You ran five ads. Four of them did roughly nothing. One of them is quietly booking most of your jobs, and every time you open the account it is still there at the top, doing the work.
So you do the obvious thing. You tell whoever makes your ads, or you tell yourself, to make more like that one.
And that is where most of the money goes sideways, because "more like that one" almost always means copying the most visible thing in the ad. The color. The person in it. The photo. The first line. Whatever you would describe if someone asked you what the winning ad looks like. The trouble is that what an ad looks like and why it won are rarely the same thing, and the part you can see from across the room is often the part that did nothing at all.
Here is the short version, and then we will take it apart properly. A winning ad is a bundle of decisions that all arrived together, and your ad account hands you one number for the whole bundle. You find out which decision mattered by lining the winner up against the ads that lost in the same batch, because those losers ran against the same people at the same time and they rule suspects out for free. Then you prove your best guess by building two new ads that differ in that one thing and nothing else. That is the whole method. The rest of this article is how to do each step without fooling yourself.
A Winning Ad Is Seven Decisions Wearing One Result
Every ad you have ever run, however simple it looked, was the product of about seven separate choices. Somebody decided them, even if nobody decided them on purpose.
Who it speaks to. The homeowner after a storm, the new parent, the owner of an older car, the person who has already been burned once by a contractor.
The reason to buy. The argument the ad is making. Fast, cheap, safe, local, honest, better than what you have now. Two ads can show the same product and argue completely different cases for it.
The opening. The first line of text or the first second of video. The part that decides whether anyone sees the rest.
The proof. A review, a number, a before-and-after, a license, a face people recognize. Or nothing.
The offer. What you are asking them to do and what they get for doing it. A free inspection, ten percent off, a quote in an hour.
The look and format. Static image or video, phone footage or polished, bright or muted, long or short.
The person on screen. You, a customer, an employee, a stranger, or nobody.
Then there are three more things that are not in the ad at all but shaped its result just as much: when it ran, who Meta chose to show it to, and how much of your budget it was allowed to spend.
That is ten possible explanations. Your ad account reports one cost per result. Nothing on that screen tells you which of the ten earned it.
If that sounds like a problem only beginners have, it is not. Dara Denney, a creative strategist who has had a hand in more than 20,000 ads, walked through the formats working best for her agency's clients this year, and on one of them, an ad that opens with the brand apologizing, she said plainly: "I actually don't exactly get why it works." She had tested it, it kept winning, it scaled, and she still could not tell you which part was doing it. That is an honest thing to say out loud, and it is far more common than the confident teardown videos suggest.
The confident teardowns have the problem too. Fraser Cottrell, who co-founded an agency that makes ad creative for online brands, broke down a winning ad for a protein shake and credited its success to the customer review placed at the top of the image. Maybe so. In the same breath he mentioned that it ran early in the year, the stretch when a lot of people are trying to lose weight. That is two explanations for one result, and nothing in the ad account can separate them. If the review was the reason, you should put reviews at the top of everything. If the calendar was the reason, the same ad in July might do nothing at all. Those lead to very different next steps, and you only get to pick one.
Make Sure It Won the Customer, Not Just the Budget
Before you explain why an ad won, check that it actually did. A surprising share of "winners" are ads that were simply handed the money.
Meta does not split your budget evenly across the ads in an ad set. It picks favorites early and pours most of the spend into one or two of them. We have written about why five ads in an ad set usually means one ad actually running. The consequence here is simple: the ad at the top of your spend column may be at the top because Meta bet on it, not because it outperformed anything. The ads it barely funded never got a fair trial. They did not lose. They were never really tested.
So run four questions before you go looking for reasons.
Did it win at comparable spend? Look at cost per result, not share of spend. If the "winner" spent $900 and the runner-up spent $40, you know a lot about the first ad and almost nothing about the second. Biggest spender and best performer are different claims.
Is the gap bigger than luck? If the winner produced six leads and the next ad produced four, you do not have a winner. You have two ads that are about the same and one of them got lucky that week. The smaller your budget, the more patience this takes.
