How to improve delivery and win back 5.1 EBITDA points: rescuing a 3-brand dark kitchen with the Masterestaurant Demand Radar

The opening snapshot: 86% of revenue in someone else's hands

The starting point was a 62 m² dark kitchen running three virtual brands, nine people across two shifts and a 21 USD average ticket, with 86% of sales coming through Rappi and Uber Eats and barely 14% through the operator's own WhatsApp. That ticket sat below the channel's reference floor, which Lightspeed (2025) places between 20 and 35 USD per order in the United States, and far from the 27.6 GBP gross value per order that Deliveroo plc (2024) reported after growing 5%. The owner was billing inside the 500 thousand to 1 million USD annual band and still closing months in the operational red. The kitchen ran flat out, the phone kept ringing, and EBITDA was losing 5.1 points against the line that format should hold. This was never a demand problem.

Why did channel growth hide the leak for months?

The channel grew faster than the unit margin deteriorated, which is exactly why nobody looked at the right number in time.

Grand View Research (2025) sizes the global online delivery services market at 380,430 million USD in 2024, projecting 618,360 million by 2030 at a 9.0% annual pace; that tide lifts every boat, including the ones taking on water. When gross sales climb 3% a month, a 40-cent drop per order takes a year to become visible at the bank. The owner reviewed daily sales and commission paid. He never reviewed contribution margin per dish and per channel, which is the only figure that separates a healthy delivery business from one quietly financing itself with its own cash flow.

List price, not commission, was the hole

The aggregator's commission sat within market range and was not negotiable in that city, so we attacked what the operator did control: the channel list price. A dish with 29% food cost turns a profit over the counter and loses money in delivery, because between the 25-30% commission, packaging and delivery shrinkage nearly forty points of every sale disappear. Only San Francisco set a regulatory ceiling of 15%, according to Restaurant Dive (2020), and that cap is the exception proving the rule across the rest of the continent. We rebuilt pricing on all 84 items with the Masterestaurant method's per-channel margin calculator, which splits counter price from platform price and forces each channel to pay its own cost to serve before contributing a cent toward break-even.

Pruning the catalog: from 84 items to 31

We cut the menu from 84 items to 31 and average prep time fell from 18.4 to 11.2 minutes within six weeks. A long catalog in a virtual brand is not variety, it is entropy: it stretches cooking, multiplies inventory, punishes the punctuality that the aggregator's algorithm rewards with visibility, and buries the eight dishes that actually cover payroll. Concentration is the norm in this format, not a quirk of ours: Grand View Research (2024) calculates that the meal delivery segment holds more than 64% of online delivery market revenue. I got this wrong for years, recommending gradual pruning dish by dish so the team would not panic. Gradual pruning fails, because the kitchen keeps buying for the old menu while it sells the new one.

The owned channel: from 14% to 34% of sales

We lifted the owned channel from 14% to 34% of sales in five months, and that is where most of the 5.1 recovered EBITDA points came from. The mechanics were dry: a second-order coupon printed inside the box, a short WhatsApp menu with payment by link, and an in-house delivery window limited to the 3.5-kilometer radius where cost to deliver stayed under 1.90 USD. Average ticket rose from 21 to 26 USD, close to the 24 USD per online order that Ken Research (2025) reports for Spain. The format's paradox is real and has to be resolved rather than dodged: you need the aggregator so people discover you, and you need to pull that customer off the aggregator to make money on him.

What would have happened if we had only raised prices?

Raising prices without pruning the catalog would have wrecked the operation, and the hypothesis is worth following to its end.

With 84 items and 18.4 minutes of average prep, a 12% platform increase would have pushed conversion down precisely when the algorithm was already punishing punctuality; fewer orders against the same nine-person payroll means worse fixed-cost absorption, the aggregator ranking drops, impressions shrink, and orders fall again. That loop closes in roughly eight weeks. Sequence matters more than the measures themselves: prune first, raise prices second, and ask for volume only at the end. Inverting that order is the most common way to turn a correct intervention into a 20% revenue collapse.

Transferable lessons by annual revenue band

What transfers is not the final number but the order of decisions, sized to what you bill each year. Under 500 thousand USD: this week, calculate contribution margin on your ten best sellers using platform pricing, not counter pricing, and pull anything below 22%. Between 500 thousand and 1 million, the case in this piece: prune the catalog to a third before touching prices. Above 1 million: split the P&L by channel, with its own commission, packaging and delivery charged against it. Above 5 million with several sites: audit prep-time dispersion between kitchens, because the margin hides there. And in the archetype above 10 million —a media-chef group with licensed brands— the problem stops being the dish and becomes the per-channel licensing contract, which almost never accounts for commission.

Limits of this case

I would not expect these 5.1 points in three contexts, and saying so protects the reader from the survivorship bias that fills consulting case studies. First, in a restaurant with a strong dining room where delivery weighs less than 30%: dual pricing confuses the customer who sees the printed menu and the app at the same table. Second, in markets with a regulatory commission cap, such as the 15% in San Francisco documented by Restaurant Dive (2020), because the arithmetic of dual pricing loses much of its point. Third, in industrial-scale operations like the ones Global Growth Insights describes for India, with a 4,500 million USD multibrand segment inside a 12,000 million USD dark kitchen market: there the lever is logistics and purchasing, not catalog. Start with your margin per channel on Monday.

Where delivery actually breaks (and why almost nobody looks there)?

What separates a profitable ghost kitchen from a bleeding one is rarely the aggregator commission. It is the channel list price.

A dish carrying a 29% food cost earns money at the counter and loses it on delivery unless you rebuild the price, because commission of 25-30%, packaging and delivery waste together take close to forty points of the sale, and no volume rescues that. Aggregators are not the enemy; they are an expensive channel with rules of their own. The second leak is the catalogue. Eighty-four items in a virtual restaurant is not variety, it is entropy: average prep time stretches, inventory multiplies, the punctuality the algorithm measures suffers, and the eight dishes that actually pay payroll disappear into the noise. After trimming to 39 items, average dispatch time fell 4.2 minutes without touching a single workstation.

Where delivery actually breaks (and why almost nobody looks there) — in practice?

Then the third one, which owners argue with me about most: the Google listing. Three virtual brands on one unverified profile means walking away from traffic that pays no commission.

An order arriving through WhatsApp after a map search keeps 100% of gross margin, while the same order through an aggregator keeps 72%. Across 3,400 monthly orders that gap is the difference between a business and a badly paid job. I got this wrong for years: I believed the right sequence was to win volume first and fix margin later, since volume gives you negotiating power. It runs the other way. Volume on broken unit economics only accelerates the failure, and it teaches the algorithm that you are a discount seller — a position almost nobody escapes.