Why do a few ads end up with most of your ad spend?

What a power law means for creative

Most people plan creative as if results are normally distributed. Make twenty ads, expect a few good ones, a few bad ones, and a cluster of average ones carrying the account. That is not what happens.

What actually happens is that a very small number of ads absorb an enormous share of spend and return, and almost everything else does nothing. The distribution has a long tail, and the tail is where the money is. Your job is not to raise the average. Your job is to produce enough attempts that an outlier shows up, then get out of its way when it does.

This changes what a batch of creative is for. A batch is not a set of ads you expect to work. It is a set of lottery tickets bought at a known price, where the payout is concentrated in one or two of them.

The numbers

The working thresholds:

• –Your top 1 to 2% of ads should be doing 50% or more of your total spend. If spend is spread evenly across your library, the algorithm is not being allowed to concentrate, and you are funding mediocrity.
• –A blended hit rate of 10 to 15% is healthy. That is the zone.
• –A blended hit rate above 20% means you are testing too safely. Counterintuitive, but a high hit rate says your ideas are close to what already works, so you are not buying enough variance to find the next outlier.
• –Around 85% of creative will fail. Accept it up front and it stops feeling like a series of disasters.

How to apply it: the 50/50 split

For brands spending $250K+ per month, split production in half.

Half the batch is variations: new versions built from what your data already says works, drawn from winners, top angles, and proven hooks. Aim for a 20% hit rate here. These are your safe bets and they keep the account fed.

The other half is swings: ideas from outside the account entirely. Competitor concepts, customer calls, brainstorms, organic content, formats that feel risky, producers with a different style. Aim for a 10% hit rate. These are what actually find the outlier.

Blended, that lands you in the 10 to 15% zone. If your blended rate climbs above 20%, shift budget from variations to swings. You are being too safe.

A worked example

A founder came in spending $150K per month on Meta at a $110 CPA. They were proud of a 30% hit rate and thought that was the win. It was the problem: a hit rate that high meant almost every ad was a minor variation of the last one, so nothing ever broke out.

The first phase was media buying alone, letting winners actually take budget instead of capping them. That cut CPA by 48%. The second phase was creative, once the swings started landing. That took another 50% off. Total reduction was 74%.

Spend tripled to $500K per month, and their top 1 to 2% of ads ended up doing more than half of it. The company later sold for over $100 million.

The mechanism is worth being precise about. Nothing here required a higher hit rate. The hit rate went down. What changed is that when an outlier appeared, it was allowed to consume budget instead of being capped for the sake of not putting all the eggs in one basket.

When it does not hold

Caveats, because a benchmark applied blindly is worse than no benchmark:

• –Below roughly $250K per month, the split is less useful as stated. At lower spend you are not running enough ads for the distribution to reveal itself, and each test takes longer to reach a readable result.
• –A hit rate is meaningless until you define a hit. Fix the threshold first, whether that is spend absorbed, CPA against target, or a significance test, and hold it constant. Otherwise the number moves because your definition moved.
• –The 1 to 2% figure describes how spend should end up distributed. It is not a target to force by manually pushing budget into two ads you like.
• –The power law depends on the algorithm being allowed to consolidate. Capping winners to spread risk, or splitting budget evenly across ad sets, flattens the distribution and the whole model stops applying.
• –New accounts and new offers behave differently. Before there is any performance history, everything is a swing, and the variations half of the split has nothing to draw on yet.

What to do with it

Two questions answer whether this is working in your account. What share of spend is going to your top 1 to 2% of ads, and what is your blended hit rate against a fixed definition of a hit? Plot the account on a log-spend scatter and the concentration is visible immediately: either a few ads have pulled away, or everything is bunched together and nothing has been allowed to win.