Nodes 2025: Graph-Driven Feedback Loops for TikTok Shop

At Nodes 2025, the world's largest graph database conference, I presented alongside David Fagerburg on how we are optimizing TikTok Shop video marketing using Neo4j.

The TikTok Shop Challenge

Brands struggle to understand which creative attributes (hooks, shots, voiceovers) actually drive sales. Traditional metrics like "likes" or "views" don't always correlate with GMV.

Our Solution: The Knowledge Graph

We built a Neo4j-powered knowledge graph that bridges the gap between creative elements and financial performance.

Technical Workflow:

1. Extraction: Multimodal LLMs and vision models analyze videos to extract attributes.
2. Mapping: Attributes are linked to real-time sales data and engagement metrics in Neo4j.
3. Optimization: The resulting feedback loop identifies high-performing creative patterns across 20+ TikTok Shop brands (including Phillips, Thorne, and Optimum).

Beyond the Graph

By combining GraphDBs with LLMs, we've unlocked a next-gen marketing optimization system that helps brands scale what works based on data, not intuition.