ones. The owner was rebuying his own repeat customers at a 22% commission and booking them in the growth column. The number that opened the audit didn't come from the P&L, it came from a three-second phone search.

Why the moat wasn't the budget but the 47 missing reviews?

The restaurant sat in position 8 of Google's local pack for its category and neighborhood, with 87 reviews against an average of 134 among the top three.

According to BrightLocal (2025), businesses holding the top-3 of the local pack carry on average 47 more reviews than positions 4 through 10, and that 47-review gap was precisely the moat keeping him invisible while he paid for ads to leap over it. A funnel broken at the top doesn't get fixed by pushing harder at the bottom. Nobody buys enough traffic to offset a listing the local algorithm treats as second tier and the diner discards before opening it. We fixed conversion before traffic because conversion multiplies EVERYTHING else, paid and organic, present and future. At 2.7% delivery conversion, each ad dollar bought 2.7 cents of sales; taking it to 5.9% makes that same dollar deliver more than double without negotiating a single rate with the platform.

Conversion as a multiplier: from 2.7% to 5.9% with no new ad dollar

The traditional method buys traffic first for a less than noble reason: it's the only thing you can purchase with a credit card on a Tuesday afternoon. Our sequence was different. A complete Google listing with hours, attributes, menu and 34 original photos; real photography of the eight dishes carrying 61% of volume; every dish description rewritten around the ingredient that justifies the price. Three weeks of work, zero dollars of media. Reviews stopped depending on chance the day they entered the service protocol with a fixed moment. The restaurant had 87 reviews and 2.4 stars among the new ones, with the kitchen convinced the problem was the food when the complaint text talked about delivery times. We handed the review request to the server at main-course pickup, never alongside the check, which is the moment the diner is thinking about the tip and not about you.

The review as a production line, not a stroke of luck

Ninety days later: 213 total reviews and 4.6 stars average among the new ones. I was wrong for years recommending QR codes at the table, and the data corrected me. Some 75% of restaurants worldwide already use QR for digital menus (QR Code, 2025), and that very familiarity turned the QR into furniture nobody looks at. The entire budget reallocation came out of the Masterestaurant channel profitability board, which Diego F. Parra uses to separate gross sales from net contribution dish by dish. We loaded the eight dishes making up 61% of volume with their real food cost, each platform's commission and delivery packaging cost, and out came what the sales report was hiding: the three most ordered delivery dishes returned between 6 and 11 points less contribution than in the dining room, once the 22% commission and the USD 0.85 of packaging per order came off.

The Masterestaurant tool that put numbers on the wrong channel

With that table on the table, the owner pulled two dishes from the digital menu, raised three delivery prices by USD 1.60 and cut ad spend from USD 3,180 to USD 1,140 a month, all aimed at one objective: capturing the customer's email. The channel that sustained growth after the repair quarter was the owned database, not the rented platform. According to Litmus (2024), email returns USD 36 for every dollar invested, and the DMA (2024) measures that return at USD 42.24 per dollar; no paid delivery channel comes anywhere near that ratio. We built email capture at the counter and on the packaging, 1,480 records in five months, with a single birthday automation because birthday coupons get redeemed three times more than a standard email offer (Stripo, 2025). Loyalty isn't faith either: average ROI of restaurant loyalty programs runs 4.8x and 90% of operators report positive returns (Welcome Back, 2026).

Email and loyalty: where the dollar returns 36 times, not 2.7 cents

Close of period: delivery fell from 61% to 44% of sales and EBITDA rose from 4.2% to 11.8%. What transfers from this case isn't the tactic, it's the order, and it shifts with what you bill each year. Under USD 500K: complete your Google listing this week with hours, attributes and 20 original photos, hiring nobody. Between USD 500K and 1M, this trattoria's band: measure delivery conversion and put the review request inside the service protocol before touching a cent of ad spend. Above 1M: split net contribution by channel dish by dish, because cannibalization there already costs more than the salary of whoever measures it. Above 5M, the archetype of the media chef with a flagship room and two openings a year: audit whether the personal brand is subsidizing an operation with prime cost outside its band, which is what the line at the door usually hides.

Transferable lessons by annual revenue band

Above 10M, multi-unit group: unify each location's listing with a named owner per site and a weekly board, because 99% of restaurants already hold a social profile (Restroworks, 2025) and what's scarce is upkeep. This result doesn't repeat in three contexts, and it's worth saying so before somebody copies the sequence without looking at where they stand. First, in a saturated market where the local pack's top three clear 400 reviews, closing the 47-review moat BrightLocal (2025) measures takes more than nine months and demands service volume that 42 seats won't produce. Second, if the real problem is the product and not the listing, raising conversion only speeds diners toward an experience that loses them: at 2.4 stars driven by the kitchen rather than delivery times, you fix the kitchen first. Third, in a restaurant without the owner on the floor every day the review protocol collapses in week three, and with no protocol there is no review production line.

Limits of this case

This trattoria had its owner in the dining room, and that isn't a detail of color, it's the case's precondition. Order of operations. Traditional plans buy traffic first because traffic is the only thing a credit card can buy; we fix conversion first because conversion multiplies everything else. At 2.7% delivery conversion, every ad dollar purchased 2.7 cents of sales; at 5.9%, that same dollar buys more than double. Conversion is a multiplier over ALL traffic — paid and organic, today's and next year's. Reviews as a production line rather than luck. The place had 87 reviews and 2.4 stars on recent ones, with a kitchen that never learned the complaint was delivery time and not the food. We moved the review request inside the service protocol — the server asks while clearing the main course, never at the check — and online reputation stopped depending on a guest's mood.

The five differences that actually moved the needle

Smaller menu, bigger sales. QR Code (2025) reports that 75% of restaurants worldwide already run QR digital menus, and that is precisely where the sector goes wrong: it assumes digitizing the menu means removing the menu. We cut delivery from 71 to 34 items, because a digital channel demands a decision in 40 seconds, and in the dining room we KEPT the physical menu as a suggestive-selling instrument, with the QR supporting price changes and accessibility. Both, each with its own job. Guest LTV entered the report. A new customer bought with a 30% discount and a regular who shows up three times a month used to weigh the same. Once we tracked 90-day frequency and check per visit, we found that 18% of guests produced 47% of dining-room sales. Ad targeting was rebuilt to resemble that 18% instead of chasing raw volume. Ads last, and with a control group.

The five differences that actually moved the needle — in practice

We cut the budget 38% and sales rose that same month, because what the ads had been buying were people already walking in. This is the most uncomfortable correction in the whole process, and the one owners resist hardest: nobody wants to admit their growth engine was, in large part, a voluntary tax paid to the platforms.