Meta Lattice
- Also called Lattice
- Updated November 14, 2026
Meta Lattice is the model architecture Meta uses to predict how an ad will perform. It replaced hundreds of smaller models, each built for one app, placement or advertiser goal, with a shared system that learns across all of them. Meta describes it as the part that ranks ads, after Andromeda narrows the candidates.
How does Meta describe Lattice?
Meta introduced it in a May 2023 post, AI-driven ads performance and efficiency with Meta Lattice, as "a new model architecture that learns to predict an ad's performance across a variety of datasets and optimization goals that were previously supported by numerous smaller, siloed models."
Its October 2026 summary for advertisers, How Meta's Ads System Matches Your Ad to the Right Person at the Right Time, puts it in one line: "Meta Lattice ties it together as one ranking system that learns across advertiser goals and across our apps, instead of a separate model and data for each."
How is Lattice different from Andromeda and GEM?
Meta names three systems, each with a different job:
| System | Job, in Meta's words |
|---|---|
| Andromeda | Retrieval: it "narrows tens of millions of ads to the few thousand worth ranking for each person" |
| GEM | A large foundation model that "learns from trillions of signals and shares what it learns with the smaller models that choose ads in real time" |
| Lattice | "one ranking system that learns across advertiser goals and across our apps" |
None of them has a setting in Ads Manager. Meta says the gains "reach your account with no work from you."
What does Lattice change for advertisers?
Three things Meta claims in the 2023 post:
- Learning across placements and goals. On Instagram, "knowledge-sharing across its different surfaces (e.g., Feed, Story, and Reels) and across various advertiser objectives (e.g., clicks, video views, and conversions) increased performance for advertisers."
- Help with little data. Lattice "is capable of generalizing learnings across domains and objectives, which is especially crucial when the model has limited data to train on." Meta mentions the "cold start" problem on new products and surfaces.
- Slow conversions. A purchase can follow an ad by days. Meta says Lattice captures "a person's real-time intent from fresh signals" and "long-term interest from slow, sparse, and delayed signals."
What results does Meta attribute to Lattice?
| Source | Meta's figure |
|---|---|
| May 2023 post, Instagram deployment | "~8 percent improvement in ads quality" |
| December 2025 paper by Meta researchers | "10% revenue-driving top-line metrics gain, 11.5% user satisfaction improvement, 6% boost in conversion rate, with 20% capacity saving" |
Both are Meta's measurements across its whole system, not results for one account. The 2023 post also gives its scale: Lattice "has trillions of parameters" and is "trained on hundreds of billions of examples from thousands of data domains."
What it means for your ads
- What people do with your ad on one placement can shape how Meta ranks it on another, because one system learns across Feed, Stories and Reels.
- Splitting campaigns by placement or goal no longer gives each part a model of its own. Meta's push to consolidate ad sets follows from the same design.
- Small advertisers and new campaigns are the case Meta says Lattice was built for: it shares what it learns from other objectives when yours has few conversions.
- Results attributed to Lattice are system-wide averages. Your own learning phase and auction still decide what happens to a given ad set.
Common mistakes
- Treating Lattice as a feature you can turn on. It sits behind every campaign.
- Confusing it with Andromeda. Andromeda picks which ads are worth ranking; Lattice ranks them.
- Quoting the 2025 paper's figures as recent results. They describe Meta's deployment up to the paper's date.
Sources
- AI-driven ads performance and efficiency with Meta Lattice — Meta AI, May 11, 2023 (accessed November 14, 2026)
- How Meta's Ads System Matches Your Ad to the Right Person at the Right Time — Meta for Business, October 28, 2026 (accessed November 14, 2026)
- Meta Lattice: Model Space Redesign for Cost-Effective Industry-Scale Ads Recommendations — arXiv, December 9, 2025 (accessed November 14, 2026)
Meta restates these figures with each update, so check the source before quoting one.
Terrads makes each placement's file at its own size, so every surface Lattice learns from gets an ad built for it.