Delivery Algorithm Optimization: Before vs After with Masterestaurant

How the delivery algorithm works: predictability over price?

Predictable. Not cheapest, not fastest: that's what the Rappi, Uber Eats, and iFood algorithm ends up rewarding. The business behind it is enormous:

more than USD 18 billion moved through Latin American delivery in 2025 alone, with those three platforms soaking up 73% of regional digital order volume and growing 14% year over year. Under the hood, the ranking engine blends five signals: acceptance rate, preparation time accuracy, photo CTR, cumulative rating, and repeat order frequency. Score high on all five and visibility multiplies 2.5 to 4 times over a restaurant competing on price alone. At Masterestaurant, Diego F. Parra calls this the «operational reliability index»: the variable separating restaurants ranking 1-10 from the ones buried at position 30 or lower in their category. The single most cost-effective move a restaurant can make on delivery platforms costs nothing: turn on auto-acceptance. Fall below 90% acceptance and the penalty hits hard, 8 to 20 ranking positions lost on Rappi and Uber Eats.

Acceptance rate ≥95%: the first visibility lever

I've watched it happen across dark kitchens in Bogotá and Mexico City: the gap between accepting by hand and running auto-acceptance averages 12 to 15 positions over the first month. The mechanics are blunt. Every order ignored past 90 seconds logs as a negative signal. Turn on auto-acceptance and that counter hits zero. And the Masterestaurant target isn't a good week at 95%. It's a 30-day rolling average at ≥95%, which is the exact window the algorithm uses to compute the score. Twenty minutes declared, thirty-two delivered: that gap is the single most common mistake that wrecks a ranking. Platforms clock the real time from the moment the driver hits the door to the moment they walk out with the bag, then check it against the declared estimate. Past 5 minutes of sustained deviation, the reliability score drops and the algorithm trims visibility 15% to 30% during peak hours.

Declared vs. actual prep time: the mistake that destroys your score

The Masterestaurant threshold is ≤2 minutes average deviation, and hitting it means timing the 10 best-selling items across three shifts. You declare the 75th-percentile time, not the team's best time: the one they actually hit 3 out of 4 orders. At a dark kitchen in Medellín, with just 8 active items, that single adjustment lifted the reliability score from 6.2 to 8.7 out of 10 in 21 days. When a menu photo's CTR drops below 2%, the algorithm reads disinterest and stops featuring that item in prominent spots. Getting a listing actively promoted takes a CTR of 4% or higher, meaning at least 4 clicks for every 100 users who see the image. What's non-negotiable starts with the background: neutral, white or matte black. Overhead framing past 45°, full portion visible. And the file, compressed to ≤800 KB, has to load under 1.2 seconds on 4G.

Photo CTR ≥4%: the signal most restaurants ignore

Across 14 dark kitchens in the Masterestaurant ecosystem, between 2024 and 2025, swapping just the main photo of a low-CTR item lifted orders for that listing 22% to 41% within two weeks. Neither price nor prep time moved. Two out of three restaurants opening a new storefront on these platforms land below position 30 in their category before day 60. The reason is almost always the same: they upload the menu, flip the store live, and sit waiting for orders to fall from the sky. The platform doesn't read that as calm. It reads it as disinterest, and penalizes visibility from day one. What if that same owner, instead of waiting, generated their own 30 opening orders from staff, family, or targeted coupons? They'd seed the algorithm's reliability history before the grace window closes: 14 to 21 days on Uber Eats, 7 to 14 on Rappi.

The first 60 days: the critical window most restaurants waste

Past that margin, historical score outweighs any tactical fix that comes later. That's why the Masterestaurant protocol for the first two weeks combines three moves: a menu cut to 6-8 items, auto-acceptance from opening day, and those seeded orders that break the inertia. Delivery optimization isn't a one-time setup. It's a weekly 45-minute routine that decides, Monday after Monday, whether the restaurant climbs or slides in the ranking. Four concrete levers hold it up. First, review the metrics dashboard every Monday: acceptance rate, actual versus declared average prep time, rating, and CTR by item. Second, temporarily pull any item whose actual prep time exceeds the declared time by more than 4 minutes; six high-scoring items beat twelve dragging the average down. Third, answer 100% of negative reviews within 24 hours, because Rappi and Uber Eats have counted review response rate as a score factor since 2024.

