Meta's New Architecture Validates Scaling Laws in Ads, Lifting IG Conversions by 6%

Meta_Engineers · x · 2026-08-07

Meta's engineering team detailed architectural breakthroughs in their ads recommendation system. The newly introduced multi-layer sequence learning architecture demonstrates LLM-like scaling laws while delivering high performance in latency-sensitive ads ranking models.

Core innovations include:

This unified platform is now a core component of Meta’s Generative Ads Recommendation Model (GEM). In production, the advancements have contributed to a cumulative lift of 6% in conversions on Instagram, 3% in conversions on Facebook, and 3.5% in ad clicks on Facebook.

Original post →

More from Research

Research channel →