Professional Meta Advertising Company

What Modern Meta Advertising Actually Looks Like

Meta is still the largest paid social platform on earth. The Family of Apps (Facebook, Instagram, Messenger, WhatsApp, Threads) reaches over 3 billion daily active users. Meta’s advertising revenue grew 21% year over year in Q4 2025 to a record quarterly figure, and the platform’s average ad price keeps climbing. CPMs sit in the $8 to $14 range across most categories. Average CPCs run between $0.97 and $1.50. The platform that everyone keeps writing the eulogy for is, in fact, more profitable than it has ever been.

What has changed is how the platform actually works. The Meta of 2018, where a media buyer could win by stacking detailed interest layers and lookalikes on top of well-segmented retargeting funnels, no longer exists. Meta replaced its core ad delivery system in late 2024 with a new algorithm called Andromeda, retired most of the old detailed-targeting playbook in favor of Advantage+ Audience and broad targeting, deprecated dozens of legacy interest categories, and shifted ad delivery from being driven by audience signals to being driven by creative signals. The platform now reads your ad to decide who should see it, rather than asking you to tell it who to send the ad to.

The accounts winning right now are the ones that have rebuilt their playbooks around that shift. The accounts losing are the ones still running 2019 campaign architectures with 2024 creative against 2026 algorithms. Most agency-managed accounts we audit fall into the second group. The opportunity for any business willing to update the playbook is significant, because the gap between accounts that have adapted and accounts that haven’t is the largest it has been in the platform’s history.

This page exists to be honest about what Meta is now, what it requires, and what we actually do for clients who hire us to lead their paid social strategy.

Andromeda Changed Everything

In late 2024, Meta replaced its underlying ad delivery infrastructure with a new system called Andromeda, built in partnership with NVIDIA and running on the GH200 Grace Hopper Superchip. The technical specifications are worth understanding because they explain the entire shift in what wins on the platform.

Andromeda is roughly 100 times faster at matching users to ads than the system it replaced, and it can hold and evaluate roughly 10,000 times more ad variants in parallel during a single auction. The retrieval model uses deep learning and sequence learning to predict which user, in which moment, in which feed context, is most likely to take the action you’re optimizing for. The matching is no longer driven primarily by who you targeted. It’s driven by what your ad looks like and what the algorithm has learned about which kinds of creative resonate with which kinds of users in which kinds of contexts.

This has three consequences that have rewritten the playbook for the entire platform.

First, creative is now the primary targeting signal. Meta’s Andromeda model reads your ad assets, the imagery, the captions, the hooks, the music, the pacing, the on-screen text, the product context, and uses that read to decide who should see the ad. The advertiser still sets exclusions and budget and bidding, but the audience finding is now substantially algorithmic. Detailed interest layering as a primary targeting strategy is dead. Broad targeting plus diverse, well-built creative is now what wins. Industry voices like Andrew Foxwell at Foxwell Digital, who audits 20 to 30 accounts a month and whose Foxwell Founders community spends over $500 million per month on Meta, have been documenting this shift extensively through 2025 and 2026. Foxwell Digital blog

Second, creative variety beats creative volume. For years, the playbook for testing was to take a single winning image and run twenty variations with different headlines, different hooks, different CTAs. Andromeda interprets minor variations of the same visual asset as essentially the same ad. The system now wants real variety, distinctly different angles, distinctly different formats, distinctly different storytelling approaches, not iteration on a single creative. Brands that try to feed the algorithm 30 versions of the same ad with different copy get treated as if they uploaded one ad. Brands that feed it 30 genuinely different creative concepts give the algorithm something to actually optimize against.

Third, simpler campaign architecture outperforms complex architecture. Many advertisers running on Andromeda have collapsed their account structures from dozens of campaigns segmented by audience temperature into one or two main campaigns. The algorithm is sophisticated enough to recognize which users need cold prospecting messaging and which need retargeting, and to serve each appropriately within a single campaign. The old architecture, three separate campaigns for cold audiences, warm audiences, and existing customers, often actively underperforms the new architecture because each segmented campaign starves the algorithm of the volume of conversion data it needs to optimize properly.

This is why the accounts running on 2019 playbooks are losing ground every month. The platform they’re optimizing for no longer exists. The accounts that have adapted are scaling efficiently because they’re working with the algorithm rather than against it.

---

Creative Is the Strategy Now

The hardest concept for most marketing teams to internalize about modern Meta is that the strategic layer of the work has moved. It used to live in account architecture, audience segmentation, bidding strategy, and budget allocation. Those things still matter, but they no longer drive performance the way they did. The strategic layer of Meta now lives in the creative itself.

Andrew Foxwell makes this argument repeatedly in his work, and the framing he uses is useful. The ad is no longer the deliverable. The ad is the strategy. Every choice about which angle to lead with, which awareness state the ad addresses, which persona it speaks to, which objection it answers, which format and pacing it uses, is now a strategic decision that determines who sees the ad and how the algorithm decides to spend the budget. Briefing a designer to “make us a Facebook ad” no longer produces results, because the brief itself has skipped the actual strategy.

What this means in practice is that the work of running Meta successfully has shifted from media buying to creative strategy. The media buyer is still important, but the creative strategist, the person deciding what concepts to test, what messages to lead with, what awareness states to target, what kinds of creative to produce, has become the most leveraged role on the team. The accounts that scale predictably are the accounts where someone is doing this work seriously every single week, with a structured testing framework, a defined set of angles to explore, and a clear feedback loop between creative output and account performance.

The volume question is downstream of this. The right answer to “how many ads should we run per month” is “as many as the strategy calls for, with real variety, briefed by someone who actually understands the platform.” Top-performing brands in the Foxwell Founders community, who collectively spend over $500 million per month on Meta, typically ship 20 to 50 genuinely distinct creative concepts per month per brand, sometimes more. Most agency-managed accounts ship 2 to 5. The gap is not because the top brands have bigger budgets. It’s because they understand that creative variety is the lever that actually compounds performance under Andromeda, and they’ve built production capacity to match.

This is the part of Meta that most agencies are not built to deliver on. Running ads is one discipline. Producing 30 distinctly different creative concepts every month, briefing them strategically, getting them through compliance, and feeding them into the account at the cadence the platform needs is an operational discipline most agencies do not have. They subcontract it, lose weeks in back-and-forth, and end up shipping less than what the algorithm needs.

The accounts that win are run by teams who treat creative strategy as the central operational discipline of the program. The ads agency relationship of 2018, where the agency runs the account and the brand handles the creative, is no longer viable for most businesses. Either the agency owns the creative strategy at the volume and quality the platform requires, or the brand has to. Outsourcing the strategic thinking and hoping for the best is the most reliable way to produce a Meta account that quietly bleeds budget.

---

The “Ugly Ads Always Win” Argument and Why It’s Oversimplified

A few years ago, Barry Hott, a well-known performance creative practitioner who’s been running Meta ads at scale for over a decade, started publicly arguing that “ugly ads” outperform polished ones. The thesis was that overproduced, agency-style brand creative was getting smoked on Meta by what looked like organic, native, scrappy content. He famously demonstrated the point by running ads built from nothing more than handwritten Post-it notes and watching them outperform fully produced video creative from the same brand.

The argument was right, and it was an important corrective to a marketing culture that had spent years pouring budget into beautiful brand creative that performed badly on the platform. The reason ugly ads work is that they look like content, not advertising, and Meta’s algorithm rewards content that holds attention in a feed full of other content. The user scrolling Instagram is not in a shopping mindset. They’re entertaining themselves, catching up with friends, looking at things their cousin posted. An ad that looks like an ad gets scrolled past. An ad that looks like a real person talking about a real problem gets watched.

The problem is that the framing got oversimplified by the time it filtered through the broader marketing community. “Ugly ads always win” turned into a license for agencies to ship low-effort UGC-style content and call it strategy. It also turned into a comfortable excuse to stop investing in production quality entirely. Brands started shipping single-take iPhone videos with bad audio and worse hooks, and when those ads underperformed, the explanation was usually “Meta doesn’t like polished creative anymore,” which was the wrong takeaway from the original argument.

Andrew Foxwell has been pushing back on the oversimplification recently, and his framing is more useful. The right approach is not “always ship ugly ads” but rather build a suite of ads that covers a real range. That suite includes founder-led content, where a real person from the brand talks on camera about why the product exists or why it matters, often shot on a phone in a natural environment. It includes polished brand creative for higher-AOV product categories where the production value itself signals legitimacy. It includes objection-handling ads that address the specific concerns prospects are raising on sales calls, in reviews, or in support tickets. It includes persona-specific ads that speak to different customer archetypes in their own language. And in 2026, it increasingly includes AI-generated creative, sourced from tools that scrape ad comments and product reviews to surface real customer language and produce variants automatically, typically running at 10 to 15% of the creative mix.

The reason this matters for performance is that different creative styles unlock different kinds of buyers. The founder-on-iPhone ad converts the buyer who values authenticity and is suspicious of polished marketing. The polished brand video converts the buyer who values quality signals and is suspicious of scrappy production. The objection-handler converts the buyer who has a specific concern preventing purchase. The persona ad converts the buyer who needs to see themselves reflected in the marketing before they trust the brand. None of these formats replace the others. The accounts that scale are the accounts running all of them in parallel, watching what the algorithm rewards, and feeding more of what wins.

The takeaway for any business spending real money on Meta is that “creative quality” is no longer one variable to optimize. It’s a portfolio decision. The question is not “should our ads be polished or rough” but “what mix of polished, rough, founder-led, persona-driven, and objection-handling creative does our audience actually need in order to convert.”

---

The Funnel Has Collapsed

The other 2018 playbook still being run by most agencies is the three-tier funnel. Top of funnel campaigns target cold audiences with awareness messaging. Middle of funnel campaigns target engaged users with consideration messaging. Bottom of funnel campaigns retarget high-intent users with conversion messaging. Each tier has its own campaigns, its own creative, its own budgets, its own KPIs. The advertiser maintains tight control over which messaging hits which audience temperature, and the funnel is supposed to walk buyers through a defined sequence of touchpoints from cold to converted.

The architecture made sense when ad delivery was driven by audience signals and when conversion windows were long enough to track buyers across multiple touches. Both of those conditions broke down years ago. Andromeda now decides on its own which user is cold, which is warm, and which is ready to buy, and it serves the appropriate creative from a single campaign rather than relying on the advertiser to segment manually. iOS 14 and the broader signal collapse made multi-touch attribution unreliable enough that the funnel sequence the advertiser thought they were building was often not actually happening in any traceable way.

What’s replacing the three-tier funnel is what some practitioners are calling collapsed-funnel architecture. The structure varies, but the pattern is consistent: one or two main campaigns running broad targeting with diverse creative, the algorithm handling segmentation internally, the advertiser focusing budget and attention on creative variety rather than audience structure. Many top-performing brands have stopped running separate retargeting campaigns altogether. The algorithm is sophisticated enough to recognize when a user has been to the site, when a user has engaged with previous ads, and when a user is seeing the brand for the first time, and to serve appropriate creative within a unified campaign.

This does not mean structure is dead. It means the work of structuring has shifted from audience segmentation to creative segmentation. Instead of building three campaigns for three audience temperatures, the modern approach builds one campaign with a creative library that addresses different awareness states inside the same campaign. Foxwell’s framework of brand-aware, product-aware, problem-aware, and unaware creative is doing the segmentation work that audience-temperature funnels used to do, but the segmentation is happening at the creative level, with the algorithm matching the right ad to the right user automatically.

The advertisers who are still running rigid three-tier funnels are usually doing so because the architecture feels like control. Three separate campaigns with three separate budgets feels more deliberate than one campaign with broad targeting. The problem is that the architecture is now actively hurting performance in most accounts, because each segmented campaign is starved of the conversion volume the algorithm needs to optimize, and the creative variety inside each campaign is necessarily smaller than what the platform now rewards. The control feels good. The performance gets worse.

There are still cases where audience segmentation matters, particularly for retention campaigns, customer-list excludes on prospecting, and certain considered-purchase B2B contexts where the buyer journey genuinely is multi-touch and trackable through CRM data. But the default architecture for most consumer brands and most direct-response B2B accounts has flipped. Simpler is better. Trust the algorithm to do the segmentation. Spend the time on creative.

---

iOS 14 and the Honest Measurement Stack

In April 2021, Apple released iOS 14.5 with App Tracking Transparency, requiring users to opt in to third-party tracking on a per-app basis. Roughly 75% of users opted out. Meta’s measurement infrastructure, which had been built on browser-side pixel tracking with long attribution windows, broke. Reported conversions on Meta dropped overnight by 30% in many accounts. Last-click attribution stopped working as a reliable signal. Click-through windows compressed. The reporting numbers Meta was showing advertisers stopped matching what was happening in the actual business.

Most agencies pretended this didn’t happen, and many still are. The pixel still fires. The Meta Ads Manager dashboard still shows numbers. The numbers still go up and to the right when ads do well. So agencies kept reporting them, even though every serious practitioner knows the underlying signal is incomplete and increasingly disconnected from real revenue.

The honest measurement stack in 2026 looks completely different from the 2020 version, and the gap between accounts running modern measurement and accounts running pixel-only is one of the biggest performance gaps on the platform. Here’s what actually matters now.

Conversions API (CAPI). Server-side conversion tracking that sends data directly from your server to Meta, bypassing the browser entirely. CAPI restored a meaningful portion of the signal lost to iOS 14, particularly for purchase events where the conversion fires on the server after a successful transaction. Every serious Meta account in 2026 runs CAPI. Accounts running pixel-only are operating with one eye closed. CAPI is not optional. It is the floor, not the ceiling.

Server-side tagging via Google Tag Manager Server or equivalent. A layer above CAPI that gives advertisers more control over what data is sent, how it’s matched, and what privacy controls are applied. Server-side tagging also enables enhanced match keys (email, phone, customer ID) to flow into Meta in a privacy-compliant way, which materially improves both attribution and audience matching. Most accounts that have invested in server-side infrastructure have seen reported conversions stabilize or recover, and have seen Advantage+ campaigns perform meaningfully better because the algorithm has more conversion data to learn from.

Meta’s Customer Lifecycle Strategy and the new attribution model. Meta released a new attribution model in early 2026 that separates click-driven conversions from engagement-driven conversions, and a Customer Lifecycle Strategy feature that lets advertisers control whether campaigns optimize toward acquisition or retention. Foxwell wrote about both extensively. The implication is that reported conversions may drop in some accounts as the model gets stricter about what counts as click-driven, but the visibility into what’s actually driving the reported numbers improves significantly. Advertisers who understand both signals can manage to a clearer picture of true performance.

Third-party attribution platforms. Northbeam, Triple Whale, Rockerbox, and similar tools have become standard infrastructure for serious DTC and ecommerce accounts. Northbeam’s Clicks + Deterministic Views (C+DV) framework, which counts both ad clicks and verified ad views into the attribution model, has been widely adopted as a more honest replacement for last-click. The framework recognizes that on Meta, the view often does the conversion work, even when the eventual click happens on a different channel. [link: Northbeam C+DV documentation]

Incrementality testing. The most rigorous answer to the measurement problem, and the one most agencies skip. Geo-split tests, holdout audiences, and lift studies measure the actual incremental impact of Meta spend by comparing exposed populations against unexposed control populations. The results are sometimes uncomfortable, because they reveal that some accounts are getting credit for conversions that would have happened anyway. They’re also the only true measurement of what Meta is actually producing. Foxwell has been writing about incrementality as the discipline that separates serious agencies from agencies who report whatever the platform reports back. Foxwell Digital writing on incrementality

The reason this matters for any business spending real money on Meta is that most performance debates between agencies and clients are debates about reported numbers that nobody fully trusts. The agency points at the dashboard. The CFO points at revenue. The two don’t reconcile, and the disconnect erodes trust on both sides. The accounts that scale predictably are the accounts where measurement infrastructure has been rebuilt to produce numbers both sides can trust, and where the conversation moves from “what does Meta report” to “what is Meta actually contributing to the business.”

That work is not glamorous. It involves CAPI implementation, server-side tagging setup, third-party attribution onboarding, and the discipline of running incrementality tests on a regular cadence. It also takes weeks of upfront investment before the new measurement is producing useful data. Most agencies skip it because the work is expensive and the client doesn’t always know to ask for it. We start with it because every other decision downstream depends on the data being trustworthy.

---

AI in Meta

Meta is now an AI-first advertising platform. Andromeda is an AI system. Advantage+ campaigns are AI systems. The bidding, the budget allocation, the audience matching, the creative optimization, the placement decisions, all of it is now driven by machine learning models running at speeds and scales that no human media buyer can replicate. Saying “I’m doing Meta ads” in 2026 is in large part saying “I’m directing AI systems to allocate budget on my behalf.” The advertiser’s job is no longer to make the decisions the algorithm makes. It’s to give the algorithm the inputs it needs to make good decisions, and to verify that the decisions it’s making actually serve the business.

There are three layers of AI inside the Meta ecosystem worth understanding, because they affect the work in different ways and require different responses from the advertiser.

The first layer is the algorithmic infrastructure of the platform itself. Andromeda for ad delivery, GEM (Generalized Embedding Model) for content understanding, the ranking systems that decide which ad wins each auction, the optimization systems that decide where to spend each dollar of budget. This layer is not optional and not configurable. It runs every account on the platform whether the advertiser likes it or not. The question for the advertiser is whether you’re feeding the algorithm in a way that lets it work for you, or starving it in ways that force it to work poorly. Diverse creative, clean conversion signal, sufficient budget for the algorithm to exit learning phase, and broad targeting that doesn’t artificially constrain delivery are the inputs that let this layer perform well. Hyper-targeted audiences, a handful of similar creative variants, broken pixel data, and budgets that keep campaigns stuck in learning are the inputs that produce poor performance regardless of how much manual optimization the advertiser does on top.

The second layer is the AI features Meta surfaces to advertisers as optional tools. Advantage+ Shopping Campaigns, Advantage+ Audience, Advantage+ Creative, generative AI for image variations and background generation, automated copy generation, dynamic creative optimization, automatic placements. These features have improved meaningfully over the last two years. Most are now genuinely useful for most accounts, and ignoring them has become more costly than using them. The honest framing on these tools is that they’re not magic, and they’re not always right, but the cost of testing them against your existing setup is low and the upside in many accounts is significant. The accounts that win are running structured tests on Advantage+ features against control campaigns, keeping what wins, and dropping what doesn’t. The accounts that lose are either ignoring the features entirely on principle, or accepting them blindly because Meta recommended them.

The third layer is third-party AI tools that have emerged outside Meta. Tools that scrape ad comments, product reviews, and Reddit threads to surface real customer language, then generate ad variants automatically. Creative analytics platforms like Motion that score creative performance and surface patterns across thousands of ads. Brief-generation tools that translate sales call transcripts into ad concepts. Image and video generation tools that produce variants at speeds and costs that would be impossible with traditional production. These tools are reshaping what creative production capacity even means for a Meta program. A team of three people with the right AI stack can now produce more usable creative variants per week than a team of ten could two years ago.

The implication for any business spending real money on Meta is that AI is not an optimization layer on top of the work. It’s the work. The accounts that win are run by teams who think about AI as a co-worker that handles the parts of the job humans are bad at, freeing the humans to focus on the parts of the job humans are still better at. The algorithm is good at finding the right user for an ad. It’s bad at deciding what angle the brand should lead with this quarter. The algorithm is good at optimizing budget allocation across thousands of micro-decisions per day. It’s bad at recognizing when the entire creative library has gone stale and the brand needs a new angle. The human role is upstream of the algorithm, not in competition with it.

This is also why the agency model is changing. The traditional agency value proposition was expertise in account management, audience targeting, and bidding strategy, all of which the algorithm has progressively automated. The new agency value proposition is creative strategy, production capacity, measurement infrastructure, and the judgment to direct AI tools effectively. Agencies that haven’t made the shift are running 2018 playbooks on 2026 platforms and producing 2018 results. Agencies that have made the shift are scaling accounts that look impossible from outside the platform.

We have made the shift. Our team uses AI throughout the creative and analytics workflow, but the strategy, the angles, the awareness-state framework, the creative briefs, and the judgment about what to test and when stay with senior strategists. AI accelerates the work. It doesn’t replace the work.

---

The Future of Meta Advertising

Predicting where Meta is heading is genuinely hard right now, because the platform is changing faster than at any point in its history. Andromeda was the largest under-the-hood change to ad delivery in over a decade, and it shipped less than two years ago. The AI layer the platform sits on continues to evolve quarter by quarter. The competitive landscape from TikTok, YouTube Shorts, and emerging platforms keeps shifting where attention sits. Privacy regulation continues to compress the signal data the platform can use. Any prediction longer than 18 months out has to come with humility attached.

That said, several patterns are clear enough to act on.

The creative discovery engine framing will keep deepening. Meta has effectively become a content distribution platform that monetizes through advertising. The line between organic content and paid content has blurred to the point where the highest-performing ads look indistinguishable from the highest-performing organic posts. The accounts that win in the next 24 months will be the ones that treat their Meta presence as a content operation first and an ad account second. That means a real content production cadence, real editorial judgment about what to publish, real understanding of what your audience actually wants to see. Brands without a content discipline will keep struggling on Meta no matter how good their media buying is. Brands with a strong content discipline will scale on Meta even with mediocre media buying, because the platform now rewards content quality more than account-management cleverness.

Brand and direct-response are converging. For most of Meta’s history, the platform was treated as a direct-response channel. Brand campaigns ran on TV, on YouTube, in print. DR campaigns ran on Meta. The two operations were managed separately, often by separate teams with separate budgets and separate KPIs. That separation is breaking down because Meta is now a meaningful brand-building channel for any consumer or considered-purchase business, and the same creative library that drives short-term DR also drives long-term brand recognition. The advertisers winning are the ones who have integrated brand and DR into a single operation, with creative that does both jobs simultaneously. The advertisers losing are still running brand campaigns to TV and DR campaigns to Meta and wondering why the channels feel like they’re competing.

Attribution will keep getting harder before it gets easier. Privacy regulation, browser-level signal restrictions, and platform-level walled gardens are all pushing in the same direction: less data, more uncertainty, more reliance on modeled and probabilistic measurement. The accounts that scale will be the ones that have built incrementality testing as a regular discipline, that have invested in modeled-attribution platforms like Northbeam or Triple Whale, and that have learned to make decisions on a measurement stack that includes some uncertainty rather than waiting for the certainty that’s not coming back. The accounts that struggle will be the ones still chasing last-click attribution and arguing about what Meta reports versus what the CRM shows.

The ecosystem play matters more than any single channel. Meta is no longer a standalone advertising decision. The platform performs best when it sits inside a connected media operation. The pattern that’s emerging across high-performing brands looks something like this: Meta drives efficient acquisition and audience-building, YouTube and YouTube Shorts scale the awareness layer for buyers in considered categories, TikTok handles discovery for younger and trend-driven audiences, email and SMS lock in retention and lifetime value, and the connective tissue between these channels is the brand work that makes each one perform better than it would in isolation. Treating Meta as a standalone channel and judging it on standalone ROAS misses how the channel actually contributes to business outcomes when it’s run as part of a larger system.

The agency model will keep getting smaller and more specialized. The Foxwell Founders 2026 State of Digital Marketing Agencies Report documented what the industry has been feeling for two years: AI is compressing the value of generalist execution work, and the agencies that are growing are the ones with deep specialization, strong creative production, and senior-led engagements. Big-roster agencies running mid-tier Meta accounts on autopilot are being squeezed from both sides. AI is taking the bottom of the value chain. Specialized senior-led agencies are taking the top. The middle is hollowing out. [link: Foxwell Founders 2026 State of Digital Marketing Agencies Report] This is a structural shift, not a cyclical one. The businesses making smart Meta investments in the next 24 months will be the ones working with senior strategists who own outcomes, not large teams of junior account managers running standardized playbooks.

AI-generated creative will become a standard line item, not a novelty. The 10 to 15% of the creative mix that’s now AI-generated in top-performing accounts will likely climb to 30 or 40% over the next 24 months as the tools improve. That doesn’t mean human creative is going away. It means the highest-leverage human creative work will shift toward strategy, judgment, and the kinds of authentic founder and customer storytelling that AI cannot replicate. The accounts that win will have a clear point of view on which parts of the creative library should be AI-generated and which parts must remain human, and they’ll have built workflows that produce both efficiently. Brands that resist AI creative entirely will lose ground on volume and cost. Brands that lean entirely on AI creative will lose ground on authenticity. The accounts that win will use both deliberately.

Vertical, sound-on, and short-form will remain the dominant formats. Roughly 90% of Meta inventory is now 9:16 vertical. Short-form video continues to grow as a share of consumption. Sound-on content has overtaken sound-off in many demographic segments because Reels and Stories normalized the behavior. The accounts shipping creative built for desktop horizontal viewing are leaving meaningful CPM efficiency on the table every day. This is the easiest fix in many accounts we audit, and it’s the one most often overlooked.

The honest framing on the future is that nobody fully knows how the next 24 months will play out. Meta itself does not fully know what Andromeda will surface as it continues to learn. What we do know is that the accounts adapting quickly are the accounts winning, and the accounts holding onto 2019 playbooks are losing more ground every month. Our job is to be ahead of the platform, not behind it, and to make sure our clients are running the version of Meta that exists today, not the version that existed when their last agency learned the platform.

Meta runs the largest advertising system on earth across Facebook and Instagram, and the most expensive mistake businesses make is treating them as one thing. They share an ad platform and almost nothing else: different audiences, different creative, different jobs. What follows is an honest account of what works on each, and how to run the whole machine without pouring money into the wrong one.

---

What actually works on Facebook ads

Facebook is the platform everyone underrates, usually because they have decided it is “for old people,” and that read is both wrong and the exact reason it is one of the best values in advertising. Start with the scale, because it is easy to forget how large it still is: roughly 250 million users in the United States alone, reaching close to 72 percent of all American adults, on a platform people open out of habit several times a day. Nothing else in advertising puts you in front of that share of the country.

The “it’s just old people” myth does not survive the actual breakdown either. The single largest US age group on Facebook is 25 to 34, at around a quarter of users. Another 37 percent sit in the 35-to-54 band, the prime earning and spending years. Yes, the 55-and-up audience is over 30 percent and growing, but that is not a weakness, it is the part competitors are too snobbish to value. Adults 50 to 64 are among the most engaged daily users on the entire platform, more than half check in every day, and over half of Americans 65 and older are active. Those are the people with paid-off mortgages, real disposable income, and the highest intent to actually buy. While every younger advertiser piles into Instagram and TikTok, that older, wealthier Facebook audience sits comparatively under-competed, which is precisely why it is cheaper to reach.

The economics make the case plainly. Facebook lead ads routinely bring in leads around $1.92 each, against more than five dollars for a comparable lead on Google, because you are generating demand from a targeted audience rather than bidding against everyone for a scarce keyword. Traffic clicks average well under a dollar, CPMs sit in the low-to-mid teens, and conversion rates across Meta average near 9 percent, though that figure swings enormously with the audience and the offer. For the right business, this is some of the most efficient paid reach available anywhere.

Then there is the targeting, which remains Facebook’s quiet superpower even after Apple’s privacy changes thinned the signal. The depth of first-party data Meta holds, combined with custom audiences built from your own customer lists and lookalike audiences modeled off your best buyers, still lets you find and clone your ideal customer at a level of precision most channels cannot approach. Layer in the surfaces that are unique to the platform, the main feed, Stories, Marketplace with its own enormous user base, and Groups, and you have reach and targeting working together rather than one at the expense of the other.

What Facebook is genuinely best at follows directly from who is there and how they behave. Lead generation is the standout, especially through native instant forms that let someone convert without ever leaving the app, which is why it dominates for local services, home services, healthcare, finance, real estate, events, and B2B lead capture. Considered and higher-ticket purchases do well because the audience is older and financially established enough to make them. And the creative can carry far more information than Instagram tolerates, because a Facebook user will actually read a longer caption and click a link, which makes it the better home for anything that needs to explain itself before it sells.

It is the wrong platform for a narrow but real set of cases, and saying so is the honest part. If your customer is squarely Gen Z, or your product lives and dies on aesthetics and visual discovery, you are describing Instagram, not Facebook. But for the broad universe of lead-gen, local, service, considered-purchase, and B2B businesses, Facebook consistently outperforms the channels that get all the hype, for the simple reason that it gets less of it.

---

What actually works on Instagram ads

Instagram is the opposite platform in nearly every way that matters, and the clearest proof of where the momentum went is the money: Instagram now generates more than half of Meta’s US advertising revenue, up from under eight percent a decade ago. It overtook Facebook as the company’s growth engine, and it did so on the back of a fundamentally different audience and a fundamentally different way of using a phone.

That audience is young and visual. Roughly 62 percent of Instagram users worldwide are between 18 and 34, and about three-quarters of US adults under 30 are on it, though the platform has aged up enough that the late-30s and 40s are now a real presence too. The unifying trait is not age so much as intent: people are on Instagram to be shown something, to scroll and discover, not to read and deliberate. That one fact dictates everything that works, and everything that does not.

Creative is the entire game here, and the bar is specific and unforgiving. An Instagram ad has to look like it belongs in the feed, polished the way genuine content is polished, not the way a corporate ad is. The post that obviously came from a marketing department gets thumbed past in half a second; the one that looks like a creator or a real customer made it gets watched and shared. This is why the highest-performing Instagram creative now leans heavily on user-generated and creator-style content, and why advertising on the platform seriously means running a creative pipeline, not a campaign you set once. You are feeding a machine that eats creative, and the businesses that win are the ones producing enough of it, fast enough, to keep finding the winners.

Format matters as much as content, because Instagram is really several surfaces with different rules. Reels now account for around 35 percent of all time spent on the platform and are the primary discovery engine, which makes short vertical video the dominant unit and the one the algorithm pushes hardest. Stories carry the largest share of ad impressions, full-screen and immediate, ideal for urgency and quick offers. The feed holds the rest, and Explore and Shop sit alongside for discovery. Each surface wants creative built for it, and reusing one format across all of them is a reliable way to underperform on every one. This is also the deeper shift worth understanding: Instagram, like TikTok, now ranks content by predicted interest rather than by who follows whom, so your reach has almost nothing to do with your follower count and almost everything to do with whether the creative earns the impression.

What Instagram is best at follows from the audience, the format, and the behavior. Brand building and visually driven products win here, fashion, beauty, food, fitness, travel, home, anything that photographs and films well. Direct-to-consumer and ecommerce thrive on genuine shopping behavior, with surveys consistently finding that around 70 percent of users have used Instagram to research or buy products, and a large share planning purchases directly through the platform. Impulse and discovery purchases, the things people never woke up wanting but buy when something beautiful stops their scroll, are native to Instagram in a way they are not to a search engine. And creator-led promotion, through partnership and whitelisted ads that run the creator’s own content as the ad, is some of the strongest-performing media on the platform.

Where it struggles is the exact mirror of Facebook’s strengths. Dry, information-heavy, or B2B offers tend to underperform, and lead generation that depends on a user reading and weighing before acting usually does better on Facebook. Instagram rewards the visual, the immediate, the authentic, and the young, and it quietly punishes anything that needs a paragraph and a moment of patience to make its case.

---

You can’t turn a cold stranger into a customer with one ad

The most common way businesses waste money on Meta is also the most intuitive: they build a single campaign, point it at a cold audience, ask for the sale, and wait. It does not work, and the reason is structural, not something a better headline fixes. Meta is interruption advertising. Unlike Google, where someone is actively searching for what you sell, on Meta you are reaching people who were not looking for you at all, mid-scroll, between a friend’s photo and a funny video. Asking a stranger to hand over a credit card the first time they have ever seen you, at the exact moment they trust you least, is asking for the hardest possible action at the worst possible time.

The data on the gap is stark. Cold social traffic typically converts somewhere between half a percent and one and a half percent. Warm audiences, people who have already engaged with you or visited your site, convert several times higher, with well-run retargeting commonly landing in the mid-teens and lookalikes sitting in between. The same studies put retargeting’s cost per acquisition 40 to 70 percent below cold, and retargeted ads routinely earn many times the click-through rate of cold ones. A dollar spent on a warm audience is not slightly better than a dollar spent cold, it is multiples better, because the trust that makes someone buy has already been partly built.

That is what a funnel actually is, and why Meta requires one. At the top, cold campaigns do the job cold campaigns are good at: introducing you, earning attention, getting a video watched or a post engaged with, building an audience of people who now know you exist. In the middle, you retarget those engagers and visitors with content that deepens interest and answers objections. At the bottom, you retarget the people who showed real intent, the add-to-carts, the repeat visitors, the past customers, with the direct ask. Each stage uses different creative and a different objective, because a stranger and a warm prospect need completely different things from you. Collapse all of that into one cold conversion campaign and you are running only the hardest stage, aimed at the coldest people, and then calling the poor result a Meta problem.

There is a real exception and a real complication. The exception is low-cost, visual, impulse products, which genuinely can convert cold, because the decision is small and a scroll-stopping image does most of the work. The complication is that Apple’s privacy changes shrank retargeting audiences and degraded the signal that powers them, which makes the funnel harder to build than it was a few years ago and makes feeding Meta clean first-party data through the Conversions API more important than ever. Neither changes the core truth: for anything considered, anything with a real price tag, the sale is earned across several touches, not demanded in one.

This is also why judging Meta on a single cold campaign’s day-one ROAS leads people to kill the exact campaigns that would have worked, and why “boosting a post” almost never produces real customers. The work that makes Meta profitable is building and sequencing the funnel, matching the creative and the objective to each stage, and keeping the warm audiences fed at the top so there is anyone to retarget at the bottom. Done right, the cold spend that looks unprofitable in isolation is what fills the warm and hot stages that quietly produce the return.

---

What a Meta ads audit actually covers

Whether you call it a Meta, Facebook, or Instagram ads audit, it is the same exercise, and the first thing to understand is that a Meta account can look perfectly healthy on the dashboard while it quietly bleeds money. Meta grades its own work, and it grades generously, so the numbers on the screen are the last place a real audit starts. The job is to ignore the self-congratulation, reverse-engineer what is actually happening to the budget, and find the specific reasons the account is leaking or refusing to scale, which are different every time and almost never the things a checklist flags.

The structure is usually the first thing that does not survive scrutiny, and the failure is counterintuitive. The instinct is that a struggling account is too simple, when far more often it has been shattered into too many pieces. We routinely open accounts with thirty ad sets each scraping a handful of conversions a week, none of them ever clearing the roughly fifty weekly conversions Meta needs to exit its learning phase, which means the algorithm has been guessing for months and was never actually allowed to learn. The dashboard shows a busy, fully built-out account. The reality is a budget sliced so thin that nothing in it can optimize, and consolidating that sprawl is often the single highest-impact move available.

Creative is the next place the truth hides, because it is where Meta accounts actually live or die and where the dashboard is most misleading. A real audit is looking for fatigue the platform will happily let you ignore: the ad that worked three months ago, still running, frequency climbing as the same people see it for the ninth time, performance decaying a little every week while the report calmly shows it “active.” It is looking at whether there is any creative volume at all, because the account cycling thirty native variations a month is going to bury the one running three stretched-square images across every placement, and it is looking at whether the creative is built for where it runs, vertical video native to Reels and Stories or the same static image dropped everywhere and ignored. Most underperformance that gets blamed on targeting or budget is creative starvation, and you only see it by looking past the headline metric at frequency, at refresh cadence, at what is actually in the account.

The funnel is where the money leaks in plain sight. We regularly find accounts paying, every single day, to introduce themselves to cold strangers and then doing nothing with the people who engaged, no retargeting, no warm audiences built from video viewers or site visitors, the add-to-carts and repeat visitors left to go cold. That is the equivalent of buying a room full of interested prospects and walking out without taking a single phone number. The audit maps whether a real cold-to-warm-to-hot structure exists or whether the whole budget is stuck demanding sales from people meeting the brand for the first time, and whether custom and lookalike audiences are built off the right source data or off nothing.

The signal layer is where a lot of invisible damage lives, especially since Apple’s tracking changes. We check whether the pixel is even firing on the events that matter, whether the Conversions API is set up so Meta still receives server-side data rather than guessing, whether the optimization is pointed at a real purchase or a cheap proxy event that trains the algorithm to chase the wrong thing, and whether the attribution window is sane. An account feeding Meta broken signal is asking the algorithm to optimize blind, and no creative or budget change fixes a machine that cannot see.

Measurement gets the hardest look of all, because this is where Meta lies to you most. Platform ROAS routinely over-attributes, claiming conversions that would have happened anyway and credit that belongs to other channels, so a real audit sets what Meta reports against actual revenue and blended performance instead of taking the in-platform figure at face value, and it is common to find the in-platform number is roughly double the truth. An account “hitting a 4x ROAS” on the dashboard while the bank account says otherwise is not a success, it is a measurement problem, and finding that gap is often the most valuable thing the audit produces.

Underneath all of it sits the wasted spend those findings add up to, and the output is never a list of two hundred settings. It is a prioritized diagnosis: the few things actually costing you, why, in what order to fix them, and what each is worth in real revenue, not platform-reported revenue. That is the line between an audit and a screenshot of your account. One tells you what is switched on. The other tells you what to do, why, and what it will return, and it is investigative work for someone who has seen what breaks Meta accounts, not a template, and certainly not the platform grading its own homework.

---

Stats to verify before publish

• Facebook: ~250M US users, ~72% of US adults; largest age group 25-34 (~24%), 35-54 (~37%), 55+ (>30%); 50-64 heavy daily use.
• Facebook lead ads ~$1.92 CPC vs ~$5 Google; CPM low-to-mid teens; Meta avg conversion ~9% (varies widely).
• Instagram: ~62% of users 18-34; ~50% of Meta US ad revenue; Reels ~35% of time on platform.
• Instagram shopping: ~70% of users have used it to research/buy.
• Cold social converts ~0.5-1.5%; retargeting several times higher (mid-teens), ~40-70% lower CPA.
• Meta learning phase ~50 conversions per ad set per week (Meta’s documented threshold).