JD-BP: A Joint-Decision Generative Framework for Auto-Bidding and Pricing

Linghui Meng; Chun Gan; Shengsheng Niu; Chengcheng Zhang; Chenchen Li; Chuan Yang; Yi Mao; Xin Zhu

cs.GTApr 7, 2026

JD-BP: A Joint-Decision Generative Framework for Auto-Bidding and Pricing

Authors: Linghui Meng, Chun Gan, Shengsheng Niu, Chengcheng Zhang, Chenchen Li, Chuan Yang, Yi Mao, Xin Zhu,

Organizations: JD.com China · The Chinese University of Hong Kong China

Abstract

Auto-bidding services optimize real-time bidding strategies for advertisers under key performance indicator (KPI) constraints such as target return on investment and budget. However, uncertainties such as model prediction errors and feedback latency can cause bidding strategies to deviate from ex-post optimality, leading to inefficient allocation. To address this issue, we propose JD-BP, a Joint generative Decision framework for Bidding and Pricing. Unlike prior methods, JD-BP jointly outputs a bid value and a pricing correction term that acts additively with the payment rule such as GSP. To mitigate adverse effects of historical constraint violations, we design a memory-less Return-to-Go that encourages future value maximizing of bidding actions while the cumulated bias is handled by the pricing correction. Moreover, a trajectory augmentation algorithm is proposed to generate joint bidding-pricing trajectories from a (possibly arbitrary) base bidding policy, enabling efficient plug-and-play deployment of our algorithm from existing RL/generative bidding models. Finally, we employ an Energy-Based Direct Preference Optimization method in conjunction with a cross-attention module to enhance the joint learning performance of bidding and pricing correction. Offline experiments on the AuctionNet dataset demonstrate that JD-BP achieves state-of-the-art performance. Online A/B tests at JD.com confirm its practical effectiveness, showing a 4.70% increase in ad revenue and a 6.48% improvement in target cost.

Figures & tables

Figure 1. Overview of the JD-BP framework. (a) Dual-Stream Decision Transformer: parallel Bidding (blue) and Pricing (orange) streams over a return/state/action token sequence. (b) Auction Environment emitting per-step value $v_{t}$ and cost $c_{t}$ . (c) Memoryless RTG conditioned only on the future $[t_{m},T]$ . (d) GCA: a gate $\sigma$ fuses bidding context into the pricing query. (e) Energy-Based DPO placing $a^{+}$ / $a^{-}$ at low/high energy w.r.t. $a_{\mathrm{ref}}$ . (f) Offline training pipeline ending at the deployed JD-BP agent.

\tau_{\text{joint}}=\{(R_{t}^{b},R_{t}^{p},\hat{s}_{t},\hat{a}_{t}^{b},a_{t}^{p},A_{t})\}

Algorithm 1 Trajectory Generation Algorithm

Dataset	Budget	DiffBid	USCB	CQL	IQL	BCQ	DT	CDT	GAS	GAVE	GRAD	JD-BP
AuctionNet	50%	9.9	11.5	12.8	16.5	17.7	14.8	11.2	18.4	19.6	20.0	21.2
	75%	15.4	14.9	16.7	22.1	24.6	22.9	18.0	27.5	28.3	28.5	32.9
	100%	19.5	17.5	22.2	30.0	31.1	29.6	31.2	36.1	37.2	37.4	42.5
	125%	25.3	26.7	28.6	37.1	34.2	34.3	31.7	40.0	42.7	43.2	44.9
	150%	30.8	31.3	35.8	43.1	37.9	44.5	39.1	46.5	47.4	47.5	47.5
	Avg	20.2	20.4	23.2	29.8	29.1	29.2	26.2	33.7	35.0	35.3	37.8

Table 1. Score comparison on AuctionNet (period-7, 48 advertisers) under five budget levels (50%–150% of the original budget). The highest score in each row is in bold; the best baseline is underlined. Higher is better.

Variant	50%	75%	100%	125%	150%	Avg
JD-BP (Full)	21.2	32.9	42.5	44.9	47.5	37.8
w/o DPO	18.6	27.6	38.9	43.7	47.4	35.2
w/o GCA	17.8	31.3	37.7	40.6	44.2	34.3
w/o m-RTG	20.7	29.4	37.4	38.3	43.5	33.8

Table 2. Ablation of JD-BP on AuctionNet. Each row removes one component from the full model: DPO, the GCA module, or the memoryless RTG (m-RTG). Higher is better.

