Thinking · Newsletter
What AppLovin's CEO says the channel rewards, and what it means for your creative
Adam Foroughi's October 2026 Northbeam interview, quoted and read for the people making the ads: the 30 to 60 second unit, frequency coming down, AI not ready for long clips, 60 percent of conversions within an hour.
TL;DR. In a closing-session interview with Northbeam published in October 2026, AppLovin's CEO Adam Foroughi described the ad unit his platform runs for commerce brands: a video, an end card ("an interactive page, an animated GIF effectively"), and a dynamic catalog ad, combined into one fullscreen experience that "on average the customer's watching for well over 30 seconds. They can't do anything else." He said the format "works really well if a customer invests time to make really good video content for our platform. It doesn't work as well if the customer doesn't actually make that investment." He also said video frequency on the platform should fall, that current video generation models cannot yet make a good 30 to 60 second clip, and that roughly 60 percent of transactions happen within an hour of the click. Here is what each of those means if you are the one making the ads.
Who said it, and where
Adam Foroughi, CEO of AppLovin, in conversation with Northbeam's Austin Harrison, published on Northbeam's channel as "Inside the Rise of AppLovin" in October 2026. The quotes below are from the recording. The interpretation is ours.
"They can't do anything else": the attention model
Foroughi described the unit in three parts: "There's a video, there's an end card that we call an interactive page, but an animated GIF effectively, that showcases some sort of sale, and then there's a dynamic catalog ad. All of that is combined into one full-screen ad experience, which on average, the customer's watching for well over 30 seconds. They can't do anything else."
Then the comparison that explains why Meta creative stalls here: "most of your videos and most of your experiences are 5 to 10 seconds long cuz that's all the time you have to hook a user on social, and in our case, they can be 30 to 60 seconds long."
What it means for the maker. The hook still matters, because the first seconds decide whether the viewer watches or waits. But the ad is no longer a hook with a tail; it is a 30 to 60 second argument with a hook on the front. A Meta cut engineered to compress is being asked to sustain, and it runs out of material. Our full read of the channel's creative rules is in the AppLovin creative guide.
"If the customer doesn't actually make that investment"
The most useful sentence in the interview is the honest one: "that works really well if a customer invests time to make really good video content for our platform. It doesn't work as well if the customer doesn't actually make that investment."
He followed it with the fix AppLovin is building: "how to bring a variety of templates to the market so that we can actually meet the needs of the customers." Templates lower the floor. They do not raise the ceiling. The brands that scale on the channel are the ones producing native 30 to 60 second video in volume, which is a production decision before it is a media one. That is why we describe AppLovin as a creative-supply channel, and why the first question to ask any partner is how many finished assets one shoot day yields.
"The fifth time, it's pretty boring": frequency is coming down
Asked where video is going, Foroughi said something few platform CEOs say about their own format: "when you serve users, you tend to find a user that's interested in a brand algorithmically and serve them videos multiple times. The first time, it's great... The fifth time, it's pretty boring. If you've seen 30-60 second clips for the same shop five times, you're going to be bored. So I think one thing that's going to probably change is frequency of video will go down at least on our platform. Dynamic catalog ads will grow."
What it means. The answer to frequency is variety, not one hero film. If the platform is going to show a brand fewer times, each showing has to be a different argument to a different motivation. That is the concept-system method: one idea built to vary across opening, performer, problem frame and offer, so the viewer's third exposure is a new case, not a rerun. And it is a reason to treat the end card and the catalog unit as creative, not afterthoughts, since they are the parts that will carry more of the load.
"Not at a point where you can create a great 30-60 second clip"
On AI: "The video generation models, at least that we've been trying to use to automatically create video, are not at a point where you can create a great 30-60 second clip. You can certainly create statics." He expects that to change over years.
This matches what we see in production. Generated video is good for environments, product shots and variant volume; a human face carrying a 40 second argument is still cast, not generated. Our position is in AI UGC: where it wins, where actors still do.
"Roughly 60 percent of all transactions happen within an hour of the last click"
On attribution: "We typically optimize to one or seven-day windows... roughly 60% of all the transactions happen within an hour of the last click. So you're getting a user who's in a game. They click, they go complete the shopper behavior, and then they go back to the game." And on what the platform cannot price: "the brand aspect of the video ad that has to be watched for 30 to 60 seconds, we're unable to capture... we price that for free."
What it means. Two things. The page the click lands on has to close inside that hour, which is the ad-to-landing-page handoff problem: a specific ad pointed at a generic product page wastes the intent the video built. See landing page vs product page for paid ads. And the longer-term lift the video creates is yours to measure, which is what a holdout is for; the platform will not report it. See incrementality testing for ad creative.
The advice he gave hesitant brands
"Everything can be plug and play. If someone has videos that are short, they're not going to get the same performance, but it doesn't mean the performance is not going to work." And on why to test at all: "your next hundred thousand dollars you spend on that one platform will have much worse ROAS than the average ROAS that you got on the platform. The less budget you have, the higher the ROAS, the more you push budget, the lower the ROAS. So on the incremental dollar," a second channel earns its place.
Our reading of all of it: start with what you have, measure it against native creative in a bake-off, and treat the result as a production brief, not a verdict on the channel.
Questions people ask
What ad format does AppLovin use for ecommerce brands? Per Foroughi: a video, an end card ("interactive page") showing an offer, and a dynamic catalog ad, combined into one fullscreen unit watched for over 30 seconds on average.
How long should AppLovin video ads be? 30 to 60 seconds, by the CEO's own description of the format; social-length cuts of 5 to 10 seconds "are not going to get the same performance."
Does AppLovin work with short Meta videos? It can run them; Foroughi says the performance is lower, not zero. Native long cuts are the investment the channel rewards.
How fast do AppLovin conversions happen? Foroughi cited roughly 60 percent of transactions within an hour of the last click, which is why the landing page has to close quickly.
Will AI make the videos? Not yet for 30 to 60 second clips, by his account; statics yes.
Source: Northbeam, "Inside The Rise of AppLovin: Adam Foroughi", YouTube, October 2026. Quotes transcribed from the recording; emphasis and interpretation are Flixr's.