Stability and Controllability of Revenue Systems via the Bode Approach

Yichuan Niu Jianhui Chen · 2025

Read Stability and Controllability of Revenue Systems via the Bode Approach by Yichuan Niu Jianhui Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-23663.txt.

Introduction

Yichuan Niu and Jianhui Chen present a 2025 paper on keeping ad budgets on pace with control theory. Online ad systems must spend a budget smoothly despite swinging traffic and auction pressure. Industry practice leans on trial and error, which can wobble or overshoot. The authors model the pacing system as a simple linear system and design a compensator using Bode plots, the classic frequency-domain tool of control engineering.

Their controller aims to cut pacing error while staying stable when conditions shift. Simulations and tests on the authors' in-house auction system showed tighter budget delivery than the usual practice. The method proved resilient to traffic and auction swings. The authors present the work as a bridge: old, proven control methods can discipline modern revenue systems. It offers a systematic alternative to hand-tuned pacing rules.

Catalog record

CREATORYichuan Niu Jianhui Chen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-23663

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Stability and Controllability of Revenue Systems via the Bode Approach Note: Corresponding author email: [email protected] Note: The work was done while the author was with Snap Inc. Yichuan Niu Jianhui Chen Abstract In online revenue systems, e.g. an advertising system, budget pacing plays a critical role in ensuring that the spend aligns with desired financial objectives. Pacing systems dynamically control the velocity of spending to balance auction intensity, traffic fluctuations, and other stochastic variables. Current industry practices rely heavily on trial-and-error approaches, often leading to inefficiencies and instability. This paper introduces a principled methodology rooted in Classical Control Theory to address these challenges. By modeling the pacing system as a linear time-invariant (LTI) proxy and leveraging compensator design techniques using Bode methodology, we derive a robust controller to minimize pacing errors and enhance stability. The proposed methodology is validated through simulation and tested by our in-house auction system, demonstrating superior performance in achieving precise budget allocation while maintaining resilience to traffic and auction dynamics. Our findings bridge the gap between traditional control theory and modern advertising systems in modeling, simulation, and validation, offering a scalable and systematic approach to budget pacing optimization. 1 Introduction In most online revenue systems, monetary flows are controlled and optimized to achieve specific goals. For example, an advertising system’s revenue flows from advertisers to opportunity suppliers (i.e., advertising platforms) are often dictated by the advertising products on advertiser’s choices. Some advertisers prefer to distribute their budgets evenly throughout the day, while others may align spending with request traffic patterns. In this context, budget pacing control is essential to achieve their desired spending objectives. A pacing system creates a budget spending plan (referred to as the desired spend velocity) based on the advertising product being managed. This plan typically spans a single day or multiple days over the course of an advertising campaign. Equipped with one or more feedback loops, the pacing system monitors actual spending performance (the actual spend velocity) and actively adjusts it in near real-time to align with the spending plan. By minimizing the error between the desired and actual spend velocities, the system ensures targets are achieved [ 1 , 2 ] . However, in such systems, designing budget pacing and monitoring actual spending performance can be challenging due to several factors [ 3 , 4 , 5 ] : • Auction Dynamics: Fluctuations in ad retrieval, ranking, and competitive bidding strategies. • Stochastic Traffic Behavior: Variability in user traffic and impression opportunities. • Signal Noise: Discreteness in observed and desired spending signals. • Environmental Variability: Changes in advertiser participants, auction intensity, budget constraints, and other time-dependent conditions. In practice, pacing control is often developed by trial-and-error without a principled methodology for quantitatively addressing the stability, dynamics, and controllability of the system. Fortunately, while this presents a novel challenge for software engineers, many of these concerns can be addressed effectively using Classical Control Theory established for decades [ 6 , 7 ] . In this work, we apply Classical Control Theory to analyze and optimize online revenue systems, with a primary focus on advertising budget pacing. By modeling the pacing system as a linear time-invariant (LTI) proxy, we systematically derive compensators that improve stability and responsiveness while minimizing pacing errors. Using Bode analysis [ 8 , 9 ] , we design a robust control framework that ensures spending velocity remains within desired thresholds, even under dynamic auction conditions. Our approach offers several key contributions: We formulate A control-theoretic framework for budget pacing, introducing mathematically grounded methods for stability and controllability analysis; we design and evaluate compensators, selecting optimal parameters meeting Bode stability criteria; The compensated system actively adapts to traffic fluctuations, bid competition, and auction intensity, ensuring consistent ad delivery and budget utilization; Our methodologies are validated through both simulated experiments and tested under real-world online auctions, demonstrating superior performance over the legacy step-based feedback control systems. By bridging the gap between control theory and modern advertising platforms, this work introduces a scalable, systematic approach to budget pacing. Our findings highlight the advantages of compensator-based pacing over empirical tuning methods, providing a principled alternative for revenue optimization in dynamic, auction-driven environments. 2 Related Work Budget pacing in online advertising is a fundamental problem that ensures advertisers maximize their return on investment while adhering to spending constraints. Approaches to budget pacing include heuristic-based methods, machine learning-driven optimization, and control theoretic frameworks. Below, we survey relevant literature across these key areas. Early budget pacing strategies relied on rule-based heuristics, such as uniform spending over time or proportional allocation based on past performance. For example, a proportional pacing algorithm is introduced in [ 10 ] that adjusts bid multipliers to control spending rates while maintaining competitiveness in ad auctions. Similar approaches have been widely used in industry due to their simplicity and interpretability [ 11 ] . However, heuristic-based methods often lack robustness to dynamic market conditions. Recent research has leveraged machine learning for budget pacing by modeling demand and optimizing bid strategies. In [ 12 ] , Reinforcement learning is applied to dynamically adjust budgets in real-time bidding (RTB) environments. More recently, a deep reinforcement learning framework proposed in [ 13 ] learns optimal bidding strategies from historical auction data. These approaches improve efficiency but require extensive training data and may struggle with interpretability. While heuristic and machine learning-based methods have been widely adopted in industry, control-theoretic approaches provide a rigorous foundation for budget pacing. Niklas Karlsson has contributed significantly to the feedback control methodologies for online advertising from optimization point of view through years. Extensive research, simulation, and experiments have been conducted in his work [ 3 , 4 , 5 , 14 , 15 , 16 ] . Similar study on RTB control can be found in Zhang’s work [ 17 , 18 ] as well as Yang’s work [ 19 ] . The existing literature lacks a connection between control theory and pacing performance in terms of stability and controllability via the Bode approach. Our work addresses this gap by leveraging classical control techniques for budget pacing, ensuring both stability and efficiency in online advertising environments. 3 Real-time Revenue System To analyze and design a compensator for pacing, the “plant” must first be modeled. Since the advertising revenue system exhibits time-varying, non-linear, stochastic, and discrete characteristics, we need to identify a linear time-invariant (LTI) proxy that is sufficiently simple to model yet capable of capturing the key dynamics. 3.1 Bidding and Auction In an advertising system, the probability of conversion is often used to represent the likelihood of a desired event. In this work, we use the probability of an ad click as the conversion event. This probability {p(click)} is provided in real-time by ad-ranking machine learning models based on the ad and the impression opportunity. Assuming {p(click)} is perfectly calibrated, an ad will eventually receive a click when the cumulative summation of {p(click)} across multiple impressions reaches 1.0 . To participate in an auction, advertisers must place a bid, {max\_bid} , representing the amount they are willing to pay for a click. For a single impression opportunity, each ad submits its final bid as follows [ 20 , 21 ] : final\_bid=max\_bid~\times p(click) (1) The ad with the highest bid

Read the complete text