Meta

Meta's auction adds organic value to the bid so ads and posts compete together. Peter Sellis credits Meta's 2010 to 2020 ads team with solving the modern auction: each bid factors in the organic value to the user, so advertisers pay more to reach people who may not want to be reached.

He suggests that in chat products like ChatGPT, trust could replace organic value as that term, and that maximizing trust likely correlates with retention and advertiser value. He points to Amazon search showing 16 ads before the first result as eroding trust. Quote: "how influenced I should be should essentially be how much trust I'm trading off."

Lenny's Podcast▶ 50:50

After Andromeda, better creative beats tinkering with targeting in Ads Manager. Since Meta's Andromeda update, much of the targeting happens through the creatives themselves. One subscription app team spends less time on bids and targeting settings, which are largely out of their control, and more on making better content.

They feed their highest-intent users back to Meta and use a custom dashboard to check which trials truly come from Meta. Nick Lawton calls Ads Manager "a big scary monster." Quote: "a lot of the targeting on meta is specifically done through the creatives."

Acquired▶ 4:33:54

Click-bait ads bring cheap trials that rarely convert; target real pain points. Ethan Ethier's team started Meta ads in January 2026 and got a baseline trial-to-paid in about a month to set allowable acquisition cost. Click-bait ads produced cheap trials but trial-to-paid dropped sharply, so they switched to ads built on customer avatars and the pain points the app solves.

Avatars came from customer surveys and warm traffic, and they kept iterating on winners. Meta trial-to-paid sits around 25%, below their YouTube traffic. Quote: "we want people to start the trial at the cheapest cost, but also we don't want just anyone to start the trial."

Acquired▶ 2:18:19

Build Meta ads that don't depend on the founder-influencer. Ethan Ethier's team runs Meta ads with B-roll, testimonials and AI-made statics instead of Jeremy, to solve "the bus problem" of all growth relying on his YouTube channel.

Jeremy's time stays on long-form YouTube, which they judge the highest-ROI activity, while clippers post on Reels and TikTok and their viral hooks inform future videos.

Acquired▶ 4:08:07

Manage Meta in-house so ads don't over-promise the product. The same team chose in-house Meta management over an agency to get someone passionate about the product who would "promote the app in a way that isn't faking anything."

The goal is a seamless handoff from the ad to the product experience.

Acquired▶ 49:50

Fast Meta growth can bring in the wrong customers. Pok Pok, whose main paid channel is Meta and especially Instagram, learned that it can grow fast there while acquiring worse-fit customers. Melissa Cash calls Meta "the golden handcuffs."

Pok Pok now invests in brand, partnerships such as a collab with Calm, and a brand department started about six months before the interview to bring in more organic and better customers.

Lenny's Podcast▶ 17:29

Run winning organic creator videos as Meta ads, not scripted copies. Cedric Roberge takes videos that already won on TikTok and Instagram, runs them on Meta and raises budget weekly on winners, at about $2.50 per download on roughly $400 a day. Scripts given to creators to mimic other GLP apps failed and lost money.

Pok Pok does the same with its paid creator program: content that pops off is then promoted with ads, which is how it adapted to weaker classic influencer marketing. Cedric's rule of thumb is that under $5 per download is good.

Lenny's Podcast▶ 9:35

After Andromeda, better creative beats tinkering with targeting in Ads Manager. Since Meta's Andromeda update, much of the targeting happens through the creatives themselves. One subscription app team spends less time on bids and targeting settings, which are largely out of their control, and more on making better content.

They feed their highest-intent users back to Meta and use a custom dashboard to check which trials truly come from Meta. Nick Lawton calls Ads Manager "a big scary monster." Quote: "a lot of the targeting on meta is specifically done through the creatives."

Lenny's Podcast

Use Meta for conversions and TikTok Spark ads for cheap reach. Spark ads push organic videos into the For You page but have no button, so only CPM is attributable. Meta ads have a button and track conversions, so Nick Lawton sees real scale and attribution coming from Meta.

Spark ads suit videos that already carry a strong organic call to action.

