Delivery app data and algorithms: traditional method vs the Masterestaurant method

Chapter 1 — What does a delivery app algorithm actually measure?

The algorithm measures operational signals, not kitchen quality: order acceptance time, restaurant-side cancellation rate, minutes of menu availability, listing conversion and order density per zone.

No platform publishes its ranking function, and that opacity carries measurable cash consequences: with Uber Eats holding 26.1% of the US delivery market at the close of 2024 according to Earnest Analytics, the operator negotiates against a counterpart that sets rules without any duty to explain them. What you can do is feed those five signals properly, because they are free. A location that cuts average acceptance from 120 to 40 seconds bought nothing: it moved the tablet to the hot line and named a shift owner. That is the lever the dashboard will never suggest, because the dashboard sells ads.

Chapter 2 — Commission is not the lever; ranking is

Arguing commission percentages from a single location is a negotiation lost before you sit down. Earnest Analytics measured 26,1% share for Uber Eats in the United States at the end of 2024, and in Spain Ken Research 2025 places Glovo near 31% and Just Eat around 26%: these are counterparts with market power who set the price rather than discuss it. The Masterestaurant method flips what gets optimized. Instead of fighting for three commission points, we work the variables that move ranking inside the zone, which cost nothing and respond within 14 to 21 days. An honest concession belongs here: for years I recommended auditing contracts before operations, and I had the order wrong. The contract defines the ceiling of your margin; ranking defines whether there is any volume for that margin to exist on.

Chapter 3 — Orders climbing while margin falls: the dashboard trap

Growing 18% in orders and losing money on the channel is a normal outcome, not an anomaly. The app dashboard reports transactions and gross billing; it never reports contribution margin per dish, and that is where the reading breaks down. When a 30% promotion pushes the mix toward high food-cost dishes, the ticket rises and profit evaporates, because the discount applies to selling price while ingredient cost stays exactly where it was. The hard rule of the Masterestaurant framework applied by Diego F. Parra is straightforward: no dish enters a delivery promotion if its food cost exceeds 32%, and that 32% is a ceiling, never a target. Delivery volume in the region will keep arriving regardless; the question is whether it arrives with margin or only with noise.

Chapter 4 — The effect changes by annual revenue band

One algorithmic signal produces different economic effects depending on business size, which is why generic recipes fail. Below 500 thousand dollars a year, the operator lives off a single zone: two minutes of acceptance delay cost positions that never come back that month. Between 500 thousand and 1 million you meet the problem that opened this document, with three straight months of decline and no change to menu or pricing. Above 1 million the operator already splits volume across platforms and can measure real elasticity per channel. Past 5 million, the conversation shifts to category agreements and aggregate data. And beyond 10 million, with a multi-city structure, the risk inverts: cannibalization among your own brands inside one zone weighs more than any commission.

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

Above 5 million dollars a year, the celebrity-chef restaurant or the large-format themed venue runs a cost structure that invalidates the standard manual. Its listing converts on brand rather than promotion, and its real prep time collides with the platform's delivery promise: the algorithm penalizes the delay even when the dish justifies it. What works is segmenting the digital menu, keeping off the channel any preparation that exceeds 18 minutes on the line. Kitchen robotics enters at this scale too, with Grand View Research attributing 40,8% of that market to North America, and so does the temptation of a parallel virtual brand. Here I am categorical: launching a virtual brand out of a destination restaurant's kitchen dilutes the quality signal that holds the price, and incremental revenue rarely offsets the reputational damage.

Chapter 6 — What would happen if the platform reshuffled your zone tomorrow

Suppose the app changes the weight of geographic density and your location drops six positions on Monday. Week one, orders fall 22%; week two, the team cuts purchasing; week three, menu availability breaks from stockouts and the algorithm downgrades you again, this time on a signal that does depend on you. That negative feedback loop is what ruins healthy operations, and you cut it at step two, not step one. So the 90-day roadmap starts by instrumenting acceptance, cancellation and availability with daily measurement, before spending a peso on ads. Your listing competes inside that mass.

Chapter 7 — The paradox of the saturated ghost kitchen

Opening a ghost kitchen to capture more territory works and fails at the same time, and both statements hold. It works because it lowers fixed cost per brand and lets you test concepts without leasing a dining room; the cloud kitchen model is not reserved for large chains. It fails when the zone is already saturated with virtual restaurants from the same operator, because then you compete against your own supply for the same courier and the same impression. The bridge between both truths is geographic and gets calculated before signing: count how many active listings in your category sit inside the delivery radius at peak hours, and if they pass twelve, the new kitchen adds no demand, it redistributes what you already had.

Chapter 8 — The 90-day roadmap a CFO can sign off on

The first thirty days you touch no advertising: you instrument. Tablet on the hot line with a named owner per shift, daily logging of acceptance and cancellation, and an audit of menu availability by time band, which is where invisible sales bleed out. From day 31 to 60 you recalculate contribution margin per dish on the delivery channel, pull from promotion anything above 32% food cost, and reorder the digital menu by profitability rather than popularity. From 61 to 90 you test geo-targeted advertising, and only then, because by now you know whether your problem was visibility or conversion. Automation comes afterward: White Castle runs over 100 drive-thrus with voice AI according to Restaurant Dive 2024, at 90% order completion and roughly 60 seconds per order, a timing benchmark worth holding as reference. Start by moving the tablet.

Chapter 9 — The five differences that move margin

The first difference concerns the object itself: the traditional method tries to optimize commission, a price set by a third party with market power, while the Masterestaurant method optimizes the operating signals feeding the ranking, which cost nothing and sit entirely under the operator's control. With Uber Eats holding 26,1% of the US delivery market at the end of 2024 according to Earnest Analytics, arguing percentages with a platform that size from a single location is a lost negotiation before it starts. Second comes the measurement horizon. The app panel shows orders and revenue; it never shows margin. A restaurant can grow 18% in orders and still lose money on the channel if the mix drifted toward high food cost dishes pushed by a promotion. Our framework forces the month to close on contribution margin per channel rather than revenue per channel, and that single discipline changes which dishes get promoted. Third, geography. Dispatch algorithms prioritize order density per radius because their marginal delivery cost falls when two orders land close together.

Chapter 10 — The five differences that move margin — in practice

That explains why hidden kitchens cluster where demand already exists, and why opening a virtual restaurant in an empty zone almost never works, however cheap the entry looks. Fourth, how the catalog gets treated. Under the traditional approach the digital menu mirrors the dining room. Under ours it is a separate product, with fewer SKUs, combos designed to lift average ticket without adding kitchen time, and packaging costed inside the dish. The physical menu still exists and still governs the in-person experience; the QR and the app catalog serve a different purpose and do not replace it. Fifth, and most neglected: ownership of the customer. The platform keeps the contact record, the history and the frequency. Your structural defense is your own local digital engine —a complete Google Business Profile listing, Maps presence, a steady flow of 5★ reviews and a direct ordering path— because that is the only asset that survives a unilateral change in fees or ranking rules.