Two tides,
two cities,
one calendar.

Miami-Dade is the most foreign-born large county in the United States. Roughly 54 percent of residents were born outside the country. Hispanic and Latino residents are about 70 percent of the population. Haitian Creole speakers are the second largest non-English first language community after Spanish. The implications for an ordering system are unambiguous: an English-only phone tree is not a system, it is a leak.

The American Community Survey, in Table B16001, publishes the language spoken at home for every zip code in the country. The five zip codes mapped below illustrate the spread. In Little Havana (33125), more than three quarters of households speak Spanish at home. In Brickell (33130), the share is still nearly four in ten. In Little Haiti (33150), Haitian Creole is the modal language at 62 percent. Across the western neighborhoods, Spanish is over two thirds. The voice that picks up the phone has to match.

A trilingual Voice AI changes the arithmetic in two ways. The first is order capture. A bilingual operator routinely misses 8 to 14 percent of inbound calls during the rush because the phone rings while both hands are busy. A Voice AI that picks up in the caller's language, takes the order, reads it back, and prints to the kitchen ticket printer captures those orders without adding a labor line. The second is repeat business. A regular who hears their language on the line trusts the system enough to use it again. Trust compounds. The order log fills up. The marketplace becomes the marginal channel, not the main one.

In Miami the trilingual configuration is English, Spanish, and Haitian Creole, with Portuguese as a graceful fourth where the customer base demands it. There is no reasonable substitute for that configuration in this city. There is, however, a great deal of restaurant software that pretends there is.

Set the scene: it is late September. The National Hurricane Center is tracking a Category 3 storm 72 hours west of Miami, with a forecast cone that includes Homestead, downtown, and Aventura. The 7 day inventory has already been ordered. The line cooks are watching the news between tickets. The kitchen TV is showing the same loop of palm trees bending on Brickell Avenue that it has shown four times in eight years.

Forty-eight hours out, a smart operator is running a different playbook from a normal Friday. Bread orders are halved, since perishables that arrive day-of will sit through a power outage. Stone crab claws are pulled forward into a Thursday special, because the Friday delivery is already cancelled. Two of the four line cooks are sent home Friday night to ride the storm out with their families. The remaining staff are stationed within walking distance of the restaurant.

Twenty-four hours out, the operator is on the phone with regulars: the 80 year old in apartment 14B who refused to evacuate, the law firm whose Monday lunch was already paid for, the catering client whose Thursday office order shifted to a Friday lunch in his own apartment. None of those conversations happen on a marketplace app. Direct text messages, with the operator's own customer list, are doing 100 percent of the communication work.

The third party marketplaces are the first channel to deprioritize Miami during a strike. Couriers stop accepting orders when wind gusts exceed safe biking thresholds. Delivery times balloon from 30 minutes to 90, then to "no couriers available". The operator who relies on those apps for 40 percent of revenue loses 40 percent of revenue. The operator who has spent two years building a direct customer list, with phone numbers consented for SMS, can pivot to curbside pickup, neighborhood walk-up, and pre-paid orders to be collected when the wind dies down.

Direct ordering with SMS to the customer list is the only channel that the operator controls during a hurricane week. It is the channel that survives the storm, because it does not depend on a third party's labor pool, a third party's app, or a third party's algorithm that decides Miami is not worth dispatching couriers to. After Irma in 2017, Miami operators who had built direct lists recovered in days. Operators who had not took weeks.

Forty-eight hours after the storm, the power is back on Brickell. Two cooks return, the third is still cleaning up a flooded living room in Hialeah, the fourth is unreachable. The first text message goes out to 1,200 customers at 5pm: "We are open. Limited menu. Cash only until card terminal reconnects. Curbside available." Three hundred and twelve people place orders inside two hours. The kitchen runs out of croquetas by 7pm. This is the channel that survives.

Most restaurant software is sold to a national market. It speaks one language by default. It charges by transaction, which means the operator pays more in months they need help most and less in months they barely need the software at all. It assumes one set of delivery couriers, one set of payment processors, and a building that lets the courier walk to the door. It is built for a suburban strip mall and rented to a Brickell tower.

The Miami reality, top to bottom, contradicts every one of those assumptions. The operator needs a flat monthly cost because the snowbird tide and the resident tide bracket twelve months of volume that varies by a factor of three. The operator needs trilingual Voice AI because the language on the phone in 33125 is not the language on the phone in 33150. The operator needs a delivery layer that can flex between Uber Direct, DoorDash Drive, and a local courier when the marketplace dispatch breaks down. The operator needs same-day payouts because cash flow in the resident tide is tight enough that a 72 hour bank settlement is the difference between making payroll on Friday and not.

DirectOrders is configured around exactly that set of assumptions. The price is flat. The Voice AI is trilingual, English, Spanish, and Haitian Creole, with Portuguese as a graceful fourth. The delivery layer routes through Uber Direct in Miami-Dade by default, with the option of DoorDash Drive when capacity demands. The payouts run through Stripe, same day where the bank network supports it. The customer list belongs to the operator, not to the platform, which is the entire point of building direct in a city where marketplace fees on a $1,200 catering order eat the equivalent of a senior cook every month.

It is the only configuration that handles the language-friction, the tide swing, the hurricane reality, the condo-tower friction, and the snowbird catering rhythm in one platform. It is, in plain terms, a Miami fit.