Weekly optimization routine: the four adjustments that sustain ranking

Fourth, check pricing against the three nearest direct competitors, not to undercut them, but to confirm the price range doesn't exclude the restaurant from budget-based search filters. Eighty to one hundred twenty daily orders in a medium-density area, that's what a restaurant at position 5 pulls in. The same restaurant at position 25 gets 10 to 20. For years I told clients that any spot inside the top 20 was good enough. I was wrong. The drop isn't linear, it's exponential: past position 15, order volume halves for every 10 spots lost. At an average ticket of USD 12, that positioning gap swings monthly revenue by USD 28,800 to USD 43,200, with no change to menu or price. That's why position 15 is the real profitability ceiling: above it the model breathes, below it fixed costs (rent, payroll, utilities) stop being covered by digital order volume alone.

Position 15 as the profitability ceiling: the ranking math

The Masterestaurant method sets position ≤12 as the management target for at least 80% of days each month, verified with daily ranking screenshots taken at the same hour, 7:00 p.m. Forty active listings on a delivery platform sounds like variety; the algorithm reads it as a visibility trap. That's the paradox almost no owner solves well: more dishes should mean more customers, and in practice each low-volume item drags the average score down, because the algorithm weights every item's order history on its own. The Masterestaurant rule cuts straight through it: only items with at least 15 orders in the last 30 days stay active. In practice that means running 8 to 14 core items and rotating 2 or 3 seasonal specials a month, just enough to keep the content-update signal alive, which Rappi rewards with a 5-to-12-position visibility boost in the first 72 hours.

Algorithm-optimized menu: fewer items, higher score

A dark kitchen in Guadalajara cut from 35 active items to 11. Its average position climbed from 28 to 9 in 45 days, same total order volume, now concentrated in fewer dishes and run with tighter consistency. **Acceptance rate vs. rejection rate.** Rappi and Uber Eats track, in real time, what share of orders each restaurant accepts. Drop below 90% and the visibility penalty hits hard: 8 to 20 ranking positions lost. What's the gap between accepting by hand and running auto-accept? Twelve to fifteen positions on average over the first month, based on what I've measured across dark kitchen operations in Bogotá and Mexico City. **Declared prep time accuracy.** The courier logs the real time; the platform checks it against whatever the restaurant declared when it built the menu. Once the sustained deviation passes 5 minutes, the reliability score drops and the algorithm pushes the listing down.

The 5 differences that define your algorithm position

Restaurants that recalibrate their declared time weekly, adjusting by day and time slot, hold deviations at ≤2 minutes. That habit earns the 'reliable time' badge on iFood and Rappi, worth 1.8 points of CTR. **Cover photo with technical brief.** Before the algorithm reads the menu, it reads the cover photo's CTR. That's the first relevance signal for that user. Without a technical brief, the average sits at 1.8%. With a neutral background, three points of natural light, and the dish front and center, the range climbs to 4.5-6.2%. Those 2.7 points translate into 38% more orders, no price change, no position change. **Active ratings management.** The star average matters to the algorithm, sure. What actually moves the needle is the speed of improvement: a restaurant climbing from 3.6 to 4.2 in 30 days triggers an internal 'rising restaurant' boost inside Rappi.

The 5 differences that define your algorithm position — in practice

The Masterestaurant protocol solves this with an order-close message; it lifts the review response rate from 4% to 22% and earns the score jump without spending a dollar on discounts or paid campaigns. **Availability calendar without gaps.** Showing up as 'unavailable' during peak hours (Fridays 12-2 pm, Saturdays 7-9 pm) is a pure negative signal, and the algorithm cuts the restaurant's exposure for the next 48 hours. Mapping real capacity by time slot and shutting the door on improvised closures reverses that: 18% to 31% of the lost volume comes back within that same window.