Some advertisers have noticed that Meta’s ad system has been faltering in the last two months.
The ad buyers note that the cost of advertising in Meta Platforms is up significantly while results are mixed.
“Our ads system is working as expected for the vast majority of advertisers,” a Meta spokesperson said in a statement. “We recently fixed a few technical issues and are researching a small amount of additional reports from advertisers to ensure the best possible results for businesses using our apps.”
As a result, some ad buyers have shifted budgets to TikTok and Alphabet.
Temu, one of Meta’s largest advertisers said it hasn’t notice any advertising issues with Meta.
I=6
Meta said its Andromeda, a proprietary machine learning (ML) system that utilizes NVIDIA Grace Hopper Superchip and Meta Training and Inference Accelerator (MTIA) and is used for retrieving relevant ad creatives resulted in a +6% recall improvement to the retrieval system and a +8% improvement in ads quality in selected segments.
AI innovations at Meta that driver advertiser performance
Meta recently detailed four AI tools and technique driving ad performance: Andromeda, Sequence Learning, Lattice, and GEM.
Meta GEM (Generative Ads Recommendation Model) is a large-scale, generative AI model trained on thousands of GPUs to process trillions of data points and deliver hyper-personalized ad recommendations. Meta said during initial launch on Reels this year, it has increased ad conversions by up to 5%.
Sequence Learning is a modeling technique that tracks the order of user actions before and after ad interactions, helping predict what ads are most relevant next. Since being adopted last year, it has delivered a 3% increase in conversions in tested segments.
Andromeda is a new end-to-end ML system introduced in 2024 (hardware + software) that enables much more complex ad retrieval and personalization at scale. It has led to an 8% increase in ads quality, enhancing tools like Advantage+ and GenAI ad creative generation.
Meta Lattice is a unified ad ranking system that allows Meta to generalize learning across ad formats and improve efficiency by operating fewer models. Launched in 2023, this model has increased ad quality by almost 12% and ad conversions by up to 6%.