From 11 to 24 Daily Orders and +3.1 EBITDA Points: How to Increase Restaurant Sales on Rappi by Fixing the Channel Leak with the Masterestaurant Demand Radar

The diagnosis: more sales in the app, less money in the till

The restaurant had no demand problem, it had a channel governance problem, and that distinction reshaped the entire intervention. We are talking about a casual dining operation with 28 seats, 9 employees, seven years of history and annual revenue between 500 thousand and 1 million USD, in a Latin American city of 900 thousand people. Average check in the dining room was 21 USD, on Rappi barely 14.80 USD, with delivery carrying 22% of total revenue. The owner arrived with a sentence I hear in many versions: he was selling more than ever in the app and earning less than ever at the till. He closed the P&L mid-way through the following month, so every wrong decision held for two months before showing up as a number. The market pushed against intuitive diagnosis too: off-premise already accounts for 29% of sales and is projected at 35% by 2026 according to the National Restaurant Association (2025).

What does the Rappi algorithm actually measure?

Rappi's algorithm rewards whoever creates no friction for the user, not whoever buys more advertising, and that is the point almost no owner accepts at first.

The platform optimizes for completed orders without incident, so three variables govern visibility: acceptance rate, real preparation time against the declared one, and store rating. At the start, acceptance sat at 84% —one in six orders rejected during peak— and real prep averaged 23.6 minutes against 15 declared, which punished the ranking twice, for rejection and for delay. Once we lifted acceptance to 98% and brought prep down to 14.2 minutes, listing visibility grew without a single extra dollar of promotion. And the context supports the bet: 37% of adults order delivery at least once a week according to UpMenu (2024). A digital catalog is a different product from the dining room menu, with fewer references, its own channel pricing and dishes that survive 25 minutes inside a closed bag.

A digital menu is not a photograph of the dining room menu

Of the 74 active products, 31 had not recorded a single sale in 90 days and were polluting the app's internal search, pushing down the dishes that did convert. We cut to 22 references grouped into four families, and listing conversion rose because users decide in seconds and a long list paralyzes them. We pulled three dishes that left the pass perfect and arrived sad: a risotto, a tempura and a salad with emulsified dressing. I got this wrong for years, recommending broader catalogs to capture more searches; the till says otherwise, and with more than 40% of adults ordering delivery or takeout three to five times a month according to UpMenu (2024), repetition beats discovery. Channel pricing is built from contribution margin after commission and packaging, never by copying the dining room price with a discount stapled on top. That was the most expensive finding of the case: with platform commission, packaging and transport shrinkage, fourteen of the twenty-two surviving dishes were selling below the channel's real cost once everything was counted.

Channel pricing: commission is designed, never absorbed

We rebuilt the matrix with the hard Masterestaurant rule —food cost per dish capped at 32%, with payroll, rent and utilities kept off the dish because they live in the break-even point— and then loaded commission and packaging as channel variable cost, not as diluted overhead. Average check on Rappi moved from 14.80 to 19.40 USD with bundles built by margin rather than popularity. The paradox of the trade resolved itself: raising prices dropped orders 6%, and the channel's gross margin still grew. The instrument was the Masterestaurant menu engineering board applied to each channel separately, and that separation is half the result. Most operations cost a dish once and assume it performs the same in the room as in the app, which is false from the first line: the same loin at 28% food cost in the room runs to a real 44% on Rappi once you add commission, packaging and the protein portion that compensates for the trip.

The tool we used: menu engineering board split by channel

We crossed popularity against contribution margin in two parallel matrices, one per channel, and out came six dining room stars that were dogs in the app, plus two minor starters that returned 61% margin in delivery. The working sequence was plain: real costing per channel for two weeks, crossed matrix, cut, reprice, and only then touch the listing and the photos. By day ninety the Rappi channel had gone from destroying margin to contributing cash, and the case figures hold without any dressing up. Revenue through the app rose 34% despite cutting the catalog from 74 to 22 references and raising prices, because the ranking improved and the listing converted better. Channel average check grew from 14.80 to 19.40 USD, acceptance settled at 98%, real prep at 14.2 minutes, and the store rating climbed from 4.2 to 4.7 stars. Channel contribution margin, which started at 9%, closed at 27% after commission and packaging.

Result measured at 90 days

Delivery moved from 22% to 31% of total revenue, a path consistent with the 35% off-premise projected for 2026 by the National Restaurant Association (2025). None of this required additional advertising spend. What applies depends on the size of the till, which is why these lessons run by annual revenue band rather than by adjective. Under 500 thousand USD: this week, count how many references you have live in the app and deactivate every one that has not sold in 90 days, with no further analysis. Between 500 thousand and 1 million, this case's band: cost your ten most ordered dishes with commission and packaging inside, and you will see how many sell at a loss. Above 1 million: separate the digital channel P&L from the dining room P&L, because aggregates hide the bleeding. Above 5 million, the celebrity-chef archetype running high-volume formats on a personal brand: audit the coherence between room price and app price, because a visible gap damages the brand more than the commission does.

Transferable lessons by revenue band

Above 10 million, multi-site groups: install real prep time measurement per location before negotiating any commission with the platform. I would not expect this result in three contexts, and saying so matters more than the 34% headline. First, in ghost kitchens with no dining room: here the physical channel absorbed fixed costs that allowed aggressive repricing in the app without losing the in-person customer, and a pure delivery operation has neither that cushion nor that brand anchor. Second, in cities where more than five platforms compete for the same courier, where delivery time depends on a fleet you do not govern and ratings fall for reasons unrelated to your kitchen. Third, in businesses under two years old with no rating history: the algorithm weighs accumulated reputation, and a new listing does not recover ranking with the same moves. This case is an anonymized composite of recurring patterns in the practice of Diego F.

Limits of this case

Parra and Masterestaurant; sector figures carry their cited source and result figures come from the intervened operation. Measure your acceptance rate this week. The Rappi algorithm does not reward whoever pays most, it rewards whoever creates the least friction for the user. Acceptance, real versus declared preparation time and rating outweigh any paid placement, because the platform optimizes for completed orders without incidents. When acceptance climbed from 84% to 98% and preparation dropped from 23.6 to 14.2 minutes, visibility rose without a single extra dollar of promotion. A digital catalog is not a photographed dining room menu. It is a different assortment, with fewer references, channel pricing and dishes that survive 25 minutes inside a bag. Of the 74 active products, 31 had not sold in 90 days and were polluting the app's internal search; cutting down to 22 lifted listing conversion, because the user decides in seconds and a long catalog paralyzes them.

The four differences that moved the number

Channel pricing is not a tax, it is a menu engineering decision. With regional commissions running between 18% and 30%, plus packaging, a dish delivering 68% gross margin in the dining room may deliver 34% in the app. That is not fixed by raising everything 25%: it is fixed by choosing what travels, what gets reformulated to travel, and what simply never enters the channel. Dining room and delivery compete for the same kitchen and the same cook. When nobody assigns capacity, the loudest channel eats the most profitable one; that is why the Demand Radar and hourly forecasting defined production windows and a separate assembly station, and order timing stopped depending on who shouted louder on the line.