Evolution of Effectiveness Advertising Bidding Strategies: From Single to Adaptive Dual Bidding

Effectiveness advertising is a game‑theoretic scenario: media platforms aim to sell traffic at the highest price, while advertisers seek to meet cost and downstream performance targets. As the industry matures, advertisers increasingly demand optimization of deep‑level metrics such as second‑day retention, payment rate, and first‑day ROI, beyond shallow conversions like activation.

The article reviews the evolution of bidding models for effectiveness advertising.

Project Background

Media platforms typically use a dual‑bidding mechanism to protect both shallow conversion costs and deep‑level outcomes. Advertisers set three goals: shallow cost compliance, deep cost compliance, and an implicit deep conversion rate target. Platforms must design an appropriate eCPM (effective cost per mille) formula to satisfy advertisers while maximizing platform revenue.

Project Timeline

2.1 Traditional Single Bidding

Advertisers bid for a single conversion goal (e.g., download, activation). Platforms estimate click‑through and conversion rates, converting conversion value into exposure value. An additional bid factor α calibrates the bid to ensure advertiser cost targets.

The bid factor serves two roles: (1) calibrating the estimated value, and (2) filtering high‑quality traffic by dynamically adjusting bids based on estimates and audience tags.

Before 2022, the factor evolved from simple inverse proportional control to minute‑level, multi‑dimensional dynamic aggregation, supporting cold‑start, regular, and aggressive strategies. As prediction models improved, the factor shifted toward quality‑traffic selection, with shallow costs stabilized by accurate models.

2.2 Weighted Dual Bidding

Advertisers set two bids—shallow and deep (e.g., activation + second‑day retention). The platform adds a deep‑cost control factor to the shallow‑only formula, creating a weighted cost factor. This addresses deep‑cost control but introduces issues:

Shallow and deep costs influence each other, causing oscillations.

The bid factor, a feedback controller, only uses same‑day data and lacks historical deep‑conversion estimates (no pDCVR).

Iterative improvements were pursued.

2.3 Pareto Dual Bidding

Weighted bidding yields overall optimal cost but not separate optimality for shallow and deep layers. Treating the problem as a Pareto‑optimal solution, the platform incorporated a deep‑conversion rate estimator pDCVR into a new dual‑bidding formula, aiming for stable control of both shallow and deep costs.

Because deep conversions are sparse, a correction coefficient w was introduced to boost volume while maintaining deep‑cost control. Real‑time online control of w further refined performance.

2.4 Adaptive Dual Bidding

To overcome remaining challenges, the next version leveraged continuous improvements in pDCVR accuracy and introduced a non‑linear S‑shaped φ function to generate a deep‑effect control factor. The φ function’s curvature η adapts based on the gap between actual deep conversion rate (DCVR) and target (T):

If DCVR ≥ T, η is minimal; φ≈constant, allowing aggressive volume while meeting shallow cost.

If DCVR < T, η increases, penalizing low‑pDCVR traffic and rewarding high‑pDCVR traffic until DCVR approaches T.

If DCVR ≪ T, η reaches maximum, discarding low‑quality traffic and aggressively bidding high‑pDCVR traffic.

This adaptive mechanism, combined with deep‑conversion estimations (payment rate, retention, ROI), enables selective bidding that favors high‑value traffic while automatically adjusting to performance gaps.

Online Results

Multiple iterations were deployed, cumulatively driving improvements in effectiveness advertising metrics (specific numbers omitted for brevity).

Summary and Outlook

All bidding strategies rely on robust model predictions, audience tagging, budget allocation, and auction design. As advertiser goals become stricter, bidding models must evolve. Future work includes:

Deep‑effect control under extreme data sparsity.

Optimization under delayed deep conversions.

Mitigating performance volatility caused by real‑time auction environment changes.

Developing an environment‑aware automatic bidding framework.

These directions aim to further enhance the alignment between shallow cost compliance and deep conversion performance.