# RTBAgent: A LLM-based Agent System for Real-Time Bidding

**Authors:** Leng Cai, Junxuan He, Yikai Li, Junjie Liang, Yuanping Lin, Ziming Quan, Yawen Zeng, Jin Xu  
**arXiv:** [2502.00792](https://arxiv.org/abs/2502.00792)  
**Date:** 2025-02-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2502.00792v1/

## Key points

- Real-Time Bidding (RTB) revolutionized advertising industry by allowing competitive bids on impression opportunities in real-time for cost-effectiveness
- Challenges persist due to dynamic nature of online bidding environments and discrepancies between online and offline settings
- RTBAgent introduced as first RTB agent system based on large language models (LLMs), synchronizing competitive bidding environments and leveraging LLMs for informed decisions
- Enhanced with auxiliary modules like click-through rate estimation model, expert strategy knowledge, and daily reflection tailored for demands of RTB
- Features two-step decision-making process and multi-memory retrieval mechanism to analyze historical decisions and adapt quickly to market changes
- Empirical testing shows RTBAgent significantly boosts profitability for advertisers using cutting-edge technology
- Authors include Leng Cai, Junxuan He, Yikai Li, Junjie Liang, Yuanping Lin, Ziming Quan, Yawen Zeng, Jin Xu
- Code for RTBAgent publicly accessible at https://github.com/CaiLeng/RTBAgent

## Summary

, , , , 

Real-Time Bidding (RTB) has revolutionized the advertising industry by allowing advertisers to place competitive bids on impression opportunities in real-time, aiming for cost-effectiveness in a highly competitive landscape. RTB has seen advancements through technologies like deep learning and reinforcement learning, but challenges persist due to the dynamic nature of online bidding environments and discrepancies between online and offline settings. To address these challenges, a groundbreaking solution called RTBAgent has been introduced as the first RTB agent system based on large language models (LLMs). This innovative system synchronizes real competitive advertising bidding environments and leverages LLMs to make informed bidding decisions. RTBAgent is further enhanced with auxiliary modules such as a click-through rate estimation model, expert strategy knowledge, and daily reflection to tailor it specifically for the demands of RTB. One of the key features of RTBAgent is its two-step decision-making process and multi-memory retrieval mechanism. This allows the system to analyze historical decisions and transaction records, enabling it to adapt quickly to market changes in real-time bidding scenarios. Empirical testing using real advertising datasets has shown that RTBAgent significantly boosts profitability for advertisers utilizing this cutting-edge technology. The team behind this groundbreaking research includes authors Leng Cai, Junxuan He, Yikai Li, Junjie Liang, Yuanping Lin, Ziming Quan, Yawen Zeng, and Jin Xu. Their work titled "RTBAgent: A LLM-based Agent System for Real-Time Bidding" showcases how advanced technologies can be harnessed to optimize advertising strategies in today's fast-paced digital landscape. The code for RTBAgent will be made publicly accessible at https://github.com/CaiLeng/RTBAgent for further exploration and implementation by interested parties.