Was something else carrying it? A seasonal rush, a promotion running on your website at the same time, a week you were mentioned in the local news, a storm that sent every homeowner in town looking for a roofer. If the ad launched into a tailwind, the tailwind deserves some of the credit.
Are you judging it on the number that pays the bills? An ad that wins on clicks or on how long people watched can lose badly on booked jobs. Judge it on the result you actually bought it for, which is the argument behind the four numbers that decide your Meta ads. And if your tracking was broken for part of the period, the history is not evidence of anything. The same test we apply to an inherited account's old data applies here.
If the ad passes all four, it genuinely beat its neighbors. Now it is worth understanding.
The Ads That Lost Are the Most Useful Thing You Own
Most owners look at a winning ad on its own and try to reason out why it is good. That almost never works, because every part of a winning ad looks good once you know it won. The review looks persuasive. The color looks striking. The person looks trustworthy. Hindsight flatters all of it equally.
The fix is to stop looking at the winner alone. The ads that lost ran against the same audience, in the same weeks, on the same budget, through the same season. Anything the winner shares with a loser cannot, on its own, be the reason the winner won, because the loser had it too and lost anyway. That makes your losing ads a control group, and you already paid for it.
Here is how to use them. Take the winner and every ad from the same batch that got a real share of spend (leave out the ones Meta starved; they tell you nothing). Write the seven decisions down the side of a page and the ads across the top. For each decision, note what each ad did. Then look for the row where the winner stands alone, the one decision where it did something that no losing ad did.
A worked example makes this concrete. It is illustrative, not a client account, but it is built on a pattern we see constantly. A local roofing company ran five ads through the spring:
- Ad A, the winner: a 30-second phone video of the owner standing on a customer's roof, pointing at the damage, offering a free inspection, with a customer review in the caption. Around $38 per inspection booked.
- Ad B: a photo of a finished roof, the same free inspection, the same review. Around $95.
- Ad C: a 30-second phone video of the owner sitting at his desk, making the same offer, with the same review. Around $88.
- Ad D: a finished-roof photo aimed at owners of older roofs, with no review. Around $140.
- Ad E: a short drone video over a neighborhood, a financing offer instead of the free inspection. Around $110.
The owner's instinct is that Ad A won because customers like seeing the owner talk to them, and his next ten ads would have been him talking to camera. But Ad C had the owner talking to camera too, in the same video format, with the same offer and the same review, and it lost. So it was not the owner, it was not phone video, and it was not the offer or the review. Down the grid, only one row has Ad A standing alone: the opening. It is the only ad that opens on a real roof with real damage on it.
That is the suspect. Not proven yet, but narrowed from ten explanations to one, and the losing ads did most of the work.
One warning. If every ad in your batch differed from every other ad in five ways at once, the grid has nothing to compare and the losers cannot help you. That is not a reason to test tiny changes; you still want genuinely different ideas in the ring, as we argued in You Don't Find the Winning Ad. You Test Your Way to It. But it is a reason to know, going in, which idea each ad is carrying, so that when one wins you can read the result instead of guessing at it.
Name the Winner in Five Words
Once you have a suspect, give it a short name. Not "the blue one" or "the video from March." A name that says what the ad is doing: "the owner on a real roof," "price in the first line," "the customer who almost didn't call," "before and after, same angle."
This sounds like a small thing. It is the step that stops you copying the wrong part. A name forces you to decide which piece of the ad is the idea, and everything outside the name becomes free to change. "The owner on a real roof" can be shot on a different roof, in different weather, with a different offer, in fifteen seconds instead of thirty. What it cannot be is the owner at a desk, because that is the version you already know loses.
It is also worth checking who the winner won with before you finalize the name. Open the ad's results and use the Breakdown menu to split them by age, gender and placement. Sometimes a winner is winning everywhere. Sometimes it is winning with one slice, homeowners over 55 on the Facebook feed, say, and doing nothing much with everyone else. That changes the name to "the owner on a real roof, for older homeowners," and it changes what you make next. Denney has noticed the same pattern with her clients: audiences over 55 tend to respond better to stock footage than to the phone footage that works so well on younger ones. The audience is sometimes part of the answer.
