Iterate Winners, Bury Losers: The Only Iteration Strategy That Survives Meta's Algorithm

I'll say the unpopular thing first. Most of the "iteration" your team is doing right now is wasted effort, and a fair chunk of it is actively hurting you.

I know that's not what you want to hear when you've just spent a fortnight tweaking headlines and swapping the first three seconds of your best video. But I've come round to a hard view on this, and it's changed how we extend winners and how we walk away from losers.

Here's the short version. Meta is no longer judging your ads one at a time the way you think it is. It's quietly grouping them, and it's passing the reputation of the old ad onto the new one. Once you understand that one mechanic, the whole "should I iterate this?" question gets a lot simpler, and a lot less democratic.

Let me walk you through how it actually works, then exactly what I'd do with it.

What's actually happening under the hood

For years we all treated an ad account like a clean slate per ad. Upload a new creative, it gets its own fresh shot, the algorithm learns it from scratch, may the best ad win. Tidy and fair.

That's not what's happening anymore.

Meta's delivery system clusters creatives that look similar to it into the same bucket. People in the industry have taken to calling this an entity, and the detail that matters is this: the grouping is mostly visual. If two ads look roughly alike, the system tends to treat them as the same thing.

And here's the part that changes everything. When it groups a new ad in with an old one, it doesn't start the new one fresh. It hands the new ad all the prior learnings of the old one. The good reputation, or the bad reputation, gets inherited.

So when you take a concept that flopped and "iterate" it - same footage, same setup, a new hook bolted on the front - you are not giving Meta a clean new ad to fall in love with. You're handing it the same entity with the same baggage. You've inherited the loser's bad name and asked the algorithm to forget it ever happened. It won't.

I think this is why so many teams feel like they're sprinting and going nowhere. They're producing volume, but a lot of that volume is just the same handful of entities wearing slightly different hats.

Why iterating a loser almost never saves it

Let's be blunt about the failed concept first, because it's the cheaper lesson.

If an ad concept tanked, and you make a small change and re-upload it hoping this version finds new life, you're betting against the system. The new iteration gets assigned the negative priors of the same concept. You're trying to launder a bad reputation with a haircut, and the algorithm sees straight through it.

I'm not saying nothing about that concept can ever work. I'm saying the small iteration is the wrong tool. A new hook on a dead video isn't a new ad to Meta. If you genuinely believe the angle has legs and the execution let it down, that's not an iteration job. That's a from-scratch rebuild: new footage, new setting, a visually different ad that won't get clustered back in with the corpse.

So my rule for losers is unsentimental. Bury them. Don't tinker, don't nurse, don't "give it one more variation". Kill the concept, learn what you can, and put the production hours into something the algorithm hasn't already decided it dislikes.

To put a number on the waste: say a team ships ten "iterations" in a week and seven of them are minor reworks of concepts that already failed. That's seventy per cent of the week's output handed straight back to entities Meta has already filed under "no thanks". The fix isn't working harder. It's pointing the same hours at the right targets.

The bit almost everyone gets wrong about hooks

This is where I have to break some bad news about a tactic that used to be gospel.

For a long time, "shoot one video, cut five hook variations off the front" was considered best practice. Same thirty-minute founder interview, five different opening lines, call it five ads. It felt like free volume.

In reality, to the system, those five are very likely one ad. You've changed the first few seconds of a video that's otherwise identical, and the grouping is visual. Meta is not handing each hook its own clean learning curve. It's recognising them as the same entity and pooling everything.

That doesn't mean hooks don't matter. The hook is still what makes a human stop and a customer convert, so it matters enormously for performance. It just doesn't do what you thought it did inside Meta's delivery. It's a conversion tool, not a way to manufacture five genuinely distinct ads.

Once that clicked for me, I stopped letting teams count hook variants as creative volume. They're polish on one entity, not five shots at the board.

How to extend a winner so the algorithm treats it as new

Right, the good part. You've got a winner. It's working, you want more of it, and you don't want to accidentally squander its hard-won good reputation. How do you ride it properly?

