How Meituan’s Joint Marketing Revolutionizes Ad Auctions with Fundraising Auctions

Instant retail is booming, and Meituan retail advertising has become a key driver for merchants and brands to expand their business. In this ecosystem, Meituan introduced a joint‑marketing model where multiple parties co‑fund a single ad slot, creating a new auction problem called “fundraising auction”.

01 Background Introduction

1.1 Joint Marketing Introduction

Joint marketing enables brands and retailers to share the same traffic resource, balancing interests of the platform, retailers, brands, and users, and improving ROI, reducing costs, and increasing platform revenue.

1.2 Fundraising Auction Introduction

In joint‑marketing, multiple parties need a scientific flow‑selling mechanism. The “fundraising auction” abstracts this as a multi‑party bidding scenario for the same item, addressing flow allocation and charging while ensuring incentive compatibility.

02 Joint Marketing Ad Mechanism Algorithm Design

The fundraising auction evolved through three stages: rule‑based → model‑based → overall.

Stage 1 : Rapid rollout for efficiency verification using GSP with reserve price.

Stage 2 : End‑to‑end model‑based solution with multi‑party incentive compatibility.

Stage 3 : Overall efficiency improvement with externality modeling and dynamic value fusion.

2.1 Rule‑based Fundraising Auction

2.1.1 GSP+Reserve Price Fundraising

Early on, GSP was compared with VCG. GSP offers clear billing but lacks full incentive compatibility for multi‑party scenarios, while VCG guarantees compatibility but yields lower revenue. To balance speed and revenue, a GSP+reserve price scheme was adopted.

2.1.2 Two‑Stage Fundraising Auction

A two‑stage mechanism integrates public‑domain and private‑domain traffic values, modeling the full user journey (recommendation, landing page, conversion) to capture downstream value and mitigate “free‑riding”. This improves overall traffic value and ad revenue.

2.2 Model‑based Fundraising Auction

To achieve incentive‑compatible multi‑party auctions, three neural‑network‑based solutions were explored.

JAMA : A generalized AMA approach that samples allocations and optimizes with a multilayer perceptron, addressing weak budget balance.

JRegNet : Regret‑based network with separate allocation and payment networks, converting DSIC constraints into regret constraints for better revenue.

JTransNet : Ensures anonymity and deterministic allocation, improving fairness and deployability.

2.2.1 JAMA

VCG was extended to a joint‑auction RVCG mechanism to satisfy weak budget balance, then combined with AMA to form JAMA, which aggregates retailer and brand bids into a joint bid for allocation and charging.

2.2.2 JRegNet

JRegNet introduces a bundle‑aware allocation network and a payment network, handling joint bids and optimizing revenue while respecting regret constraints.

2.2.3 JTransNet

JTransNet solves both anonymity and deterministic allocation, ensuring that bid order or participant identity does not affect outcomes, making it suitable for industrial deployment.

2.3 Overall Fundraising Auction

Building on JRegNet, the overall solution incorporates natural externalities by jointly modeling ad and organic candidates, introducing the JAE adaptive neural network to fuse multi‑party values and optimize platform revenue and user experience.

2.4 Technical Summary & Deployment

2.4.1 Technical Summary

The joint‑marketing ecosystem introduced a multi‑party auction paradigm, constructing a “fundraising auction” stack that progressed from rule‑based to deep‑learning‑based to holistic solutions, advancing auction theory and setting industry benchmarks.

2.4.2 Deployment

These technologies have been deployed in Meituan’s retail ad platform, improving CPM and ROI by over 20% compared to standard ads, and have been published at KDD 2024, TAMC 2024, SIGIR 2025, attracting academic and industry interest.

03 Summary and Outlook

The joint‑marketing model and its fundraising auction algorithms have significantly boosted retail ad revenue and enabled systematic productization. Future work will continue to advance both product and technology to accelerate commercialization.