Meta’s Andromeda boosts ad retrieval with 6% recall gain
Meta has developed Andromeda, a personalized ads retrieval engine leveraging NVIDIA Grace Hopper Superchip and Meta Training and Inference Accelerator (MTIA) hardware, to enhance its Advantage+ automation for ad recommendations. This system utilizes deep neural networks and hierarchical indexing to improve personalization and ad performance, achieving a +6% recall improvement and +8% ads quality improvement on selected segments within Instagram and Facebook applications. Andromeda is designed to scale with the exponential growth of ad creatives generated by generative AI tools, increasing efficiency and return on ad spend for advertisers.
Key Takeaways
- Retrieval is the first stage in Meta’s ads recommendation system, filtering tens of millions of ad candidates down to a few thousand.
- Andromeda’s deployment across Instagram and Facebook delivered a 6% recall improvement and an 8% ads quality improvement on selected segments.
- Meta says more than a million advertisers used its GenAI tools in a month to create more than 15 million ads.
- Advantage+ automation is increasing eligible ads through audience creation, budget allocation, dynamic placement, and creative generation.
- Meta says its hierarchical index and model elasticity are designed to scale retrieval as ad creatives grow from AI-generated content.
Why It Matters
Meta is pushing more complexity into the retrieval layer, where ad candidates are narrowed before ranking models make the final selection. The immediate effect is better personalization and higher return on ad spend for advertisers, at least on the segments Meta measured. More broadly, Andromeda shows Meta adapting its ads stack to the growth in AI-generated creatives and to the demands of near real-time automation in Advantage+. The key watch item is whether Meta extends the reported 6% recall and 8% ads quality gains beyond the selected Instagram and Facebook segments.
Read full article at engineering.fb.com