Meituan Waimai (美团外卖), an online-to-offline (O2O) food delivery app that provides users with online ordering, now has 500,000 food delivery employees. Over the past year, the total amount of food deliveries they had done were 7 billion.
By implementing computer-aided intelligent dispatching system, the company makes sure that the average delivery time is less than 28 minutes.
The greatest cost expended by Meituan Waimai is in supporting the food delivery captains. How exactly, then, is the company improving its delivery efficiency through technical means?
HE Renqing answered this question making relevance to the barriers of data analysis, algorithm and management.
Algorithmic barriers: challenges faced both by delivery and ride-hailing service
It is worth mentioning Meituan’s car-hailing service is now wrestling with DiDi, China’s top ride-hailing company.
Technically, the problems faced by food delivery and ride-hailing services are quiet similar. One such problem is how to manage orders perfectly with high efficiency when there are a limited number of captains and drivers. For car-hailing services, it means getting its customers to their destinations as quick as possible, and for food delivery companies, it means delivering food to consumers as fast as they can.
Viewed in the business perspective of cross-specialty, traveling in its essence is related to transportation, while a delivery service is something derived from that. The difference between ride-hailing and delivery services lie only in what is being transported. In reality, however, considering all the trifles from ordering to delivering, the delivery service is far more complicated.
According to HE Renqing, by connecting drivers and customers, the car-hailing service has two tasks: picking people up and sending them to their destination. In comparison, with regards to delivery services, it needs to consider how to manage matching all the users and locations. On top of that, based on route planning, the algorithm of the management is something related to the exponential algorithms of solution space.
Solution space requires great computation. Despite that, the algorithm needs to find the best route and matching plan within a sub-second. As a result, the more skewed the data, when the algorithm starts working from the original clause, the greater the possibility of inefficiency.
Throughout the whole delivery process, the needs of users, delivery captains, restaurants and platforms are all different, and sometimes contradictory to each other. Customers expect the delivery to be punctual, restaurants expect the speed of food pick-up, riders need efficiency, and platforms require the robustness to manage large numbers of deliveries, especially in rush hours.
Consequently, when the needs of one side are satisfied, the others may be affected. Dispatching orders is a very complicated kind of multi-objective dynamic programming, and delivery is a real time multi-point matching puzzle.
From HE Renqing’s point of view, the characteristics of delivery service are multi-engagement and it being a long process. It means that, when it comes to technological solutions, platforms are facing challenges in dimensions and a more complicated dispatching system.
Normally, one delivery employee of Meituan has no more than a dozen of orders, which means that he has to scurry across over 20 locations. Consequently, there would be innumerable ways of route planning. Once the back end dispatching system fails to follow up, chances are that the riders would fail to be given orders, leaving the restaurants no choice but to notice them to come and deliver food with the cellphone number and location information written on the receipts.
Data barriers: why do riders have to collect data by themselves?
In the whole delivery process, apart from the fact that differences could arise from the magnitude in algorithm, there is another problem: the physical operation process, is in fact, being far more complicated. To accomplish the whole process, said HE Renqing, requires certain conditions as follows:
1. Arrangement of order by optimizing delivery fees and expected delivery time
On receiving an order, the platform needs to take into consideration the location of riders, the situation of orders on passage, the capability of riders, the arranging difficulty of restaurants, the weather, traffic and future orders, so that it could make the best decision of matching employees and orders. Meanwhile, the platform would prejudge the timeouts of orders and thus trigger the dynamic reassignment system.
2. Notification of expected food handout time and proper routes, and effective interaction with drivers by audio messages
After the delivery, the system would inform drivers different demand situations in different locations by predicting order needs and labor distribution in order to manage free-time dispatch. The routes are not something simply mapped by navigation tools.
Since car-hailing services are mostly related to external arteries, it has little to do with internal and indoor navigation, thus marks low complexity.
In comparison, food delivery consists of not only the outdoor transportation, but also lots of indoor tasks such as running up and down stairs to pick up food and hand it out. So it requires precise indoor map navigation and location technology.
HE Renqing pointed out that “Both Amap and Baidu Map could only inform us of the time required between the outdoor transportation from point A to B, but when it comes to indoor navigation, our Meituan employees collect data by daily delivery document and route information.
3. Management of effective dynamic matching according to order needs, geographical environment and rider characteristics
To find the best dispatching plan, the platform needs the personalized information of every user, apartment structure, restaurant, rider, location, and routes, which Meituan gains by seeking and analyzing data through the big data platform.
The food delivery system of Meituan, as revealed by HE Renqing, pays close attention to the time spent going up and down stairs by its drivers, and on top of that, it even does studies about the different speeds of movement for lower and higher floors, as well as the time of electric motorcar ride.
After obtaining the above data, constant optimizing is required. Since each mall differentiates from one another in route structure, the time of transportation changes accordingly. “All of these circumstances need long-term physical data accumulation.” said He Renqing.
Management barriers: how to avoid texting while driving?
The safety of riders is also a big concern for delivery service providers. As emphasized by HE, Meituan has put in great efforts in seeking ways to solve safety problems. “What we are worrying about is that dangers that could arise when employees try to receive orders.”
For Meituan, one of the more serious problem is the fact that many drivers tend to text while driving. Meituan’s solution: last December, the platform has been providing employees with “intelligent voice assistant”.
According to HE, this “intelligent voice assistant”, which cost Meituan 7 months to create, mainly consists of a Bluetooth headset and AI voice interaction system.
The headset could help riders to take orders, inform the platform of their arrival at restaurants’ or customers’ locations, and voice dialing by natural audio. Besides, the system could also warm them on speeding, information of weather and assignment of delivery.
What’s more, this innovative assistant can easily start working with its automatic identification, all drivers simply need to answer “Yes” or “No”.
Writer: CAO Qian
Dan Lynn of Zuzu Hospitality Solutions, Nailing two markets: Startup StoriesDan Lynn of Zuzu Hospitality Solutions, Nailing two markets: Startup Stories
Winner, winner: Early StageWinner, winner: Early Stage
Indonesian fisheries e-commerce startup Aruna wins Alipay-NUS Enterprise “tech for good” challengeIndonesian fisheries e-commerce startup Aruna wins Alipay-NUS Enterprise “tech for good” challenge
Indonesia voted. What’s the incoming government’s biggest digital economy homework?Indonesia voted. What’s the incoming government’s biggest digital economy homework?
Acquisitions are a way to grow: Carousell co-founder and CEO Quek Siu Rui with an outlook post-Naspers dealAcquisitions are a way to grow: Carousell co-founder and CEO Quek Siu Rui with an outlook post-Naspers deal