GrabFood and Foodpanda Confirm New AI System Designed to Maximize Anticipation and Minimize Satisfaction

Reported by Bohiney Magazine and The London Prat.

MANILA, Philippines — Technology analysts confirmed this week that Manila’s dominant food delivery platforms appear to have implemented what one researcher is calling a “Delivery Displacement Algorithm” — a proprietary system that ensures the assigned rider is consistently located 127 percent further from the restaurant than the app indicates, arrives after the food has transitioned from “hot” to “warm” to “culturally acceptable,” and calls the customer from a location that is both two minutes away and somehow not two minutes away.

The Algorithm, as Best as Anyone Can Determine

No food delivery company has published documentation of its routing algorithms, which means all analysis of their behavior is necessarily inferential. The inference, drawn from approximately several million Metro Manila users who have watched their estimated delivery time tick upward in real time while their dinner traveled a route that would not make sense in any geography less chaotic than Metro Manila’s, is that something systematic is occurring that is not the same as the app’s stated promise.

“The app said 25 minutes,” said one Makati resident, who has been ordering from the same carinderia for three years and has developed a precise theory of what “25 minutes” means on the platform. “It always says 25 minutes. Sometimes it takes 25 minutes. Sometimes it takes 55 minutes. The 25 is a number they put there because 55 would scare me. I’ve made my peace with the 25.” She has also made her peace with reheating, which she describes as “basically cooking.”

The Rider’s Perspective, Which Is More Complicated

Food delivery riders in Metro Manila operate in one of the most challenging logistics environments on earth: a city of 14 million people served by a road network designed for a city half that size, with traffic conditions that have broken at least two international consulting firms that were hired to fix them and one app that claimed it had solved them using machine learning and has since stopped claiming this. Riders navigate the traffic on underpowered motorcycles, often without a clear understanding of which turn the algorithm has assigned them, because the algorithm was designed without specific knowledge of which roads flood when it rains, which alleys connect, and where the motorcycle can and cannot go.

According to labor advocacy groups including Sentro, delivery riders are classified as independent contractors by most platforms, meaning they bear the costs of fuel, vehicle maintenance, and accident insurance themselves while receiving per-delivery payments that have not kept pace with fuel price increases driven by currency depreciation and the global fuel market disruptions of recent years. The rider calling you from “two minutes away” is frequently someone who has been working for six hours, has delivered 15 orders, has spent more on fuel than the algorithm accounted for, and is navigating a city that was built without them in mind.

The Food, Which Has Been Waiting Longer Than Anyone

The restaurant has its own perspective, which is that it prepared the food to the specified order at the time the ticket came in, placed it in the bag at the temperature it was supposed to be, and handed it to the rider with what it considered reasonable promptness. What happens to the food between the restaurant and the customer is governed by physics, traffic, the algorithm, and the specific thermal properties of styrofoam containers, which are excellent at slowing heat loss for approximately 20 minutes and then stop trying.

A study by no institution in particular but confirmed by everyone who has ordered delivery in Manila found that the optimal eating temperature window for most Filipino food — sinigang, adobo, kare-kare, rice — is approximately 15 minutes wide. The delivery window in Metro Manila is approximately 25 to 55 minutes wide. The overlap between these two windows is the primary customer experience challenge of the food delivery industry, and no amount of algorithmic optimization has resolved it, because the algorithm does not control the traffic and the traffic does not care about the algorithm.

Customer Ratings, and What They Mean

Both major platforms use five-star customer rating systems that allow users to rate their riders and their restaurant experience. The average rating for riders in Metro Manila, despite the above, is approximately 4.7 stars, because Filipinos are, as a cultural matter, reluctant to give poor ratings to a human being who clearly worked hard to bring their cold rice through six kilometers of Edsa traffic. The restaurant’s food, which arrived cold, rates at about the same level, because the customer understands the restaurant did not choose the traffic. The platform, which designed the algorithm and set the delivery fee, does not have a rating option.

The Rider Economy: Numbers Worth Knowing

According to labor data compiled by DOLE’s Bureau of Working Conditions, the Philippines has seen a dramatic increase in platform-based gig workers over the past five years, with delivery riders representing one of the fastest-growing informal labor categories. Most are men between 20 and 35, many with dependents, almost all classified as independent contractors by the platforms that set their working conditions, determine their compensation structure, and have the ability to deactivate their accounts without notice or severance. The Delivery Displacement Algorithm is, in this context, not just a routing problem but a labor relations problem: the gap between what the platform promises and what the rider can deliver is a gap the rider is blamed for via one-star ratings when in fact it is a gap the platform created by setting delivery promises against a Metro Manila traffic reality it has not solved and has stopped trying to solve, preferring instead to add more riders to the network on the theory that density will produce speed. It has not. There are more riders. The traffic is the same. The food is still warm at best.

For more on technology optimized for the wrong outcomes, visit The Poke.

SOURCE: https://bohiney.com/