The honest answer is the inverse of the loser rule. With a winner, inheriting the priors is a gift. The concept is roughly working, the algorithm likes the entity, and you want to extend that as far as it'll go. So you iterate on purpose - but you iterate to be as visually different as you can while keeping the core idea intact.

That sounds contradictory, so let me make it concrete. Messaging is not a big part of how Meta groups things. Visual is. So the play is to keep the winning message and structure and change the visual setting as hard as you can.

Take a founder-talking-to-camera ad that's winning. Here's how I'd extend it:

• Same script and structure, filmed in a car instead of the office.
• Then the same thing again, filmed walking through a warehouse.
• Then again, at a kitchen bench, different shirt, different time of day.
• Then a version where a different person delivers the same lines entirely.

Each one keeps the thing that's working - the message, the beats, the reason it converts - while looking different enough that Meta is more likely to read it as a distinct creative rather than the same clip on a loop. You get the best of both: you're riding a proven concept and you're feeding the system genuine visual diversity instead of five near-identical cuts.

The contrast with the failed-hook approach is the whole point. Five hooks on one video is one entity pretending to be five. One winning script shot in five genuinely different settings is a real attempt at five.

A before and after, so this isn't just theory

Let me make this tangible with an invented but very typical example.

Picture a skincare brand spending around A$40k a month. Their best performer for three months was a founder piece, shot at her desk, explaining why most serums do nothing. Strong hook, real mechanism, the lot. Then it fatigued, the way every winner eventually does. CPA drifted from roughly A$30 up past A$45, and the obvious panic move was on the table: kill it and start again.

Here's the wrong version of what came next, the one I see constantly. The team cut four new hooks off the same desk footage and shipped them as fresh ads. Nothing changed. The CPA stayed ugly, because to Meta it was the same tired entity with new intros. They'd "made four ads" and launched zero.

Now the version I'd run. Same script, same beats, rebuilt visually. One take filmed in her car on the way to work. One walking through the lab. One at her bathroom mirror doing her actual routine. One handed to a second creator who delivers the identical lines. Four genuinely different-looking ads carrying the proven message.

In a case like that, you'll often see one of those new settings re-ignite the concept and pull the CPA back down towards where it started, because the algorithm gets something it reads as new while the underlying thing that made customers buy is untouched. I want to be careful here: that's the pattern, not a promise. Sometimes a fatigued concept is genuinely spent and no setting saves it. But far more often the concept had life left, and what was actually dead was the visual, not the idea.

That's the distinction the whole strategy hangs on. Fatigue is usually visual. The message can have months left in it if you stop strangling it with hook cuts and start re-shooting the setting.

The rule, stripped right down

Pull all of it together and it's almost embarrassingly simple:

• Losers: bury them. A small iteration on a failed concept inherits its bad reputation. Don't nurse it. Kill it and rebuild from scratch if the angle's worth it.
• Hooks: stop counting them as volume. Five hooks on one video is one entity to Meta, not five ads. The hook earns conversions, not distinct learning.
• Winners: extend them with maximum visual diversity. Keep the message and structure, change the setting as hard as you can, so the system reads genuine new creative instead of a near-duplicate.
• Fatigue is usually visual. Before you bin a winner, try the same idea in a completely different visual world.

None of this needs a bigger budget. It needs you to stop spending production hours on iterations the algorithm has already decided about, and start spending them where the priors actually help you.

If you want to know which of your current concepts are winners worth extending and which are losers you're quietly resurrecting every week, that's exactly the kind of thing a Signal/Noise Audit pulls apart. We go through your creative history and lay your fatiguing winners next to your zombie losers, so it's obvious which deserve a fresh visual life and which just need to be left in the ground.

So before your next upload day, I'd ask one question of every "iteration" in the queue: is this a genuinely different-looking shot at a winner, or am I just handing Meta the same entity again and hoping it forgets?