Dynamic Click Destinations: Personalizing Post Ad-Click Landing Experiences
Apurva Pathak, Yuya Shiraki, Iman Jenabzadeh, Nikhil Bhale, Yuwen Tan, Shashank Rangarajan, Amirmehdi Sharifzad
Useful problem framing and large-scale deployment evidence, but a simple, loosely 'A2C'-branded distillation method with no uncertainty estimates, ablations, or external baselines.
Abstract
Apps that monetize outbound (offsite) ad clicks have multiple post ad-click levers that influence user and business outcomes, including the landing page shown on the advertiser's site, whether the click resolves to a webpage or an app on the user's device, and whether it resolves to a native in-product surface or an external website. We refer to personalization over these levers as post ad-click destination automation. The optimal destination is heterogeneous across users, advertisers, placements, devices, and inferred intent, making static, one-size-fits-all destination policies suboptimal. We present Dynamic Click Destinations (DCD), a production personalization framework that selects a destination for each eligible offsite ad click to maximize Advertiser Value (AV), a value-weighted conversion outcome. DCD applies to any post-click destination decision where the system can choose among destination options at click time and downstream conversion feedback is available. We describe an evolution of learning objectives from post-conversion destination inference, to two-task destination-conditioned conversion prediction, to a single-task policy model that directly predicts which of two destinations is better for a given click, trained with Advantage Actor Critic (A2C)-inspired soft labels derived from a higher-capacity destination-conditioned "critic" model. In a web-vs-app case study, online A/B experiments at scale on a major advertising platform show that per-click personalization with DCD outperforms both static routing policies, with the single-task objective delivering the largest gains. These results highlight post ad-click destination as an important and under-explored lever in conversion optimization and demonstrate that personalized destination automation can outperform one-size-fits-all destination policies.
AI Impact Assessments
(1 model)Core Contribution
The paper names and formalizes "post ad-click destination automation": choosing at click time where an ad click lands. The options are a landing page, web versus app, or native versus external surface. The deployed framework, Dynamic Click Destinations (DCD), is a Meta production system. The case study is the binary web-vs-app decision. The modeling contribution is a progression through three generations:
1. Post-conversion destination inference, using the conversion destination as a proxy label.
2. Two-task destination-conditioned CVR prediction.
3. A single-task model that predicts "alternative is better". It is trained on soft labels, σ(α·(logit_App − logit_Web)), taken from a larger two-head "critic" model.
The "A2C-inspired" label is essentially distillation of a logit difference from a T-learner-style critic into a single student model that targets the decision boundary. The actor-critic framing is loose. There is no policy gradient, no state transitions, and no actual advantage function, so the name is more branding than substance. The substantive idea is real but modest: train directly on the relative-advantage target instead of subtracting two absolute CVR heads. This is closely related to known uplift modeling, meta-learner, and direct-policy-learning approaches.
Methodological Rigor
The strongest part is the evaluation. Online A/B tests run three weeks with 10% traffic cells on a very large platform. The always-App holdout (−1.05% AV) is a useful demonstration that no static policy dominates, and a +0.1% gain on purchase-optimizing ads suggests heterogeneity. Results rise monotonically from Gen 1 to Gen 3 (+0.06%, +0.42%, +0.54%).
The weaknesses are substantial:
Potential Impact
The practical upside for large ad platforms is real. Post-click experience is under-studied, and AV gains of about 0.5% at this scale are commercially meaningful. The framing could prompt other platforms (messaging, e-commerce, AI apps) to treat the landing destination as a decision variable. Scientific influence is limited, though. The method is simple, the paper is a short industry workshop paper, and its results are not independently replicable. The generalization claims (landing page personalization, native vs. external, multi-option) are asserted but not tested. Researchers will likely cite it as an example of a deployed system and a problem framing rather than build on the method.
Timeliness & Relevance
The paper is relevant to industrial ad optimization and to the broader trend of decision-aligned objectives in recommender systems. The observation that offline metrics decouple from online decisions under a two-head formulation is a known and important practical issue, so its confirmation here is useful. There is also growing interest in AI-driven apps that mediate outbound clicks. The topic is niche, however.
Strengths
Limitations
Overall
This is a competent industrial workshop paper with a useful problem framing and credible large-scale deployment results, but modest methodological novelty and limited rigor in analysis and reporting. Its scientific impact is likely confined to ad-tech practitioners and a handful of citations.
Generated Nov 3, 2026
Comparison History (0)
Useful problem framing and large-scale deployment evidence, but a simple, loosely 'A2C'-branded distillation method with no uncertainty estimates, ablations, or external baselines.