Delivery algorithm optimization: the mistakes that cost you ranking, and the right method

1. What does the delivery algorithm actually measure before ranking you?

It measures operational reliability: prep time honored, kitchen-originated cancellations and menu availability, in that order of weight.

Everything else —the dish photo, the discount, how old your account is— comes later, once the marketplace has decided you will not break its delivery promise. The operator running three ghost kitchens in Bogota who lost 34% of his orders in five weeks changed neither prices nor menu: he launched a fourth virtual brand on the same hot line, and his real prep time jumped from 14 to 22 minutes. Eight extra minutes, sustained, weigh more than any campaign. And the pie split by that criterion is not small: according to DoorDash (quarterly results 2024), it processed 685 million total orders for the fourth quarter. That split gets recalculated daily, not by seniority.

2. Why paid placement buys impressions and operations buys permanence

Bought placement dies the day you switch off the budget; placement earned through fulfillment survives for weeks, because your reliability history does not get erased along with the credit card. It is an asymmetry almost nobody books when building the annual budget. Walk through the full counterfactual: if you spend 3,000 dollars a month on paid visibility with a 22-minute prep time and a 6% kitchen cancellation rate, you buy traffic into a listing the algorithm already punished, you pay commission on late orders, you collect two-star reviews and you end with less ranking than before you spent. That same money put into a second assembly line and a signposted pickup point drops prep time to 12 minutes, and that number stays. With 37% of adults ordering delivery at least once a week (UpMenu, 2024), the first screen is a balance-sheet asset.

3. The marketplace optimizes its delivery cost, not your margin

Understand it without romance: the app maximizes its own conversion and minimizes idle courier minutes, and you get paid only for lining up with those two goals. That is where the cheapest lever in the business comes from, because anything that cuts courier friction lifts your ranking without costing one extra peso in commission: packaging that does not spill on the bike, orders finished when the courier walks in, a labeled pickup station, one point of contact. That is the paradox of the trade, and it has a bridge: you do not control the algorithm, you control the seconds a courier loses inside your kitchen, and those seconds ARE the algorithm. Diego F.

4. Under 500 thousand USD a year: one brand, one shift, zero paid media

Here the decision is about focus, not investment. Below 500 thousand dollars in annual sales you run a single virtual brand per production line, and you ban the second one until median prep time holds under 15 minutes for four consecutive weeks, measured in the marketplace dashboard rather than in your own impression. The hard threshold: kitchen cancellations under 2% and menu availability above 95% at peak. Paid media budget: zero, until those three numbers are met. Trim the menu to ten dishes that plate in under eight minutes and let the ranking catch up with you.

5. From 500 thousand to 1 million and from 1 to 5 million: when the second brand opens

Between 500 thousand and 1 million dollars a year, the second virtual brand enters only with a dedicated assembly line and a shift lead who does not cook, with typical investment below 150,000 dollars for a full QSR format (Square, 2024). The threshold: prep time under 15 minutes BEFORE the second menu goes live, plus a weekly ranking review in every app. Past the million and up to five, the conversation shifts from kitchen to data: a half-time analyst reading cancellations by hour, by brand and by courier pays his salary in three months. AI-driven scheduling lowers labour cost compared with the manual method, and that saving already funds the extra shift holding the fifteen minutes. Paid media budget: 2% of digital sales at most, never to mask a weak operation.

6. Above 5 million and groups over 10 million: negotiation, not optimization

Once past the five-million mark, you stop optimizing inside the algorithm and start negotiating its terms. The high-end profile —a media-chef brand or a large-format themed concept— arrives with its own traffic and can demand tiered commission, cohort data and a guaranteed promotion window; the expensive mistake there is trading permanent discount for visibility, because the discount does erode margin while the visibility gets pulled the moment the app shifts priorities. In a group or chain above 10 million, the decision is architectural: one dashboard consolidating prep time and cancellations across every location, with an automatic alert whenever any site crosses 18 minutes. The governance threshold: no location sits below the group's 25th percentile for more than two weeks. Large-chain menu prices already climbed 42% between 2020 and 2025 (One Haus), so the pricing lever is spent.

7. Dark kitchen against brick-and-mortar: position replaces the corner

A restaurant with a dining room competes for square meters and foot traffic; a dark kitchen competes exclusively for algorithmic position inside a delivery radius, and that difference reorders the whole budget. The ghost kitchen pays for no storefront, so its rent drops, yet it loses the cushion of the customer who wanders in: if the ranking falls, revenue falls the same day, with no buffer. That is why the closing indicator in a dark kitchen is not average check but median position across the three main apps, reviewed Tuesdays at seven in the evening, with the result taped to the kitchen wall where the assembly team reads it. In Colombia the margin already runs tight —dish prices rose 9.8% since February 2025 to sustain 98,000 jobs, according to ACODRES (2025)— and a bad ranking week does not come back through promotion.

8. What to do on Monday: the three-number board

Start by measuring three figures per brand and per app, and touch nothing else until you have them: median prep time at peak, percentage of kitchen-originated cancellations and percentage of hours with a dish sold out. If prep time goes over 18 minutes, shut down your weakest brand this week instead of cutting price; the ranking gives back that lost revenue in ten or fifteen days. If cancellations pass 3%, the problem sits in the menu, not the crew: some dishes do not fit your line. And each additional star in your rating moves a real share of revenue, so the review is not vanity, it is cash. Write those three numbers on one sheet and compare them the following Monday. The algorithm keeps your history even when you do not.

9. What actually changes between buying visibility and earning it?

Ads buy impressions; operations buy permanence. Kill the budget and purchased position vanishes the same day, whereas position earned through compliance holds for weeks because your reliability history does not reset when the spend stops.

The marketplace optimizes its own conversion rate and courier cost, not your profitability. Everything that reduces courier friction —stable packaging, the order ready on arrival, a signposted pickup point— lifts your ranking without a single extra point of commission. A physical restaurant competes for square metres and foot traffic; a dark kitchen competes purely for algorithmic position inside a delivery radius.

10. What actually changes between buying visibility and earning it — in practice

In dark kitchen vs physical restaurant terms, that is the structural unit-economics gap: you are not paying for a storefront, you are paying for ranking, and ranking is paid in kitchen discipline. Rating is not soft reputation, it is measurable money. Michael Luca's work at Harvard Business School documented that one extra star moves a real share of revenue, which makes answering reviews within 24 hours a financial decision rather than a customer-service gesture. Menu prices at large U.S. chains rose 42% between 2020 and 2025 according to One Haus (2025), nearly double the 22% general inflation. Anyone who has not rebuilt their delivery menu in that window is selling with a food cost calculated for another country and another year.