Intelligent Advertising Delivery System: Budget‑Constrained Bidding, Multi‑Constraint Bidding, Sequential Allocation, and Multi‑Channel Optimization

In the rapidly digitizing economy, internet advertising plays a crucial role in helping merchants market their products, and advertisers aim to maximize marketing effectiveness under limited resources.

The article first outlines the three‑party advertising ecosystem (advertiser, media platform, user) and explains why sophisticated bidding mechanisms are needed to handle diverse pricing models (CPM, CPC), contract plans, and multi‑dimensional optimization goals.

1. Advertising Business Background – Advertisers pay media platforms to deliver marketing messages to target users, influencing purchases and creating a flow of information and money.

2. Historical Evolution of the Advertising System – Starting from 2016, Alibaba’s ad system progressed from basic CPM/CPC to value‑estimated OCPM/OCPC, contract‑guaranteed delivery, Budget‑Constrained Bidding (BCB), Multi‑Constrained Bidding (MCB), and finally to multi‑channel sequential budget‑constrained bidding that optimizes long‑term user value.

Key research results have been published at top conferences such as KDD, CIKM, and ICML.

3. Technical Capabilities Summary

OCPC/OCPM – Optimized Cost‑Per‑Click adjusts the original bid by a factor (pValue/baseValue) derived from a machine‑learning model that predicts the value of each individual traffic unit, enabling per‑impression value‑based pricing.

Budget‑Constrained Bidding (BCB) – The platform automatically bids under a fixed budget by selecting traffic with the highest value‑to‑cost ratio (λ*), discarding expensive impressions, and guaranteeing that the total spend does not exceed the budget.

Multi‑Constrained Bidding (MCB) – Extends BCB by adding additional constraints such as average click cost, using dual variables (p*, q*) to balance budget and cost limits, and can be solved via reinforcement learning or PID‑based feedback control.

Multi‑Stage Sequential Bidding (MSBCB) – Addresses scenarios where users need multiple exposures before conversion; it jointly selects users and optimizes per‑user exposure policies (π_i) to maximize long‑term value under budget constraints, proving that a two‑stage greedy‑adjustment algorithm converges to the global optimum.

Cross‑Channel Smart Allocation (CCSA) – Formulates the allocation of a total advertiser budget across search, recommendation, and brand channels as a concave optimization problem, enabling a minimal API that distributes marginal budget where marginal returns are equal, thus achieving global optimality with low coupling between channels.

The system’s evolution demonstrates a shift from coarse value estimation to fine‑grained per‑impression value prediction, from single‑objective to multi‑objective optimization, from manual bidding to fully automated budget‑driven bidding, and from single‑channel to holistic multi‑channel optimization.

Future work will continue to deepen data‑driven insights, mechanism design, algorithmic upgrades, and product iterations to further reduce advertiser pain points.