Research map / Learning theory and algorithms / Submodular and game algorithms
Beyond Last-Click: An Optimal Mechanism for Ad Attribution
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Formulates ad attribution as a mechanism design problem and proposes the Peer-Validated Mechanism (PVM), a DSIC rule that attributes credit based on other platforms' reports, proving it optimal for homogeneous platforms and showing last-click is not DSIC and performs poorly.
Problem
Multi-platform ad attribution under last-click rules is vulnerable to strategic report timing, so platforms can misreport to capture credit; the field lacks a theoretical incentive-compatible attribution mechanism with accuracy guarantees.
Approach
A platform's credit depends only on other platforms' eligible reports (compared against a prior-calibrated threshold), so its own report cannot affect its own allocation, making the mechanism DSIC by design while matching the true last-click probability in expectation.
Key results
- PVM is DSIC and optimal among DSIC mechanisms in the homogeneous setting, with exact accuracy 1-(1-1/n)(1/n)^(1/(n-1)); for n=2 this is 0.75 versus LCM's worst-case upper bound of about 0.3431.
- For two heterogeneous platforms, PVM's exact worst-case accuracy is 19/27 (about 0.7037), whereas LCM's accuracy can approach 0.
- In simulations with n=5 homogeneous platforms, PVM improves accuracy over LCM by 0.3041 on average.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace LCM with PVM | Advertising attribution · Homogeneous platforms, n=2 | accuracy: (2-\sqrt{2})^2\approx0.3431 -> 3/4=0.75 | Helped |
| Replace LCM with PVM | Advertising attribution · Heterogeneous platforms, n=2 | accuracy: 0 -> 19/27\approx0.7037 | Helped |
| Replace LCM with PVM | Advertising attribution · Homogeneous platforms, n=2 | accuracy: 0.0404\pm0.0396 | Helped |
| Replace LCM with PVM | Advertising attribution · Homogeneous platforms, n=3, 4, and 5 | accuracy: 0.1583\pm0.0439; 0.2444\pm0.0580; 0.3041\pm0.0490 | Helped |
| Replace LCM with PVM | Advertising attribution · Heterogeneous platforms, n=2 (6 pairs) | accuracy: 0.0655\pm0.0283 | Helped |
| Replace LCM with PVM | Advertising attribution · Homogeneous platforms, n=2, 3, 4, and 5 | fairness: 0.0248\pm0.0089; 0.0157\pm0.0034; 0.0107\pm0.0047; 0.0111\pm0.0040 | Helped |
| Replace LCM with PVM | Advertising attribution · Heterogeneous platforms, n=2 (6 pairs) | fairness: 0.1320\pm0.0598 | Helped |
| Use PVM instead of LCM | Advertising attribution · Homogeneous platforms | accuracy: 1 - \left( 1 - \frac{1}{n} \right) \left( \frac{1}{n} \right)^{\frac{1}{n-1}} | Helped |
| Use PVM instead of LCM | Advertising attribution · Heterogeneous platforms | accuracy: (19/27)^{\lceil\log_2 n\rceil} | Helped |
| Use PVM instead of LCM | Advertising attribution · Any platform distribution | fairness: 1 | Helped |
| Use LCM | Advertising attribution · Any platform distribution | DSIC | Hurt |
Surprising findings
- Under heterogeneity, LCM's accuracy and fairness can be driven arbitrarily close to zero even with just two platforms.
Limitations
- Assumes independent click-time distributions; correlated extension only sketched.
- Models strategic delay only, not click spamming or injection.
- Full proofs deferred to a full version not included in the paper.
Experimental setup
Datasets: real-world ad conversion logs from four advertising platforms (used to fit click time distributions)
Benchmarks: simulated user paths fitted to real ad-conversion logs
Metrics: attribution accuracy fairness score
Baselines: Last-Click Mechanism
Builds on: last-click attribution Myerson-style incentive compatibility
Closest papers by meaning
- Truthful Aggregation of LLMs with an Application to Online Advertising (NeurIPS 2025 · similarity 0.85)
- LLM-Auction: Generative Auction towards LLM-Native Advertising (NeurIPS 2026 · similarity 0.83)
- Merging Mechanisms for Ads and Organic Items in E-commerce Platforms (AAAI 2025 · similarity 0.82)
- Optimal Auction Design in the Joint Advertising (ICML 2025 · similarity 0.82)
- Parallel Ranking of Ads and Creatives in Real-Time Advertising Systems (AAAI 2024 · similarity 0.82)
- GenAuction: A Generative Auction for Online Advertising (AAAI 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.