Delivery app data and reviews: the traditional method against the Masterestaurant method

What is the app actually measuring when your rating drops?

The app doesn't measure goodwill, it measures PATTERN, and that's where the two methods split.

A rotisserie chicken spot in northern Bogotá fell from 4.9 to 4.6 stars in eleven days and lost 41% of its lunch orders without touching the recipe or the price: fourteen consecutive reviews said the same thing in different words, that the rice arrived cold, because the kitchen had started plating rice before protein to shave minutes off dispatch. The traditional method reads those fourteen reviews as fourteen annoyed customers and answers fourteen times with a coupon. The Masterestaurant method reads them as a single process deviation with fourteen witnesses. The second wins, and it wins on plain arithmetic: if the channel carries 30% of sales and the marketplace takes 30% to 40% of each ticket (ActiveMenus), a 41% volume drop isn't a reputation bruise, it's a cash hole that opens that same week.

Reading frequency: monthly versus weekly

Reading channel data every Monday beats checking it once the rating has already fallen, and the gap between the two practices is measured in days of bleeding. An owner who opens the dashboard once a month finds the deviation roughly twenty days late; by then review number fourteen is published and the algorithm has already recalculated his shelf position. A weekly cut catches the same problem with three or four cases stacked up, when fixing it costs a fifteen-minute conversation with the head chef. I got this wrong for years: I treated review monitoring as a job for whoever runs social media, when it belongs to operations, at the same rank as counting inventory. Diego F. Parra built it into Masterestaurant as a fixed block of the Monday committee, glued to costing, because the app metric and the dish food cost are talking about the same dish. Replying fast helps, though what brings the star back is fixing the process, and that deserves saying without decoration.

Customer reply versus process correction

Under the traditional method the reply closes the ticket: apology, coupon, archived. That route costs you twice, since you give away the coupon and keep the defect. Under the MR method the reply opens the file: every review enters a table with root cause and associated dish, and once the same cause shows up three times per hundred orders, a kitchen correction starts that same week. The practical difference is the threshold: three per hundred is a number you can police without expensive software, and with labor cost running between 25% and 35% of revenue according to the U.S. Bureau of Labor Statistics, every remake of a badly dispatched dish gets paid in man-hours already committed elsewhere. The app rewards speed and the customer rewards temperature: that tension gets fought out every day at the dispatch window and it has to be settled, not balanced. The traditional method picks speed because speed shows up on the marketplace panel and temperature doesn't.

Dispatch speed versus arrival temperature

The Masterestaurant method picks temperature and buys speed somewhere else, pushing mise en place forward and resequencing the plate so protein leaves first and the side follows. Go back to the Bogotá case: they gained 40 to 60 seconds per order and gave up 0.3 stars, meaning they traded one minute of dashboard for 41% of the lunch window. With DoorDash marketplace GMV growing 20% year over year through 2024 according to its annual results, the channel has demand to spare; what it has none of is patience with your rating. The store average hides the problem; the dish-level rating names it. A location sitting at 4.6 stars can hold twelve dishes above 4.8 and two below 3.9, and those two drag the average down and burn your shelf placement. The traditional method watches the big star because that's what appears on screen, and ends up reworking the whole menu when the damage sat in two SKUs.

Store rating versus dish-level rating

The MR method breaks it apart and cuts surgically: pull those two dishes from the channel catalog for fifteen days, fix packaging or plating sequence, then reinstate them. In markets where two players own the traffic — Zomato and Swiggy together hold over 95% of India's online delivery according to Business of Apps — losing shelf space over two bad dishes is a luxury the cash register can't absorb. Marking a dish sold out costs you more ranking than collecting a three-star review, and hardly any owner tracks it. When the kitchen switches off SKUs to keep the night simple, the marketplace logs an unreliable store and cuts its exposure in the listing, so you pay for tonight's convenience with tomorrow's orders. The traditional method has no such indicator anywhere; the Masterestaurant method puts it on the same board as the stars, with a hard ceiling of 2% of SKUs switched off per service.

Sold-out items: the metric nobody watches and the algorithm does

Fernando Machado, former global Chief Marketing Officer at Burger King and now a marketing executive in technology, argues that a food brand's digital reputation is built on operational consistency before advertising creativity, and in delivery that consistency is written into the ranking code. Follow the scenario to its end and you'll see why this leaves no room for lukewarm positions. Quarter one: the rating slides from 4.8 to 4.5 and you do nothing because total revenue hasn't moved yet. Quarter two: the algorithm pushes you off the top positions, volume drops 15% to 20%, and since the channel's effective cost already eats 30% to 40% of the ticket (ActiveMenus), delivery contribution turns negative. Quarter three: you buy in-app advertising to patch the hole, which means paying to win back traffic you lost over cold rice. That's the real ending, and a root-cause table costing nothing prevents it.

What happens if you ignore the pattern for a quarter?

Asia-Pacific already concentrates more than 41% of online food delivery according to Grand View Research; mature markets do not forgive repeat offenses.

If delivery brings you under 15% of sales and you run a single location, the traditional method will do, as long as you reply within 24 hours and review the panel every fifteen days. Once the channel passes 25% of billing, or once you run two or more kitchens, the Masterestaurant method stops being an upgrade and becomes the minimum viable setup: root-cause table per review, dish-level rating, 2% ceiling on sold-out items, Monday review sitting next to costing. And if you operate a dark kitchen, where there's no dining room to offset a bad score, there's no debate at all; worth remembering that Europe already accounts for 18.79% of the global dark kitchen market according to Global Growth Insights, with the competition that implies.

What to choose based on your operating profile?

Start tomorrow: export the last thirty days of reviews, sort them by cause and dish, and hunt for the one repeating three times per hundred orders.

The gap is not about answering reviews faster, useful as that is. It is about WHAT you do with what they say. In the traditional method the reply ends the process: apologise, hand out a voucher, close the ticket. Under the Masterestaurant method the reply starts it, because every review enters a table with root cause and dish attached, and once the same cause shows up three times per hundred orders a kitchen fix opens that same week. There is a real tension almost nobody resolves properly, and I will say it plainly: the app rewards speed and the customer rewards temperature, and those two fight every day in the despatch window.

Where the two methods genuinely part ways?

The traditional method picks speed because speed shows up on the dashboard;

the MR method picks temperature and compensates with plating sequence and packaging, since an order delivered two minutes later does not sink your ranking, whereas fourteen comments about cold food certainly do, and that is the arithmetic hardly anyone runs. Suppose you switched off every geotargeted Rappi campaign tomorrow. Under the traditional method position collapses within days, channel sales drop 25 % to 40 %, and you switch the budget back on inside a week, now convinced advertising was indispensable. Under the MR method the first drop looks similar, yet a 4.8 rating and repeat-order rate hold about 70 % of volume from day twelve onward, and that is when you discover you were paying for ads to hide a kitchen problem that cost far less to fix. The traditional method mistakes a virtual brand for a catalogue trick: same kitchen, new name, more orders.

Where the two methods genuinely part ways — in practice?

It works for six weeks. Then the virtual brand's reviews drag the parent brand down, because both share prep times and both share the same packaging errors, and the owner ends up with two mediocre listings instead of one good one.

A dark kitchen from scratch without data governance is just an expensive unit with no storefront. One honest concession belongs here. For years I told operators to answer every positive review with a personalised message, convinced the algorithm rewarded it. It does not reward it directly on any of the three big platforms in the region. What genuinely moves the needle is answering the negative ones fast and using that public reply to show the specific fix, because the next hesitant customer is reading exactly that reply before tapping order.