Delivery algorithm optimization: before vs after with the Masterestaurant framework

Chapter 1 — Why the aggregator's ranking ignores your discounts

The aggregator's algorithm ranks by the probability that an order ends well, not by price, and that single sentence reorders your entire channel strategy. Consider what the platform is defending: its business is getting the diner to open the app again, so the weight sits on FULFILLMENT signals — seconds to accept, deviation against the promised time, merchant cancellations, later complaints — because those predict repeat purchase. A discount, by contrast, buys borrowed visibility for seven or ten days and then hands the slot back with your margin already bitten. In a channel moving 1.51 trillion dollars in 2026 and growing at a 6.24 % compound rate through 2031 according to Statista (2026), giving away contribution points to buy a position that operations gives away free is the most expensive way to compete. Start by measuring your acceptance time this week. Four indicators explain almost all the variation in your position inside the app, and none of them costs money: acceptance under two minutes, merchant cancellation below 2 %, compliance with the promised time, and a complete catalogue with photo and description on 100 % of dishes.

Chapter 2 — The four numbers that govern your visibility

One operator accepting in forty seconds and another accepting in four minutes sell the same food in the dining room and live in different worlds inside the app. The National Restaurant Association (2024) reports that 76 % of US operators believe technology gives them a competitive edge, and there sits the trap: the edge does not come from the software, it comes from the operation the software makes visible. A tablet forgotten at the end of the bar destroys more ranking than any new competitor. Put that tablet where the line cook sees it without turning. A dish that survives twenty minutes inside a sealed bag is a different product from the one that reaches the table, and confusing them produces complaints the algorithm reads as your failure. Menu engineering for the digital channel runs on two axes: contribution margin after commission — which on the large platforms takes between 15 % and 30 % of the ticket — and RESISTANCE to transport.

Chapter 3 — The digital catalogue is not the dining room menu

A plate with an emulsified sauce and fried elements loses texture in eight minutes; with 30 % commission on top, every refund for a cold arrival costs the whole dish plus the ranking penalty. My criterion, after years reordering delivery menus: first pull from the app whatever does not travel, then optimise photos. Reversed, you spend money photographing a logistics failure. Statista (2026) puts the US channel at 473.49 billion dollars this year. The same operating rule produces very different economics depending on how much the house bills, and that nuance decides where you invest first. Below 500 thousand dollars a year, cutting acceptance to ninety seconds is free and usually moves orders between 8 and 15 %: do not buy software, relocate the tablet and assign one owner per shift. Between 500 thousand and one million the first real cost appears, a dedicated expediting station during peak hours, paid for by the savings on cancellations.

Chapter 4 — What changes across annual revenue bands

Above one million it pays to integrate the aggregator into the point of sale and remove manual keying, the source of 60 to 80 % of order errors in kitchens juggling three apps. Past five million we are talking about a part-time channel analyst reading the weekly panel. The thresholds stay identical; the purchase order does not. Above five million dollars a year — the media-chef restaurant, the large-format themed venue, the brand with a queue at the door — the problem stops being visibility and becomes brand protection, something no small operator needs to solve. Here a lukewarm delivery does not cost a 22-dollar refund: it costs a review with a photo that circulates and stays. That is why the right answer in this band is usually a separate delivery menu, six to ten references designed to travel, and often a satellite dark kitchen in the highest-demand zone.

Chapter 5 — The high end: celebrity, large format and its own costs

The global dark kitchen market will reach 171.3 billion dollars by 2033 according to Global Growth Insights, and cloud kitchens 203.72 billion by 2033 according to Grand View Research. Past ten million, with several units, the decision becomes a virtual brand portfolio rather than a menu. Let us run the experiment to the end, because the answer is uncomfortable. Switch the discounts off on Monday: the first week you lose between 20 and 35 % of orders, and the urge is to turn them back on by Thursday. Hold, while keeping acceptance under two minutes and cancellation below 2 %, and by the third week volume recovers 70 to 85 % of the previous level, but with full contribution margin, because ranking rests on fulfilment. The order that never came back belonged to the deal hunter, who was never going to repeat at full price. The channel's paradox is exactly that: promotion brings volume and destroys the signal that delivers volume for free, and it resolves by using discounts to LAUNCH a new unit, with an end date written down, never as a crutch.

Chapter 6 — What would happen if you switched off every promotion for a month?

At Masterestaurant we call this buying traction, not buying customers. Separate the headline from the executable investment, because the sector's technology noise exists to sell software.

The delivery robot market projects 3,236.5 million dollars by 2030 at a 32.4 % compound rate according to MarketsandMarkets, drone package delivery 5,238.8 million at 38.7 % according to Grand View Research, and Serve Robotics announced up to 2,000 robots operating with Uber Eats in its Form 8-K filed with the SEC. None of that decides your week. What decides your week is the aggregator-to-point-of-sale integration and a dashboard carrying those four indicators. Uber Eats moved roughly 74.6 billion dollars in gross bookings during 2024, and that volume is distributed by operating rules already written today. Diego F. Parra puts it this way in Masterestaurant audits: win the algorithm you have before buying the one from 2030.

Chapter 7 — How to read the panel and what you decide with it

A one-page weekly dashboard, reviewed Tuesday morning, is worth more than any channel consultancy, and this is the exact discipline. Record five figures per week and per location: average acceptance time in seconds, merchant cancellation percentage, average deviation against the promised time, percentage of the catalogue carrying photo and description, and contribution margin after commission per dish. When cancellation crosses 2 %, do not hunt for culprits inside the app: hunt for the dish that runs out at eight and stays published. According to Statista (2024), Mexico projects 18.27 billion dollars in online delivery by 2029, and at DiDi Food Mexico 70 % of the roughly 74,000 restaurants on the app are small and midsize businesses. That is your real competition. Set your acceptance threshold at 120 seconds tomorrow and measure who breaks it. The aggregator does not rank by price, it ranks by the odds an order ends well.

Chapter 8 — The five differences that actually move the needle

Acceptance time, promised prep time met and absence of cancellations outweigh any discount, and they are free: they depend on how you organize the station, not on how much you give away. Per the National Restaurant Association (2024), 76 % of U.S. operators believe technology gives them a competitive edge, though the edge sits in the operation the technology exposes, not in the software itself. The digital catalogue is not the dining-room menu. A dish that travels twenty minutes in a sealed bag arrives different, and one that breaks generates complaints the algorithm reads as failure to deliver. Menu engineering for the channel runs on two axes — contribution margin and transport resilience — and produces a shorter menu with food cost below the 32 % ceiling the Masterestaurant framework sets as a maximum, never as a target. The always-on promo destroys margin and eventually stops working. A sustained 30 % becomes the diner's reference price, so removing it reads as a price increase; a daypart campaign, by contrast, switches on to fill the dead hour and switches off when the kitchen is slammed.

Chapter 9 — The five differences that actually move the needle — in practice

Delivery carries its own unit economics and must be costed apart. Commission of 18 % to 30 % depending on city and plan, packaging of 0.60 to 1.40 dollars per order, delivery waste and assigned labor: unless all of that comes off the channel price, you cannot tell whether you are selling or subsidizing. With Uber Eats booking 74.6 billion dollars in 2024, per Statista (2024), the channel is enormous for the platform; whether it is also enormous for your bank account depends on this arithmetic. The local engine and the aggregator reinforce each other. A Google Business Profile with the right category, current photos and answered reviews feeds near-me intent; the aggregator captures the conversion. When name, category and zone disagree across the two, you are competing against your own entity.