To continue on this "Andromeda is the boogieman" narrative that's taken over X and LinkedIn from digital marketers here's some data for you... I asked X's Grok to show me the timeline of how this started. The initial blog post was from December 2024 and mentioned on X by @eric_seufert . From there it had some initial conversations (even though it was rolling out). As you can see starting in Sept - Nov it's skyrocketed in popularity. We are confusing things. I believe that yes, Andromeda caused hiccups and massive changes for brands. I believe that yes, in some instances performance was impacted by Meta's shifts. I don't believe that your massive performance declines you're all experiencing in Q3 early Q4 are directly caused by this. Get lean, get diverse, get creative, focus on offers + marketing and stop trying continuing to push this narrative that Meta's ad retrieval system is to blame for your declining business. It's more than likely all economic factors.

Whatever it takes

Google’s Network business is vestigial (part 2) mobiledevmemo.com/googles-ne…

Snap $SNAP up 23% AH after reporting Q3 earnings.

The size of the Google model which Apple will use to power the new Siri - for which Apple will pay Google $1 billion annually(!) - is 1.2T.

One observation about the election: Biden-era anti-consumer antitrust is dead, including the Brown Shoe-driven idea that we should be condemning things because they lead to lower prices.

👀

I’ve seen a good deal of commentary about Meta’s Andromeda system, and its recent update, on LinkedIn lately. I wrote about the update a few weeks ago, with some thoughts on how, generally, advertisers should think about adapting their creative testing processes to the new operating regime that retrieval tools like Andromeda were built to address: dramatic growth in the volume of creative deployed across the digital advertising ecosystem. Meta published a conceptual whitepaper back in December 2024, when it launched Andromeda. Aside from that, very little public information exists about how it functions. That’s why I was pleased to host Meta’s VP Monetization Infrastructure, Ranking, and AI on the Mobile Dev Memo podcast back in June to discuss the mechanics of Andromeda, as well as Meta’s other tentpole AI infrastructure tools. The podcast goes deep into the details of how Andromeda functions and fits into the broader digital advertising workflow. I’m biased, but I think it’s a helpful place to start in understanding these types of tools. And understanding this type of infrastructure which will be increasingly critical for digital advertisers as those tools continue to be developed and adopted. mobiledevmemo.com/podcast-me…

Google’s Network business is vestigial (part 2) In Q3 2025, Alphabet’s advertising business, broadly, saw 12.7% growth, with 14.5% growth in Search and 15.0% growth in YouTube. One segment contracted: Network. YouTube is scaling on CTV, experiencing revenue tailwinds from direct response CTV units and accelerated subscription growth. Meanwhile, Search is benefiting from a secular shift to an AI-infused consumer interaction model. Yet Network, Google's lowest-margin advertising business, is in systemic decline. Network is tethered to a surface area that is increasingly ceding consumer engagement to a host of other media and now represents less than 10% of the company's overall advertising revenue. Put another way: it is vestigial. mobiledevmemo.com/googles-ne…

Good morning, New York.

Agentic commerce is a mirage. mobiledevmemo.com/agentic-co…

"while Google was found guilty of monopolistic behavior with respect to its open web publisher and advertiser tools (and in tying DFP to AdX), it also likely kept the open web alive by preserving cookies." - @eric_seufert mobiledevmemo.com/googles-ne…

"Tweet often, but only about your winners" - Warren Buffet

Related: DAU forecasting using cohort matrices mobiledevmemo.com/modeling-d…

~10 years ago, there was a general inclination in the app ecosystem to build models for predicting user-level monetization and engagement. These predictions, often related to user LTV, would then be "rolled up" into aggregates for things like ROAS calculation. With the benefit of hindsight, this approach was mostly a dead end. The data that feeds these models has too much variance, and the decisions that these models are used to inform (marketing adjustments, product changes) generally aren't scoped to individual users, anyway. The modern approach that I see commonly implemented is to use cohort analytics for these kinds of decisions at the top of the funnel (marketing) and to intervene with specific users in ways that relate to group membership. This tends to smooth out the impact of outliers and results in less overfitting and more reliable results, although it also necessarily assumes that changes in product and marketing are moderate.

YouTube is the future of media.

The original podcast episode on the topic of Commerce at the Limit: open.spotify.com/episode/04M…

Last week, I gave a talk at Marketecture Live in New York titled “Commerce at the Limit: Frontier Applications of AI to Digital Advertising.” The presentation has two parts: The first is an overview of my concept of Commerce at the Limit, which argues that AI enablement in advertising aims to achieve two outcomes: - That every ad exposure generates value equivalent to its potential - That every business that could potentially derive value from digital advertising, does The second is a literature review of recent (published in the last ~12 months) papers on topics related to the application of AI / ML to digital advertising. Several people requested the slides, so I've made them publicly available. slideshare.net/slideshow/com…

vertical videos are eating all the attention