What counts as a win?

Analysing win rates, rocketship winners, moderate winners, and false starts.

Aug 19, 2025

Last week, I shared a preview of one of our forthcoming reports about upper funnel spend, and Fiona shared her current AI creative stack. That forthcoming report is going to be one of the first fully paywalled pieces we launch here. If you’re not yet a paid subscriber, you know what to do.

In Early Stage Growth this week, I’m looking into what really constitutes a win on Meta?

TL;DR
It’s difficult to know what to do when a new creative test ‘wins’. Do you double down? Does it reduce all CPA? How long do those ads last?

In our end of year report this year, we want to go into this in a lot more detail, but in the short term we want to share short term learnings to begin the discussion.

Things we’ll be covering:

What is a winner? How do we test?
Why do ad accounts need so many ads?
What are the types of winners – rocketships, maintainers, false starts?
What is a good rate of winners to losers?
How can I improve that rate?

Here we go.

How do you define a winner on paid social?

There are loads of schools of thoughts on how to run creative testing on Meta. There are those who throw everything into one mega ASC. There are those who test thousands of ads into heavily bidcapped ads and are content with most ads getting zero spend. There are those who us the Meta Experiments tool for each test. Some isolate every single ad in its own ad set.

There’s no perfect way. Everything in paid social is about tradeoffs.

So to understand how we make our decisions, we should first set out what we’re trying to do.

Why we test like we do at Ballpoint. The background thinking:

Ad volume is important to scale
I’ll get into this in more detail in section two, but at a high-level each ad is a chance to find a new audience. Therefore to find new customers, you need new ads.
You can improve probability of success through strategy
We are not a ‘throw everything at the wall’ agency: there are lots of agencies that find success with that method. But we aren’t, we think that strategy delivers better long term performance. It does that by increasing the probability of finding a winner.
Ability to learn improves strategy
Like any good growth process, we run insight → experiment → learning. Learning feeds the next insights. And so in our testing approach we want to optimise for ability to test.
We know that despite strategy, chance still plays a huge role
Therefore, we need to create an environment that allows us to factor chance in. We need to make sure we’re testing ads that don’t just follow what we expect to work.
Diversity AND volume are therefore needed
With that in mind, it’s not just about following what we know. We have to test lots that doesn’t fit a mould as well.
Pace is vital in startups
The most likely outcome for a business is that it will die. You will run out of money before you work all this stuff out. And so pace is vital.

Say it takes 100 tests to find the first answer. If you do 1 a week you need 2 years of runway to survives. If you do 3 a week, you need 8 months.

As a result of all of these things, we by default run tests:

Ads grouped by concept at an ad set level (variants within it, usually 1-5 depending on size)
Run on ABO. This is an intervention away from how Meta usually works by forcing it out of a CBO/ASC mindset. We opt to do that to improve our ability to learn (point 3)
Don’t use Experiments tool and don’t isolate ads too much (point 6 would mean that these ad too much cost and slow down overall learning).

All testing is about tradeoffs. This is how we think about our approach.

So what about these winners?

We compare ads for similar like for likes in the hero/scaling campaigns.

For EG, say

Ad set 1 has £44 CPA
Ad set 2 has £88
Ad set 3 has £56
Hero campaign is at £55

Ad set 1 is definite winner, promote the best variant(s).

Ad set 2 is fail of concept, but check if a variant won

Ad set 3 is likely a moderator as within a margin of error. Promote best variant(s)

Why do ad accounts need so many ads?

Meta has become increasingly good at pairing individuals with ads that are relevant to them. We think of this as ad-audience fit. A lot of this started last year, but Andromeda in particular improved this.

Everyone responds slightly differently to things.

There are some universalities. There are some ways of communicating which reach broad audiences.

But there is so much micro-nuance in our day-to-day.

We could have the exact same needs and problems to solve, but I just happen to hate black and white in imagery and you love it. Or maybe I just really hate jazz and so the background music made me swipe past, whereas you don’t care about it and ignored it.

