Pat Dean

a brand new ad on Meta may no longer need weeks of engagement history before the system understands where it belongs meta has rebuilt 3 layers of its ad system in the latest update and the easiest way to understand how it works is to imagine a personal shopper. 1. andromeda searches the store it reduces tens of millions of eligible ads to the few thousand worth ranking for each person. the new version routes ads through specialised expert LLM models and scores more than 1 billion ads per second. this means it can search more of the store instead of repeatedly checking the most familiar shelves, so more of your ads can make the shortlist 2. GEM understands the shopper gem is the foundation model learning across facebook, instagram, reels, different objectives and trillions of signals. it now reads 2x more engagement history through a 2.5x more complex model, then transfers that knowledge to the smaller models ranking ads in real time. this means it can predict what someone is most likely to want next 3. GenRecs understands new products an LLM reads the text and images in each ad, then turns attributes like brand, colour, fit and style into semantic tokens. this lets meta relate a new ad with no engagement history to similar ads and products it already understands. instead of waiting to learn everything from clicks, the system can reason about what the ad is and who might want it from day one. what this means for advertisers: 1. more creative variety gives the retrieval system more possible matches. 2. new ads and products should begin learning faster. 3. fragmented campaign structures and manual audience control make even less sense. but better AI also makes the quality of your inputs more important. do you want as many sales as possible, or only from new customers? if you report every sale equally, meta can’t know which customers or orders are more valuable. and ad quality now extends beyond the creative. checkout friction, shipping delays, unexpected charges, support and product satisfaction can all affect whether your ad wins the auction. the system is taking over more of the delivery decisions. our job is to give it better creative, cleaner data and a more honest definition of what winning actually means.