Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Feedback Control for Small Budget Pacing

Sreeja Apparaju, Yichuan Niu, Xixi Qi

TL;DR

Budget pacing in real-time online auctions is challenged by dynamic bid environments and noisy signals. The authors introduce a Bucketized Hysteresis Controller that couples hysteresis with proportional feedback to achieve stable, adaptive spend while providing a framework for principled parameter selection. Real-world experiments demonstrate concrete gains, including a 13% reduction in pacing error and a 54% reduction in lambda-volatility relative to a production baseline, with additional efficiency benefits for small-budget campaigns. By bridging control theory with advertising systems, the work offers a scalable, reliable pacing solution that improves delivery predictability in practice.

Abstract

Budget pacing is critical in online advertising to align spend with campaign goals under dynamic auctions. Existing pacing methods often rely on ad-hoc parameter tuning, which can be unstable and inefficient. We propose a principled controller that combines bucketized hysteresis with proportional feedback to provide stable and adaptive spend control. Our method provides a framework and analysis for parameter selection that enables accurate tracking of desired spend rates across campaigns. Experiments in real-world auctions demonstrate significant improvements in pacing accuracy and delivery consistency, reducing pacing error by 13% and $λ$-volatility by 54% compared to baseline method. By bridging control theory with advertising systems, our approach offers a scalable and reliable solution for budget pacing, with particular benefits for small-budget campaigns.

Feedback Control for Small Budget Pacing

TL;DR

Budget pacing in real-time online auctions is challenged by dynamic bid environments and noisy signals. The authors introduce a Bucketized Hysteresis Controller that couples hysteresis with proportional feedback to achieve stable, adaptive spend while providing a framework for principled parameter selection. Real-world experiments demonstrate concrete gains, including a 13% reduction in pacing error and a 54% reduction in lambda-volatility relative to a production baseline, with additional efficiency benefits for small-budget campaigns. By bridging control theory with advertising systems, the work offers a scalable, reliable pacing solution that improves delivery predictability in practice.

Abstract

Budget pacing is critical in online advertising to align spend with campaign goals under dynamic auctions. Existing pacing methods often rely on ad-hoc parameter tuning, which can be unstable and inefficient. We propose a principled controller that combines bucketized hysteresis with proportional feedback to provide stable and adaptive spend control. Our method provides a framework and analysis for parameter selection that enables accurate tracking of desired spend rates across campaigns. Experiments in real-world auctions demonstrate significant improvements in pacing accuracy and delivery consistency, reducing pacing error by 13% and -volatility by 54% compared to baseline method. By bridging control theory with advertising systems, our approach offers a scalable and reliable solution for budget pacing, with particular benefits for small-budget campaigns.

Paper Structure

This paper contains 29 sections, 6 equations, 1 figure, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. System Setup
  4. Bidding and Auction
  5. Pacing
  6. Pacing Controller
  7. Lambda Update
  8. Scale Selection
  9. Bucketized Hysteresis Controller
  10. Controller Design
  11. Controller Methodology
  12. Error Calculation
  13. Error Discretization
  14. Lambda Update
  15. Online Experiments
  16. ...and 14 more sections

Figures (1)

  • Figure 1: For each strategy (a-e), the left panel shows the daily cumulative spend of our Test controller (red) against the Production controller (blue) and an Ideal target (black, dashed). The right panel shows the corresponding controller output, $\lambda$.