Lenny's Podcast▶ 57:50

Meta replaced 13-person teams with pods of four to six engineers. Teams used to include 2 to 3 each of Android, iOS and server engineers plus a PM, designer, data scientist and researcher. Now pods run with four to six generalist engineers and a product staff lead, so core teams are usually six or seven people.

Specialists are added only when needed, such as a senior data scientist for pricing or a senior designer for a novel experience. Adam Mosseri says fewer people means faster moves and "a little bit less design by committee."

App Masters▶ 2:08

The product staff role blends PM, design, data and research using AI tools. Meta evolved the PM into a generalist who does part of each specialist's job with internal AI tools. A Reels creation drop-off waterfall that needed a data scientist a year earlier can now be pulled by product staff directly.

Pods can run with four engineers and one product staff and no dedicated designer, data scientist or researcher.

App Masters▶ 3:33

Keep hiring juniors in every function to grow future seniors. Mosseri warns against staffing functions only with senior people, since nobody would be left to become the next senior specialist.

Shrink teams if needed, and move people not on a path to senior into generalist roles as a soft landing.

App Masters▶ 4:55

Treat AI token spend as a budget and skip token leaderboards. Meta has no token limits yet but manages spend by shutting down silly projects. Mosseri calls token leaderboards a terrible idea because "it's not that hard to build a token incinerator."

He expects caps if a strong engineer's burn approaches their salary, sized to trust in ROI-positive use, as with GPUs, labeling budgets and headcount.

App Masters

Odd, cluttered static ads can beat polished ones, so test them. Cedric Roberge's ads person made a crowded square static listing about eight pieces of information in about 2 minutes with ChatGPT. It became the best performing ad at under $2 per install, against a $2.50 average.

Cedric thought it was "way too cluttered" but now makes more variants, including cleaner ones aimed at men or women. The lesson he drew: let teammates try ideas you disagree with.

Sub Club▶ 21:30

Run winning organic creator videos as Meta ads, not scripted copies. Cedric Roberge takes videos that already won on TikTok and Instagram, runs them on Meta and raises budget weekly on winners, at about $2.50 per download on roughly $400 a day. Scripts given to creators to mimic other GLP apps failed and lost money.

Pok Pok does the same with its paid creator program: content that pops off is then promoted with ads, which is how it adapted to weaker classic influencer marketing. Cedric's rule of thumb is that under $5 per download is good.

Sub Club▶ 17:22

Meta tolerates some categories TikTok rejects, like peptide content. TikTok keeps rejecting Cedric Roberge's Spark ads and removing videos because peptide content reads as medical advice. Meta is more lenient and performs better, and rival GLP apps in the Health and Fitness top 100 run only Meta ads.

His app ranks around 250 and aims for the top 100. He also spends about $150 a day on Apple Search Ads.

Sub Club▶ 46:30

Message, media, content type and funnel together decide who Meta reaches. Marcus Burke's formula is "message plus media plus content type times funnel equals audience." Video skews to a younger Instagram Reels audience, statics to Facebook feed and older users, and street interview styles skew younger than raw text walls.

To scale, add new pain-point messages, new formats and media types, or new funnels (web or app store), then check where ads were delivered to learn from it.

David Senra▶ 28:00

Compare cost per trial with trial-to-paid by age bucket. Marcus Burke collects age during onboarding to see trial-to-paid by age, then maps it against cost per target. Young traffic buys cheap trials that convert worst, while someone 65 plus might cost 100% more per trial but convert much better.

In his example chart, 25 to 34 was the most expensive bucket. The point is not to let Meta simply buy the cheapest trials.

David Senra▶ 28:37

Use value rules to make Meta bid more for older users. Keep targeting broad and set value rules telling Meta that someone 45 plus is, for example, 30 to 40% more valuable. This works around trial-to-paid landing about 7 days later, too late for the algorithm to react.

Also make ads that appeal to older people, since low spend on them often comes from not making content for them.

David Senra▶ 31:55

With little budget, put every ad in one ad set. Multiple ad sets would require doubling or quadrupling budget. Marcus Burke advises putting everything in one, seeing what wins, repeating weekly and learning patterns step by step.

