Delivery algorithm optimization: the decision matrix by operating profile in 2026

Which lever actually moves you up an aggregator's listing?

For an independent operator under 15 tables, with delivery between 20% and 45% of sales, the strongest lever is catalog HYGIENE and declared prep time, not paid placement inside the aggregator.

A Peruvian kitchen in an office district proved it dry: nine items that were never actually available got switched off, declared prep time went from 15 to 22 minutes, twenty-two product names were rewritten, and within eleven days daily orders climbed from 18 to 41 without a single peso spent on ads. Anyone who has looked at the mechanics understands why: DoorDash moved roughly 2,583 million orders in 2024 (DoorDash, full-year 2024 results), and at that volume the platform cannot afford to recommend listings that cancel or run late, so it buries them before it ever looks at who is paying. A delivery aggregator does not behave like a text search engine; it behaves like a recommendation system maximizing COMPLETED orders per session, and that distinction decides where you show up.

The algorithm optimizes completed orders, not pretty copy

Anything raising the odds that a user orders and receives properly — real availability, delivery times you meet, low cancellations, ratings — pushes your listing up, and anything lowering those odds buries it no matter how much budget sits on top. Best suited to small kitchens running one cook on the line: switching off a sold-out dish from the tablet beats any campaign, because a cancellation over missing product costs you the user's entire session and the signal gets recorded. With 27.5% user penetration in meal delivery in 2024 (Statista, Meal Delivery Worldwide 2024), demand is not scarce; measured reliability is. Paid placement inside the aggregator is what everybody buys first, and three scenarios argue against buying it. First, if your listing converts at 3% while the category median sits near 8%: every peso of ad spend multiplies the problem with more traffic, since you are paying for impressions rather than conversion.

When NOT to pick the popular option: in-app paid placement?

Second, if your commission already sits in the 30% tier — DoorDash restaurant plans run 15%, 25% and 30% (CloudKitchens, Delivery App Fees 2024) — ads on top drag plate margin into negative territory and you end up working for the platform.

Third, if cancellations exceed 5%: buying visibility for an operation that fails only accelerates bad ratings. Paid placement pays off only once conversion reaches the median and declared time is actually met. Promising 15 minutes and delivering in 34 costs you ranking TWICE, once for the delay and once for the rating it drags down, so treat declared time as a contract with the algorithm. Run this counterfactual in your head: say you drop declared time to 12 minutes to look faster than the shop next door; orders spike on day one, the kitchen saturates by day three, couriers wait, real deliveries stretch to 30 minutes, ratings fall below 4.5, and by week two the platform stops showing you in the main listing.

Declared time is a contract with the algorithm, not a sales pitch

You bought traffic for 72 hours and lost position for a month. Raising declared time from 15 to 22 minutes, exactly what the Peruvian operator did, is the move almost nobody dares to make. Four signals tell you, in kitchen terms, that the offer on the table will not serve you. One: the sales rep talks impressions and reach yet never shows your listing conversion rate or your cancellation rate, the two numbers that govern ranking. Two: they quote a commission plan without telling you which tier you land in — between 15% and 30% (CloudKitchens 2024) there are fifteen points of margin — while promising «more visibility» inside the same expensive tier. Three: they ask you to keep the full catalog live even with sold-out items, because a fat catalog inflates their storefront and hands you the cancellations. Four: nobody gives you real delivery time against declared time. Without that comparison, you are not optimizing anything.

Best for brands with their own demand: push the direct channel

If customers already search for your brand by name, shifting part of the volume to your own channel is the better call, and the data backs it: 58% of customers prefer ordering through the restaurant's own app or site (NCR Voyix, via Restaurant Dive 2024). The math is plain cash-register math: on a 60,000-peso ticket, escaping a 30% commission returns 18,000 per order, and at 40 orders a day that is 720,000 pesos daily going straight to the platform today. This suits you if you hold a recognizable name in the neighborhood and someone answers the phone or WhatsApp; it does NOT suit you if all your traffic comes from discovery inside the aggregator, because the direct channel stays empty and you pay for development while orders never arrive. Product names are the only field where you control the match against what people type into the aggregator's internal search, which is why rewriting twenty-two names moved that Peruvian listing fourteen places up.

Short catalog, names the internal search engine understands

My rule is strict: the name opens with the term the customer searches — «Mixed ceviche», never «Our seaside classic» — and any dish missing an ingredient gets switched off from the tablet BEFORE service, not once the order lands. Best suited to dark kitchens and operations without a dining room, where catalogs balloon because nobody pays to print menus: with the cloud kitchen market projected at USD 203,720 million by 2033 (Grand View Research), competition over that text match will only tighten. Diego F. Parra has spent twenty years between kitchens and boardrooms, and at Masterestaurant the pattern repeats across the 43 countries where we have worked: the owner who arrives saying «the algorithm has me buried» almost always carries cancellations above 5% and a catalog full of sold-out items nobody switched off. Delivery algorithm optimization starts in the kitchen and on the tablet, never in the marketing department, and that single sentence has cost me more arguments than any other with owners convinced their problem was budget.

Where optimization starts: the kitchen and the tablet?

With meal delivery penetration projected at 29.2% of users in 2026 (Statista 2026), your slot in the listing gets contested every week. Tomorrow, before service, switch off everything you do not have and add three minutes to your declared time.

A delivery aggregator algorithm is not a text search engine: it is a recommendation system optimizing completed orders per session. Anything that raises the odds a user orders AND receives well lifts your position; anything that lowers them buries you. That is why switching off a sold-out dish beats raising your budget. Advertising buys impressions, never conversion. If your listing converts at 3% while the category median sits at 8%, every dollar of ad spend multiplies your problem with more traffic. Fix conversion first —photos, names, entry price, timing— then buy volume. Declared time is a contractual promise to the algorithm, not a sales argument. Promising 15 minutes and delivering in 34 costs you ranking twice: for the delay and for the rating that follows.

The four differences that actually move position

Declare 25, deliver in 21, and you hand the system a positive signal on every single order. Delivery unit economics belong to a different business than the dining room, with a different cost structure, and treating them as the same business is the most expensive mistake I meet in Masterestaurant audits. A dish with healthy margin on a table can turn negative in the app once 27% commission, packaging and co-funded promotion walk in.