Rethinking Digital Advertising: From Guesswork to User Engagement

Traditional digital advertising often relies on guessing user needs based on incomplete data. Instead, we should focus on understanding users through interactive engagement, creating a more refined and dynamic ad delivery system.

Here are some thoughts on how to achieve that:

1. Increase User Interactions & Positive Sample Rate

• Objective: Design interactions that actively encourage users to engage with ads beyond clicks, increasing the volume of data collected per session.

• Tactics:

• Dynamic Ad Switching: Implement a “shuffle” feature within a single ad slot to allow users to change the ad if it doesn’t resonate. Track these interactions as a proxy for content preference.

• Interactive Ad Units: Create rich media formats like swipeable image carousels, quick polls, or “choose your adventure” stories to capture richer behavioral data.

2. Refine Intent Labels Using Marketing Funnel

• Objective: Move beyond simple “click or not” labels. Use refined intent labels to create a granular understanding of where users are in their journey.

• Tactics:

• User State Tracking: Implement a multi-stage label system to track the user’s stage across awareness, interest, consideration, and conversion.

3. Expand Ad Types Using GenAI & UGC

• Objective: Use generative AI and creator-generated content to diversify ad styles, ensuring there’s always a relevant format for every type of user.

• Tactics:

• Ad Content Generator: Implement a GenAI system that can auto-generate multiple ad creatives based on high-performing templates (e.g., explainer videos, parameter-based comparisons, and visual storytelling).

• Creator Ecosystem: Build an API to allow third-party creators to submit branded content and automatically match it with user segments based on their style preference (e.g., educational, humorous, inspirational).

4. Build Detailed & Reusable User Profiles

• Objective: Develop detailed user profiles that can serve as a long-term competitive advantage (moat) by enhancing both targeting precision and model efficiency.

• Tactics:

• Interest Graphs: Create a multi-layered interest graph that captures not just primary interests but also contextual and session-level nuances (e.g., “workday interest” vs. “weekend leisure interest”).

• Profile Distillation: Use fine-tuning techniques to distill session-level behavior and historical data into compact embeddings, reducing the need for large-scale models.

• Adaptive Profiles: Allow profiles to dynamically update based on real-time interactions and adapt targeting strategies (e.g., shifting from product-focused to lifestyle-based ads as interests evolve).

5. Deploy Efficient Candidate Generation Models

• Objective: Use distilled user profiles and session-level data to build smaller, cost-efficient models that can handle candidate generation without compromising performance.

System Architecture & Implementation Roadmap

1. Data Pipeline:

• Set up a real-time data ingestion pipeline to handle user feedback, engagement data, and contextual signals.

• Use feature stores to maintain session-level and long-term user features.

2. Labeling & Profile Service:

• Build a centralized user state service that maps interactions to funnel stages.

• Implement a profile aggregation service that combines session data with historical data, outputting refined user embeddings.

3. Dynamic Ad Serving:

• Design a dynamic ad delivery system that supports in-session creative switching, real-time updates, and multi-variant testing.

• Introduce a creative generation engine powered by GenAI to scale up content production.

4. Profile Distillation & Efficient Candidate Models:

• Use compact transformer models to distill user behavior into high-quality embeddings.

• Deploy multi-stage candidate generation models with fine-tuning capabilities to minimize system footprint.

By shifting from a reactive to a proactive system, we can transform digital advertising from guesswork into a structured process of user education, engagement, and demand realization. This approach builds deeper user relationships and creates a sustainable competitive advantage for the platform.