Then there’s the fact that life changes daily. Maybe you woke up grouchy or your favourite F1 team just lost or your baby is teething or it got a bit dark and rainy when you opened Instagram and it’s supposed to be summer. There is no uniform human experience.

And so ad accounts need ads for all eventualities if you want to scale.

Let’s be clear, there are advertisers out there spending tens of millions per month on paid social. That’s what success looks like. If you really have product-market fit and product-channel fit, then that’s what your aspiration should be.

And part of getting there is ads for all permutations of all problems solved, all jobs to be done, all people, all stages of the funnel, all whims of their mood.

What are the types of winners – rocketships, moderates, small contributors?

We’re increasingly adding nuance to these categories, but in general we’ve got the following groupings:

🚀 Rocketships – ads which (1) heavily improve CPA compared to the account CPA, by 20-90% and then (2) go on to become the dominant spenders in the hero campaigns.
🥈 Moderates – ads which sit within a 10% bracket either side of account CPA. They then go on to remain in the account for some time but rarely become the biggest spenders.
🎗️ Small contributors – sometimes an ad will do well in testing, you promote it and it ends up spending less. Maybe it gets the odd purchase or two at a very low rate, or maybe low levels of spend and no conversions. These aren’t fails necessarily, just maybe their audience size is small.

What does good look like? What ratio can we expect of these ads?

The million pound question. “How many ads should we expect to be winners?”

Here’s a few key stats from a small data set of ours (blend across account sizes):

Concept win rate ranges from 22% – 46% with evenish distribution
At an individual ad level 49% of ads have better CPAs than the account average that week – but it’s closer to 30-35% that receive enough spend to be considered winners
2.3% of winners are rocketships this rises to 3.5% amongst bigger accounts – indicating higher success at determining the very best ads the more we learn
17% of winners scale well – i.e. they’re spending more in month two than month 1
40% of winners maintain the account – spend stays pretty consistent, indicating either incremental reach
40% of winners fatigue within 4-6 weeks – indicating low longevity

“Maintain” might look like a dreadful word.

And if you’re pre-product-channel fit then what you obviously need is rocketships and good scalers to be able to change your trajectory.

But post-PCF, you’re fighting multiple fights: creative fatigue (which maintainers fix), new audiences, and ability to scale (scalers & rocketships fix).

Note we will be re-running this analysis in much more detail for our end of year report, where we’ll be running this across £15m of annual ad spend.

How can I improve that win rate?

When we talk about doing paid social today, this is really what we mean. Media buying is still important, yes, but ultimately how well you do creative is how you improve that win rate.

Improving paid social today is about building a creative growth engine.

Start with Jobs to Be Done
Your best ads don't come from copying competitors or following trends. They come from deeply understanding the progress that customer is trying to make in their lives.
Test with intentional diversity
Each JTBD insight should create multiple creative angles. Test different hooks, formats, and emotional territories. Get rid of your own bias. Nothing kills an ad account faster than making ads look like only ads you like. “I like it” is not a good sign of an ad.
Document relentlessly
It’s vital. But without good documentation, you aren’t improving the learning rate, or creating future scalability.
Ensure at least 25% of ads are illogical
Chance still factors in a lot. Lean into it. Throw in ideas that are totally counterintuitive. Whether it’s a messaging line you don’t think will work or a different style of creator or editing style. Break your own rules.
Speed compounds learning
You don’t go to the gym once a week and except to get a sixpack. Consistent, high volume is vital. As a startup you’re almost definitely losing money. Velocity is what will save your runway.

Final thoughts

The most likely thing an ad will do is not reach anyone. But it’s through that process that you learn how to reach as many people as your product will allow.

Diverse ads buy you new audiences. Better ads improve performance within those audiences. Both of these are needed.

What are your rocketship win rates?

Early Stage Growth