Meta
The Machine Learning Systems Behind Meta
Learn about machine learning and how it powers Meta's ad platform.
Learn how to maximize Meta's machine learning for ad buying by stacking catalog ads, Look-a-Like audiences, and bid strategies.
When it comes to maximizing Meta's machine learning capabilities, most brands and ad buyers think they are taking full advantage, when really they're just scratching the surface.
Here's three ways to go from playing Checkers to playing Chess with your Meta ad buying.
▶ ▶ Stack Catalog Ads
Catalog ads can't be very efficient and especially powerful for businesses with a wide range of products. But the settings within your catalog matter. Make sure to go head to head with these variation:
- single image vs carousel
- all products vs specific product sets
- standard catalogs vs 3rd party custom catalogs
💡 Key insight:
There are many permutations of how to set up your catalog. Have them compete against each other as separate ads in one ad set to learn what works best.
▶ ▶ Stack LaL audiences to make them broad
If you use a single 5% Look-a-Like, your audience isn’t big enough to unlock the benefits of broad targeting. But generate 5-10 unique LaL audiences from different sources and stack them in a single ad set to build a giant look-a-like super pool.
💡 Key insight:
Stacked audiences automatically deduplicate. Keep stacking LaL audiences until you achieve your target broad audience size.
▶ ▶ Stack auto-bid and manual-bid campaigns
When it comes to auto-bid vs manual bid, it’s a “yes and” situation where you want both. Duplicate an auto-bid campaign, and turn it into a manual bid campaign to unlock incremental volume.
💡 Key insight:
Take advantage of auto-bid for predictability and consistency. Take advantage of manual-bid for incremental scale when the price is right.
--------
To unlock the full potential of Meta's machine learning, think about stacks-on-stacks-on-stacks of optionality for the system to iterate through and learn from.
Here's three ways to go from playing Checkers to playing Chess with your Meta ad buying.
▶ ▶ Stack Catalog Ads
Catalog ads can't be very efficient and especially powerful for businesses with a wide range of products. But the settings within your catalog matter. Make sure to go head to head with these variation:
- single image vs carousel
- all products vs specific product sets
- standard catalogs vs 3rd party custom catalogs
💡 Key insight:
There are many permutations of how to set up your catalog. Have them compete against each other as separate ads in one ad set to learn what works best.
▶ ▶ Stack LaL audiences to make them broad
If you use a single 5% Look-a-Like, your audience isn’t big enough to unlock the benefits of broad targeting. But generate 5-10 unique LaL audiences from different sources and stack them in a single ad set to build a giant look-a-like super pool.
💡 Key insight:
Stacked audiences automatically deduplicate. Keep stacking LaL audiences until you achieve your target broad audience size.
▶ ▶ Stack auto-bid and manual-bid campaigns
When it comes to auto-bid vs manual bid, it’s a “yes and” situation where you want both. Duplicate an auto-bid campaign, and turn it into a manual bid campaign to unlock incremental volume.
💡 Key insight:
Take advantage of auto-bid for predictability and consistency. Take advantage of manual-bid for incremental scale when the price is right.
--------
To unlock the full potential of Meta's machine learning, think about stacks-on-stacks-on-stacks of optionality for the system to iterate through and learn from.