Split by message and media only once ads are near break-even or you have funding to speed up learning.

David Senra▶ 3:59

To scale, split ad sets by format and message with their own budgets. Left alone, Meta pours budget into one best ad or message, concentrating about 80% of spend on its favorite within an ad set. Burke clusters creatives by media type, style (text walls, memes, UGC video) and message (overthinking, anxious attachment), with budgets and cost-per-target goals per ad set, for example five ad sets.

Meme ad sets must deliver traffic at about half the price because their quality is lower. Quote: "I'm not looking for the very best ad and only like serve to that audience. I'm looking to scale."

David Senra▶ 1:45

Give outlier ads their own ad set so they get delivery. Chart each ad's placement mix against cost per target. Ads with good cost per target that spent 40 to 50% of budget on Facebook feed got starved when mixed with others, so Burke pulls them into a separate ad set with its own budget.

Keep broad targeting and all placements, and compare them with the other ads to learn why they deliver differently.

David Senra▶ 16:57

New managers should spend their first stretch as ICs. Fiona Fung spent her first quarter at Meta as an IC after 11 years at Microsoft, before leading Marketplace, smart glasses and the Orion AR effort. Teammates said they appreciated her using what they built.

She has managers keep making PRs part time to stay close to the product. Quote: "give yourself that maker time to actually type deep into the code."

Subversive▶ 18:50

Test creatives at high volume, because about 1 in 85 wins. Alex's app found that one video out of 85 would take most of the ad spend and work. Of roughly $1.5 to 1.8 million in total spend, something like $800k was wasted, while the winners paid back hundreds of thousands in profit.

Much of the waste was the cost of learning which kind of video Meta favored, so the whole operation was planned around producing enough ads to find winners often enough. Losers are the price of finding the winner.

Superwall▶ 13:44

A winning ad needs both Meta's prediction and the audience to favor it. A technically excellent creative can still fail. One that looked great took $20K of spend and did not work however the team sliced it.

Alex's rule: "It has to be both a Meta match and an audience match." Do your best on quality and accept that the platform decides the rest.

Superwall▶ 28:47

Expect Meta account bans and keep appealing until you recover. Alex's accounts were restricted or banned six to eight times, some permanently, even without breaking policy, and they always got them back.

Early on, support is offshore and can't fix much, so appeal persistently and explain why you should be allowed to advertise. At higher spend, a Meta rep and a warmed-up account help. Quote: "a lot of people would quit the first or the second time."

Superwall▶ 26:11

A business built only on Meta ads should diversify or sell. Bans or CPMs doubling can wreck a paid-only business whose profit rests on a "loaded-dice game of statistics." Alex sold Boost App Social for seven figures in January after a four to five month negotiation, partly to take chips off the table.

He now mixes in organic growth at Screens.design.

Superwall▶ 35:25

Web funnels let Meta learn faster than app installs do. Rise Science runs web onboarding, which ad platforms treat as standard e-commerce with no delayed data or broken attribution from app privacy limits. The algorithm finds people faster and creative results show up sooner, while on app you may wait a week or two depending on budget.

Rise's older web audience tilts toward Facebook over Instagram, though the channels perform about the same and creative decides results. Leon Sasson calls the faster iteration cycle "a huge unlock."

App Masters▶ 18:12

Meta's large AI push gained little productivity compared with Anthropic. Boris Cherny, once responsible for code quality across Facebook, Instagram and WhatsApp, says a year of work by hundreds of Meta engineers gave a few percentage points of productivity, while Anthropic saw a 200% rise in pull requests per engineer.

His takeaway is to under-resource projects a little at the start, for example one engineer per project, so people lean on AI to automate.

Lenny's Podcast▶ 25:30

Prepare for the Meta PM interview with an AI coach and human mocks. Zevi Arnovitz built a Claude project coach fed with frameworks and told it not to flatter him, had Claude play the candidate for questions he couldn't mock, and used an agent to rank the most frequently asked questions.

He says human mocks, found through cold LinkedIn outreach, were "the biggest game changer." He got the job, and engineers there now ask him to teach his no-code workflow.

App Masters