Dynamic Advertising, Target Conversion Bidding, DMP, Auction Mechanisms, and Optimal Mechanism Design

With the rapid growth of the Internet, online advertising has become a dominant monetization model, giving rise to giants such as Google, Facebook, ByteDance, Alibaba, Baidu, and Tencent. Effective advertising requires delivering the right content to the right user at the right time, which has driven the development of dynamic advertising, target conversion bidding, and data management platforms.

Dynamic Advertising

Dynamic advertising (programmatic creative and Dynamic Product Ads, DPA) tailors creative assets and product displays to individual users based on attributes (gender, age, income, location, device, behavior) and context. Key components include:

Programmatic/Dynamic Creative: Different users see different creative materials; the same user may see different creatives at different stages (first view, non‑first view, unconverted, converted).

Creative Decoupling: Content, title (m) and format (n) are set separately, allowing the system to combine them (m×n) and increase creative variety.

Exposure Strategy: Adjust exposure frequency based on conversion metrics (CTR, CVR) to favor high‑performing creatives.

DPA: Pull product assets from a product library and dynamically display items that match the user's profile.

Applicability: Suitable for multi‑product categories where personalized recommendation is needed.

Product Library Integration: Manual upload for small, low‑frequency catalogs or API integration for large, frequently changing inventories.

Exposure Principle: Match user tags with product tags to serve the appropriate ad.

Goal: Increase relevance and conversion efficiency.

Important considerations include tag mapping between media and product, closed‑loop data feedback, sufficient product variety, dynamic tag adjustment, unique user identification (cookies, IDFA, IMEI), and maintaining stable bidding strategies during model training.

Target Conversion Bidding (OCPM/OCPX)

Advertisers can choose pricing models such as CPM, CPC, CPA, CPS, CPD, but performance‑oriented advertisers often prefer target conversion bidding, where a specific conversion goal (e.g., app install, H5 lead) and an estimated price are set. The platform optimizes delivery based on a conversion‑price model, aiming to achieve the desired conversion while keeping costs under control.

Key steps:

Advertiser defines conversion goal and budget.

Platform collects post‑click conversion data to build predictive models.

Two‑stage bidding: initial machine‑learning phase uses historical average conversion price; later phase uses the platform’s estimated optimal price.

Effective bidding requires accurate conversion tracking, sufficient budget, and well‑designed creatives and landing pages.

Data Management Platform (DMP)

A DMP aggregates user data from multiple sources, creates fine‑grained audience segments, and enables precise targeting. Core functions include:

Creating segmented audiences via built‑in tags, uploaded user lists, or behavior‑driven generation.

Analyzing user profiles across dimensions (gender, age, location, device, interests) and calculating target group indices (TGI).

Look‑alike expansion to find users similar to a seed audience.

Retargeting based on conversion status.

Audience operations such as union, intersection, and exclusion.

Evaluating campaign performance and iteratively refining audience definitions.

DMPs provide the foundation for personalized ad delivery, dynamic product ads, and frequency control.

Ad Auction Core

Online ad delivery relies on real‑time bidding (RTB), which is essentially an auction. Various auction formats exist (English, Dutch, first‑price sealed, second‑price sealed). Modern ad exchanges typically use a second‑price (or VCG) auction where the winner pays the second highest bid, adjusted by estimated click‑through rate (CTR) to compute an effective CPM (ECPM = 1000 × CTR × bid).

Key concepts:

Allocation rule (who gets the impression) and payment rule (how much they pay).

Incentive compatibility: truthful bidding is a weakly dominant strategy in second‑price auctions.

Efficiency: the item is allocated to the bidder with the highest true valuation.

Mechanism Design and Optimal Mechanism

Beyond incentive‑compatible mechanisms, platforms may aim to maximize revenue. By defining a virtual valuation function Y(v) = v − (1 − F(v))/f(v) for each bidder, the platform can allocate the impression to the bidder with the highest virtual valuation and charge the inverse of that function evaluated at the second‑highest virtual valuation. This “optimal mechanism” maximizes seller revenue but may introduce price discrimination.

In practice, the platform converts virtual valuations into expected CPMs (e.g., CPM = 1000 × pCTR × Y) and determines the winner’s payment based on the next highest virtual CPM, ensuring revenue optimality while preserving the auction’s real‑time nature.

Author and Source

Author: 莫菲克 (Mophic), an advertising practitioner with a strong interest in ad technology.

Original source: Zhihu (https://www.zhihu.com/people/mophic/posts). Permission required for reproduction.

Additional Information

For further reading, see recommended articles on advertising algorithms at Alibaba, Hulu, and other industry case studies.