Figure 2. Online deployment workflow of the bidding and pricing system. User requests are processed by the Ad Server, followed by value prediction and bidding & pricing modules. The results are submitted to the auction and charging components, and all operational data are recorded and monitored by the Bidding & Pricing Perception Center for continuous optimization.

Deployment stage	Ad Revenue	Target Cost	Achievement
Stage 1: Memoryless bidding + PID pricing	$+3.40\%$	$+5.10\%$	$+3.63$ pp
Stage 2: + Generative joint (JD-BP)	$+1.30\%$	$+1.38\%$	$+0.29$ pp
Overall (JD-BP vs. base)	$+4.70\%$	$+6.48\%$	$+3.92$ pp

Table 3. Staged online A/B deployment, also serving as an online ablation of the generative pricing model. Stage 1 (memoryless bidding + rule-based PID pricing) is measured against the pre-deployment base; Stage 2 (the full generative-joint JD-BP, replacing PID pricing) is measured against Stage 1 and isolates the generative model’s contribution; the last row is JD-BP’s cumulative effect over the base. Achievement is the fraction of ads with CPA/TCPA in

[0.8,1.2]

Automated Bidding

Automated bidding studies algorithms that choose auction bids over time while balancing value, budgets, and performance constraints.

See all 35 papers

PlatformBid: An Auto-Bidding Benchmark from a Unified Advertising Platform's Perspective
Jul 29, 2026Shengtian Yang, Yewen Li, Peng Jiang +5Benchmark Design Automated Bidding
Nanyang Technological University Singapore, Singapore
Generative Auto-Bidding with Unified Modeling and Exploration
May 19, 2026Mingming Zhang, Feiqing Zhuang, Na Li +7Automated Bidding Online Advertising
Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University · Taobao & Tmall Group of Alibaba
DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation
Jun 12, 2026Miduo Cui, Haochen Wang, Shangqin Mao +6Distributional RL Automated Bidding
Shandong University, Jinan, China · Meituan, Beijing, China

Jul 29, 2026J/K move · Enter open · S save

PlatformBid: An Auto-Bidding Benchmark from a Unified Advertising Platform's Perspective

Shengtian Yang, Yewen Li, Peng Jiang, Zhiyi Lyu+4

Nanyang Technological University Singapore, Singapore

Benchmark Design Automated Bidding Online Advertising Real-Time Bidding

Open paper AskPDF arXiv

Real-time bidding is central to computational advertising, comprising three elements: Supply Side Platform (SSP) selling ad impressions, Demand Side Platform (DSP) bidding for advertisers, and Ad Exchange conducting auctions between them. Traditional auto-bidding algorithms focus solely on the DSP side, maximizing advertiser conversions by adjusting bids against competitors. However, current big ad platforms, such as social media and e-commerce companies, now integrate SSP, DSP, and Ad Exchange functions internally. From such ad platforms' perspective, the goal of the auto-bidding algorithms is not only to maximize the advertisers' conversions, but also the total revenue of the platform. Given the lack of platform-centric evaluation frameworks and the pressing need to advance auto-bidding research, we propose PlatformBid - the first comprehensive benchmark designed from a unified ad platform's perspective. To accurately reflect the real-world auto-bidding scenarios, we define three representative settings: (1) homogeneous competition with identical algorithms across advertisers, (2) heterogeneous competition with diverse algorithmic strategies, and (3) promotional competition where some advertisers surge budgets for boosting sales during promotional events like Black Friday. We systematically evaluate a broad spectrum of existing auto-bidding methods across these settings, encompassing classical control methods, RL-based methods, and recent generative methods. Besides these methods, we further propose a novel auto-bidding method based on flow-matching, termed BidFlow, which leverages the flow-matching method's expressive policy representation to effectively handle dynamic competitive environments. Online experiments on Kuaishou further show a +0.68% improvement in target cost, providing deployment evidence for the offline-online consistency of PlatformBid.

Abstract

Automated Bidding

Explore similar work

PlatformBid: An Auto-Bidding Benchmark from a Unified Advertising Platform's Perspective

Generative Auto-Bidding with Unified Modeling and Exploration

DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation