You can't out-target the algorithm. You can only out-supply it.

For most of the history of paid social, the edge lived in the account. Pick the right interests, build the right lookalikes, split the budget across the right placements, exclude the right people, and almost any reasonable ad would find its buyers. Agencies sold account structure because account structure was where performance came from, and a media buyer who could build a clean account was worth more than the creative team feeding it.

That is over. The platforms have converged on the same approach: the platform decides who sees an ad, in real time, by reading how people respond to the creative rather than the audience someone picked. Interests, lookalikes and placements still exist. They are just no longer the lever. Two accounts with the same offer and the same budget now separate on what they feed the system, not on how the system is set up.

Meta’s version is called Andromeda. Meta Engineering describes a retrieval system with 10,000 times the capacity of the one before it, narrowing tens of millions of candidate ads to a few thousand and matching ads to live behavior (Meta Engineering, Andromeda, December 2024). The name is not the point. What it replaced is. Account configuration used to be the work. Now it is largely handled for you.

Meta says the focus has shifted from niche targeting to creative diversification (Meta for Business, The Creative Advantage, 22 April 2025). Google and TikTok have moved the same way. Performance Max and Smart+ find the audience automatically, and Google’s own guidance is to add as many assets and variations as possible (Google Ads Help, How asset groups work; TikTok for Business, Smart+, May 2026). So performance no longer separates on how well an account is built. It separates on what gets fed into it, how varied that supply is, and how quickly it is replaced.

This is worth sitting with, because most of the industry still sells the old thing. Audits of account structure, bid strategy reviews, placement optimization: real work, honestly done, aimed at a lever that is mostly gone. The lever that is left is the ad.

Targeting used to work like this. You picked an interest, say pets, and built an ad to match it. The trouble was that one label never captured what actually predicts a buyer. Age, habits, values, the dozen signals sitting underneath the word pets: none of that was in the word, and a brand that targeted pets reached everyone who had ever liked a dog video.

Now it runs the other way. The ad goes in with little or no targeting attached, and the system learns who responds to the ad, whatever it was that stopped them, the face, the setting, the pace, the words, and sends it to more people like them. Every ad is a hypothesis about who it is for, and the platform tests it for you, continuously, against live behavior.

That changes what variety means. It is not a nice extra. It is the raw material the platform uses to find your buyers in the first place. A brand with one ad reaches the people that one ad fits, and no one else. A brand with twelve genuinely different ads is running twelve hypotheses at once, and the platform is sorting the market for it: this face for these people, this argument for those, this format for the ones who never watch a video.

There is a catch, and operators see it every day. Near copies of the same ad behave as one ad. A new color, a new end card, a slightly different first line: operators consistently see the system group them with the original, so they compete for the same reach instead of adding any. A genuinely new hook, a new face, a new argument behaves like a new ad and earns its own audience. Variety only counts when the platform can tell the difference.

The practical consequence is strange at first. The targeting work has not disappeared. It has moved into the creative brief. Deciding who an ad is for, which belief it argues against and which buyer it should stop is still the work. It just happens before the ad is made, in the choice of what to make, instead of after, in the account.

Nobody we know of can reliably pick the winner in advance. Not the founder, not the agency, not the creative director with twenty years of taste. Taste raises the odds. It does not remove the need to test. Every experienced team has a story about the polished ad that died and the rough one shot on a phone that carried the quarter, and the stories do not cancel out into a rule. They add up to one: make enough, test honestly, and let the buyers decide.

Run the arithmetic at the industry rate and the case makes itself. Thirty genuinely different new ads give an account roughly one or two real winners. Five give it, most of the time, none, and a team that makes five ads a quarter spends the quarter wondering what went wrong. That is not a promise about any account. It is why supply matters more than any single ad, and why a plan built around one great ad is a plan built around a coin toss.

Two things follow. First, a miss is not a failure; it is the expected outcome of most launches, and the system has to be built to absorb it. Second, the rarer the winner, the more it matters what happens to it once it is found, which is the next chapter.

Performance rarely collapses all at once. It fatigues first, then erodes. A winning ad earns strong signals, so the platform sends it more delivery. Delivery concentrates. Frequency climbs. The account leans harder on a shrinking set of ads, and the same people see the same ad until they stop responding to it.

The numbers are blunt. Buyers are 45% less likely to click an ad they have already seen four times (Analytics at Meta, Creative Fatigue). NP Digital measured click rates across 50 companies falling about 27% by the eighth week live (NP Digital, Ad fatigue: how CTR decays week over week, April 2026). None of this is a flaw. It is how the budgeting works: spend flows toward whatever is winning, which earns it even more delivery, which wears it out faster. The better the ad, the faster the platform uses it up.

Every repeat look lowers the odds of a response; by the fifth look they are down by about half. A genuinely different ad brings them back. A small variation does not, for the reason in the last chapter: operators see the platform treat it as the same ad, and so do the people who have already scrolled past it. The platforms are built to keep pushing what works. They will not refresh your creative for you. That part is on the brand.

What looks like volatility usually is not random. It is a timing gap between how fast the platform scales a winner and how fast a team can produce its replacement. The winner falls below the useful floor and stays there until the next ad climbs over it, and the account pays for every week of that gap. Close the gap and the line goes flat. Most teams never see the gap as a gap. They see a good month, a bad month, and a reason to blame the platform.

So the four facts of Part I stack into one picture. The platform finds the buyers from the ad. Winners are rare and cannot be called. Every winner wears out. And the only thing a brand controls in all of that is the supply: how many genuinely different ads go in, how good they are, and how soon the next one arrives.

The first easy answer to Part I is the one most brands with a real budget reach for. Spend it on one beautiful production. Hire the director, build the set, run it for a quarter, and make it the ad everyone remembers. The thinking is honest: if winners are rare and quality raises the odds, then put every chip on quality.

Three things go wrong at once. The ad takes months, so by the time it runs the brief is old. It is one hypothesis, so the platform can only find the people that one ad fits, and the arithmetic of the last part says the odds it is a winner are about one in twenty, taste included. And if it does win, it fades on schedule, and there is nothing behind it, because the next one is another quarter away. The account rides a single curve down and calls the decline seasonality.

The hero ad is not the mistake. Brands need ads of that quality. The mistake is asking one ad to carry a system that needs thirty. Quality without quantity is a bet, and the house wins most bets.

The second easy answer runs the other way, and it is now the default, because the tools made it free. Make as many ads as possible, as cheaply as possible. Fifty versions of the same thing with new colors. A feed full of generated faces that look like everyone else’s. A template with the product swapped in. Hundreds of outputs a week and a dashboard that says the account has never been busier.

Two things go wrong here too. Operators consistently see the platform group the near copies as one signal, so fifty versions buy the reach of one, and the frequency problem of Part I arrives with a bigger invoice. And buyers, we find, scroll past what looks generic, because they have seen that face, that template, that pace a hundred times that week from brands they will never remember. The brand that set out to out-supply the algorithm ends up cheapening itself in public.

AI makes the volume trap the default. When anyone can produce a hundred ads in an afternoon, volume stops being an advantage and quality becomes the scarce half. The brands that win the next few years are not the ones that make the most. They are the ones that make enough, and make every one of them worth a buyer’s stop. AI earns a real place in that: briefs, scripts, boards, parts of the edit, the native versions each format needs. It does not decide what is worth making, or whether it is any good, and a system that lets it decide is a system for looking like everyone else faster.

Put the two traps on a grid and the shape of the answer is obvious. Quality is one axis. Quantity is the other. The hero ad sits high and left: excellent, alone. The volume trap sits low and right: plentiful, forgettable. The bottom left is where most accounts actually live, a few ads made fast, nothing to learn from and nothing to scale. The top right is the only square that works, and nothing drifts into it by accident.

Quality and quantity are not a trade off to manage. They are one job, and the job has a shape: enough genuinely different ads to give the platform real variety, each one good enough to stop a buyer and say something true, each one made for a particular buyer at a particular stage, and each one replaced before the last one fades. Miss any part and the account slides toward one of the traps.

That job cannot be done by a brief and a vendor. It needs research that keeps finding what to say, production that can make a lot of it well, testing that reads honestly, and a replacement rhythm that never stops. In other words, a system. Part III is that system, step by step. First, though, why it is rare.

Most organizations are not set up to treat creative as the main lever for paid growth, and the reasons go deeper than production capacity.

The pace gap is real. Most brands are working with long approval chains, slow production cycles and a few hero assets asked to carry performance far longer than they can. When replacement arrives late, the cost of a customer rises, and the diagnosis is almost always wrong: the targeting, the platform, the season.

Underneath that sits an accounting problem. Most businesses we meet track ad spend as the thing that drives revenue and file creative production under overhead, something to trim when the quarter tightens. That is backwards. The platform is a distribution layer that scales whatever it is given. Creative is the input that generates the return. Cut production first and the ads go stale, and the drop gets blamed on everything except what caused it.

The third gap is structural. Even a team with the budget and the right books usually has no loop: nobody owns the question of what to say next, nobody owns a test that teaches, and nobody owns replacement as a schedule rather than a reaction. The fix is not more assets. It is an operating structure that keeps finding what to say, making it well, testing it honestly and replacing it on time. The rest of this thesis is that structure.

Every strong ad starts with something a buyer already said. The words that move people are written down long before any brief: in reviews, comments, searches, forums and messages, the places people talk when no brand is listening. The system starts there, and it uses the same five sources every round, because the sources do not run dry; the buyer keeps talking.

The output of listening is not a report. It is a list of lines, in the buyer’s own words, sorted by who said them and what they believe. Those lines become the hooks, the headlines and the objections the ads will answer. Here is what that looks like on an account, with the place each line came from.

Not every pain carries equal weight. The ones that cross a buyer’s tolerance threshold, the point where a problem becomes urgent enough to act on rather than just notice, produce, in our experience, far stronger response. So segments are ranked by how far past that threshold their pain sits, not by how big they are. A large audience that is mildly annoyed is a worse place to start than a small one that is awake at three in the morning.

Most buyers are not ready to purchase the moment they see an ad, and most of any market never is at any given moment. Some do not know they have the problem. Some feel it but do not know there is a fix. Some are comparing approaches, a few are comparing you with others, and fewer still are ready to buy today.

One ad cannot speak to all of them. An ad that names the problem bores the buyer who is ready, and an offer means nothing to someone who has never felt the problem. So the work is spread across the stages on purpose, and the account always has something running for each.

Each stage asks a different thing of the ad. The unaware buyer needs to see the problem. The problem aware buyer needs it named and explained. The solution aware buyer needs to see why this fix works. The product aware buyer needs to win the side by side. The most aware buyer needs a reason to act today. Same product, five jobs.

Any line a buyer said can become the message for any stage. A search about why a problem keeps happening becomes an ad for the buyer who feels it. A review that doubted it would work becomes an ad for the buyer who is comparing. The stage decides the job of the ad. The buyer’s own words decide what it says. And the share of buyers at each stage decides how many ads each stage gets, which is why the early stages, where most of the market sits, are never left to one ad.

Ads are not produced at random. They follow a structure built to refine the message on purpose instead of stumbling into it. Four layers, each one narrower than the last, and each one a decision that can be tested on its own.

Personas work best when they are specific rather than demographic. Beekeeper describes almost no one in particular, and an ad built for it comes out generic. A first time backyard beekeeper who just lost a hive to mites and wants a setup simple enough to keep the next one alive is a real person you can write to. The audience is smaller, the ad is sharper, and, in our experience, it converts better because of it.

Many teams collapse angle and concept into one step. Keeping them apart forces precision. Two concepts can share an angle and look nothing alike, or share a format and argue completely different things. Keeping the layers distinct is what makes a test readable and a brief usable: when an execution fails, you know which layer to question, and when one wins, you know what to remake.

One thing sits outside the layers on purpose: the offer itself, the product, the price and the packaging. That belongs to the brand. The work is built around the offer as it exists, and if testing points to an offer problem, it gets flagged rather than changed.

This is where strategy becomes ads, and where the quality and quantity problem is won or lost, because volume without a production system either costs a fortune or looks cheap. A round of ads is planned as a set, not as a list of requests, and it is planned around three kinds of work.

Not every ad needs to be invented from scratch. Remaking a proven winner is often the highest leverage thing a team can produce, and it is undervalued next to the appetite for something new. Iterating on a winner’s persona, format or hook comes next. Net new concepts are essential, because the system does not compound without them, and they are also the easiest place to overspend relative to how often they win.

Every concept gets made in more than one format, chosen to fit the buyer rather than brand preference. Operators find that a format someone never engages with is rarely served to them, so format is a real limit on reach, not only a creative choice. The choice follows a sequence, and the order is the point.

Statics hold real weight at every stage when the concept calls for one. Video is not better because it feels premium. It is better when the concept needs motion, a face or a demonstration. A round that is all video has usually skipped the sequence.

The physical side matters as much as the strategy. Volume at quality comes from planning a shoot day around many concepts at once, on sets that are already lit, with the people who planned the ads directing them, and hooks shot in runs so one body of footage yields many openings. A day planned that way yields a round. A day planned around one ad yields one ad.

One shoot day25 ads13 pre-lit setsRonin Creative Studios, Miami

AI earns its place here too: briefs, scripts, boards, the native versions each format needs, parts of the edit. It makes a well planned day go further. It does not plan the day.

Ask an experienced media buyer what actually predicts whether an ad works, and most will tell you testing runs backwards. Format gets tested first, because it is the easiest thing to swap. Persona comes next. The angle, the argument the ad makes against a belief, barely gets touched, because it is the hardest thing to isolate and the easiest to leave alone.

That order is exactly wrong. If a buyer has already shown they will buy and results drop, the fix is almost never the format. It is the message. Testing in the wrong order does not only waste budget. It teaches nothing, because it changes the variable least likely to be the problem.

So every concept launches with four things defined before it runs: a hypothesis tied to a specific belief, a test budget big enough to produce a real decision, a window for when the result gets read, and the bar it has to clear, set from the brand’s own numbers. Those numbers come from the business. The test runs against them; it does not set them. Then every concept ends a round in exactly one of four states.

Read results at the right level. A single ad is the noisiest read, an ad set is more trustworthy, and a whole round is the most reliable, which is why the health number of the system, creative win rate, is read for the round and never for one ad. For a clean read, each concept runs in its own ad set; mixing concepts in one makes it impossible to tell which is working. That is a structure recommendation for whoever runs the account.

Last click reporting credits only the final touch, so an ad speaking to buyers who are not looking yet will rarely look strong on its own, even while it makes every later ad cheaper. Operators learn to discount two numbers: dynamic product ads, which often take credit for sales that were coming anyway, and platform reported return, which counts existing customers. The round level read and the business’s own bar are the defense against both.

Most concepts will not clear the bar, including some that look promising early. That is the point of testing: to find out cheaply, before spend makes the answer expensive. And a winner is never ridden until it dies. It becomes the baseline, and the next round carries the same message in new hooks, formats and proof styles, which is where Part III ends up.

An ad introduces a belief. The page it sends people to has to carry it. When someone clicks, they arrive with expectations shaped by the ad they just saw. If the page continues the same message, they keep going. If the message shifts, they have to start over, and most, in our experience, do not. The ad can be perfect and the round still loses here.

