What is Meituan?

the key business model of the O2O platform Meituan

Meituan is an O2O platform, founded in Beijing, China in 2010. Its central business lines are food delivery and on-demand services, earning tens of billions of RMB each year by connecting local restaurants and retailers with customers in China's top-tier cities.

Meituan is showing a bullish growth. In 2024, Meituan earned a profit of 35.8 billion RMBs, which increased 158.4% compared to 2023. It earned 10 billion RMBs profit in the Q1 of 2025, maintaining a strong momentum with a 87.3% year-on-year increase.

Some latest news from Meituan: Meituan plans 1,200 centralized kitchens for food safety

Project Introduction

In the Business Analytics competition for college students supported by Meituan, I obtained de-identified business data of user activities in the first half of 2023. Using this data, I identified two main problems:

• Coupon Abuse Detection: There are non-trivial coupon abuse cases. To reduce the unintended cost in offering discounts, we need to detect malicious user behavior aiming at exploiting promotional loopholes.

Coupon Abuse Detection

• User-centric Recommendation: The coupon redemption rate is quite low. To improve the utility of coupons, we need to build a coupon recommendation system to provide coupons catered to users real needs.

An efficient coupon targeting strategy guides the flow of user funnel from building awareness to boosting purchases, while a dampening utility of coupons won’t help to keep potential customers.

Data Exploration

Introduction of the datasets

- **Real world** business data of a cross-sell e-commerce platform from 2023/01/01 to 2023/06/30.
- **De-sensitized:** Contains ~50K user data with no specific user privacy involved.
- Includes transactions details, user log-ins history, and coupon receiving details.

## Keys/Grains: 

- The field name `User_id` is the common key across all 3 tables.
- The field name `Coupon_id` is the key across `order_detail.csv` and `user_coupon_receive.csv`.
- All time-related fields use a single TZ (Asia/Shanghai).
- The meanings of field names are as follows:

In `order_detail.csv`: transaction details of users; short as `txn`.
| Field Name     | Description                                                                 |
|----------------|-----------------------------------------------------------------------------|
| User_id        | Unique user ID after login                                                  |
| Shop_id        | Merchant ID                                                                 |
| Order_id       | Order ID                                                                    |
| Coupon_id      | Coupon ID used in this order                                                |
| Coupon_type    | Type of coupon used (distinguished by numbers, exact meaning not required)   |
| Biz_code       | Business line code (distinguished by numbers, exact meaning not required)    |
| Pay_date       | User payment date                                                           |
| Actual_pay     | Actual transaction amount paid by user (Order original price − Subsidy)      |
| Reduce_amount  | Subsidy amount                                                              |

In `user_visit_detail.csv`: users' timestamps of log-ins; short as `user_visit`.
| Field Name     | Description                                                                 |
|----------------|-----------------------------------------------------------------------------|
| User_id        | Unique user ID after login                                                  |
| Visit_date     | The date on which users logged in to the platform App or website            |

In `user_coupon_receive.csv`: user history of coupons receipts, and coupon discount details; short as `receipt`.
| Field Name     | Description                                                                 |
|----------------|-----------------------------------------------------------------------------|
| User_id        | Unique user ID after login                                                  |
| Coupon_id      | Coupon ID                                                                   |
| Coupon_status  | Coupon status (1 = Unused, 2 = Used, 3 = Other)                             |
| Coupon_amt     | Coupon face value (unit: RMB yuan)                                          |
| Receive_date   | Coupon receive time (format: yyyy-MM-dd)                                    |
| Start_date     | Coupon effective start time (format: yyyy-MM-dd)                            |
| End_date       | Coupon expiration time (format: yyyy-MM-dd)                                 |
| Price_limit    | Minimum spending threshold for coupon usage (unit: RMB yuan)                |

The key features of the platform

• The official platform is the major channel of attracting buyers:
  • Only 352 buyers made 1233 orders without leaving any visiting records on the website/mobile APP, representing 0.7% of total buyers and 1.3% of total orders.
  • Insight: While a small number of customers placed orders only through other channels, the vast majority of buyers were also visitors of the official sites. This demonstrates that the official channel serves as the main source for attracting buyers.

The majority of buyers were also visitors of the official channels such as the official website and the mobile APP. Only a small portion of buyers place orders solely from other channels.

• The platform is very effective in turning visitors into buyers:
  • High conversion rate: During the first half of 2023, there were 49598 different visitors logging in the website/the mobile APP, among which 46856 users made at least one purchase, representing a 94.5% conversion rate.
• The platform has strong stickiness and is habit-forming for users:
  • High frequency in visiting: each visitor logged in 57 times on average within half a year — more than twice a week per user.
  • High frequency in buying: there are roughly 21 orders per buyer on average.

Both the distributions of visiting times and order numbers are heavily right-skewed. In other words, most users visited just a few times and made only a couple of orders, while a small proportion of users contribute to the most of visiting records and orders.