Change One Thing and Let the Market Answer
Now you prove it, and the proof is cheap. Build two new ads that are identical in every way except the suspect. Same script, same offer, same length, same caption. One carries the suspect and one does not.
For the roofer, that means the owner delivers the same thirty seconds twice: once on a customer's damaged roof, once in the driveway in front of the truck. If the roof version wins again, the suspect is confirmed and you know something real about your market. If they come out even, the roof was not the reason and you move to the next row on the grid.
The professionals do exactly this. On one of Denney's concepts, an ad that calls out an industry villain, she found that who the villain is makes the biggest difference in performance, so that is the one thing her first round of variations changes. She also points to a brand that handed the same script to two very different kinds of creators, which is a clean test of the person on screen and nothing else. And Cottrell's own teardown had a cleaner example than the protein shake: a soap brand ran the same static layout across several scents, and one scent outperformed the rest. Same design, one variable: the scent. That result means something precisely because nothing else moved.
The mechanics of giving both ads a fair run, so Meta does not simply pick one on day two and starve the other, are covered in how to force Meta to give a new ad a fair shot. On a small budget, be honest about what the test can do. You are not going to get a laboratory result from a few hundred dollars. You are looking for a clear direction on the number that matters, over long enough that one lucky day cannot decide it. A clear gap is an answer. A narrow gap is also an answer: it is telling you the suspect was not the reason.
And keep the suspect a real decision. "The roof versus the driveway" is a question about what makes people trust a roofer. "Red button versus orange button" is not a question about anything.
Write Down Why It Won, or You Will Pay to Learn It Again
Every business that runs ads for long enough has discovered the same winning idea two or three times, because nobody wrote it down the first time. The ad wore out, the person who made it moved on, and a year later someone paid to learn it again from scratch.
Keep a winners log. It does not need software. One entry per confirmed winner, five lines:
- Name: the owner on a real roof.
- What it beat: four ads in the same spring batch, about $38 per booked inspection against $88 to $140.
- The suspect: the opening, on a customer's damaged roof. Not the owner, not the format, not the offer.
- The test: same script, roof versus driveway. Roof won again.
- Keep in every future brief: open on real work, on a real customer's property.
That last line is the point of the whole exercise. It becomes the first line of the next creative brief, which is how a brief stops being a request for "something that performs" and becomes an instruction with evidence behind it. We have written about why nobody can make you a good ad from a bad brief, and a winners log is the cheapest way to make every brief a good one. It is also what makes variations work. Once you know the idea is "the owner on a real roof," you can build a whole family of ads around it, different roofs, different damage, different hooks, and the steps for turning one winning Meta ad into several without resetting your results pick up from there.
This is the part of creative work we spend the most time on in creative production: not making the ad, but reading what the last round of ads told us and carrying it into the next one. Making ads has never been cheaper. Knowing which part of an ad to keep is what still costs money if you get it wrong.
The winning ad itself will not last. Every ad wears out eventually, and you usually cannot see it happening until the numbers slide. If all you kept was the ad, that is the end of it. If you kept the reason, you just make the next one.
When You Genuinely Cannot Tell
Sometimes you will do all of this and still not know. The budget was small, the results were thin, or the batch changed everything at once and the grid has no clean row. That is fine, and it happens to people with far bigger budgets than yours.
When it happens, do two things. First, leave the winner alone. Do not "fix" it, freshen it or swap its photo because you have a theory. An ad that is working for reasons you cannot see is still working, and every change you make to it is a guess with real money on the line. Denney's apology ads are the proof: she could not explain them, and they kept scaling anyway. The job there was to keep running them, not to improve them into something that fails.
Second, design the next batch as the experiment you could not run on this one. Build new ads that each carry one clearly different idea, and know before launch which idea each one carries. Next time something wins, the grid will have a clean row, and you will know what you are keeping.
The owners who get steadily better at advertising are not the ones who find more winners. They are the ones who find out why, write it down, and make every batch start where the last one left off. If you would like a hand reading what your own ads are telling you, that is exactly the work we do. And if you would rather do it yourself, the method above is all of it: line up the losers, find the row, change one thing.