Because ads are made for different stages, the page has to match the stage of whoever lands on it. An ad for a buyer who just discovered the problem cannot send them to a checkout and expect them to stay; it sends them to the explanation it promised. An ad for a buyer comparing options sends them to the comparison. Three things carry continuity: the same message, the mechanism and the proof, and as little friction as the stage allows.

The page is not something the creative system builds. It belongs to the brand or the vendor that already owns it. What the system owes the brand is one written finding per winner: it holds, or here is the line that breaks it. A broken page is never counted against the ad, and a winner whose page breaks is not scaled until the page is fixed.

Follow up ads work the same way. They should answer the specific reason someone did not buy, not repeat the first ad louder.

A winner arrives here already proven. The job now is replacement, and that starts with understanding why an ad is fading, not just that it is. Two kinds of fatigue look the same on a dashboard and need opposite fixes.

Every round, then, does three things. It keeps the winners and carries them deeper, into new hooks, formats and proof styles. It replaces what faded, on a schedule, before the gap in Part I opens. And it banks what it learned.

The bank is the quiet part of the system and the most valuable. Every ad is tagged by its persona, stage, angle, hook, format and proof style, and every verdict is recorded with it, including why a winner faded. So each round starts from what the last one taught instead of from a blank page: which arguments this buyer answers, which hooks this stage needs, which formats this audience will not watch.

Without structured replacement, delivery piles onto a shrinking set of winners, frequency rises and costs climb. With it, new ads arrive as old ones decline, the account stays steady while the brand decides how hard to scale, and the next round is faster to make than the one before, because more of it is remaking and iterating and less of it is guessing. That is what compounding means here. The herd keeps moving.

Creative is the lever, but it is not the whole machine. Four things have to be in place for any of this to show up in results, and a missing one gets named before a single ad is made, because no round of ads can make up for it.

None of these are the creative system’s to build, and all of them are its to check. A round launched against a page that breaks, tracking that lies, an account nobody is reading or an offer nobody wants produces verdicts that mean nothing, and the system learns the wrong lesson from them. The order of operations is to confirm the four, then make the ads.

Marketing and finance speak different languages. Marketing reports in engagement: did people stop, did they watch, did they click. Finance reports in what a customer costs and what they are worth. Most creative work we see only speaks the first, which leaves the marketing team standing alone every time the spend has to be defended.

A creative system worth the name runs inside the financial guardrails a business already has. They are captured at the start and treated as inputs: what a new customer may cost, what margin a sale has to protect, how fast it has to pay back. Every concept is judged against that bar, not against platform numbers finance does not trust. Hook rate says whether an ad stops people. The cost ceiling says whether it is a winner.

That is also what makes the scaling decision a decision rather than a bet. The system does not produce scale on its own, and it does not decide when spend increases; that is the brand’s call and its media buyer’s, made against their own numbers. What the system produces is the signal that makes the call informed: which ads cleared the bar, at what cost, with what margin left, and what is in the can to replace them. Marketing keeps creative authority. Finance keeps financial authority. The system makes their decisions legible to each other.

That is the whole system. Listen, spread the work across the stages, build in four layers, make enough and make it well, test the message first, keep the promise on the page, keep, replace and bank, confirm what has to be in place, and report in the numbers the business already uses. Nine parts, each handing something to the next, and the last one feeding the first.

Everything above is how paid advertising works now, whoever runs it. For the record, here is where The Herd sits in it. We make the ads, and the strategy behind them: the listening, the personas and angles, the concepts, the shoot days, the cuts, the reads and the bank. We do not run ad accounts, and we never have. Your media buyer, in house or a partner, keeps the account, the budget and the delivery, and we work beside them every round: we hand them the set, they run the test, we read it together, and the next round starts from what it taught.

The physical capacity comes from Ronin Creative Studios in Miami, our partner studio: 13 pre-lit sets and a working test kitchen, so a shoot day goes to shooting rather than setting up.

Recognized by the Telly Awards, Best in Biz and the Addy Awards. The awards are for the work. The thesis is for the system that keeps producing it.