Most ads you make will lose.

That isn't a creative problem. It's the math of advertising. A few winners end up carrying most of an account's spend, and the only way to find them is to test a lot of ads that don't make it.

Creative testing is how you find those winners without burning the budget on the losers.

The short version

Creative testing is a system for putting new ads in front of real people, spending just enough to read the result, and moving money to what works.

Test budget per new ad ≈ 2 × target CPA

  • Test one thing at a time. A new concept, or a new hook on a proven concept. Never both.
  • Read the creative before the cost. Hook rate and click-through rate speak long before cost per purchase does.
  • Spend enough to judge. About 1 to 1.5x your target CPA before you decide, and cut at 2 to 3x with nothing to show.
  • Test on a schedule. New ads go in every week, whether or not anything looks broken.

Why testing matters more than ever

Meta now decides who sees your ad largely from the ad itself. Its Andromeda system shortlists ads for each person based on what the ad is, so different ideas reach different people. I explain how in Meta Andromeda, explained.

Every winner also wears out. Frequency climbs, click-through rate sags, and the cost per result creeps up. That's ad fatigue, and the only cure is a replacement that was already in testing.

An account that stops testing has already scheduled its decline. It just hasn't seen the date yet.

The testing ladder

Not every test is the same size. I think of it as the testing ladder: three rungs, from biggest change to smallest.

1. Concept tests. A new angle: a different problem, a different desire, a different person speaking. These find new territory. Most of them lose, and the ones that win can change the whole account.

2. Hook tests. Same concept, same body, new first moment. The cheapest test in advertising, because the hook decides whether anyone sees the rest. More on what wins in what hundreds of ad creatives taught me about hooks.

3. Format tests. The same idea as a static, a short video, a carousel, a different crop. These squeeze more out of territory you already know works.

New concepts find winners. Hooks and formats keep them alive.

Climb down the ladder when a concept shows a pulse. Climb back up when a concept runs dry.

How to set up a test on Meta

There are three common ways. All of them work if you respect the budget rules below.

1. Meta's built-in testing. Ads Manager can run A/B tests that split the budget and the audience between versions. Meta's own A/B testing page says you can test up to five ad variants at a time. It's the cleanest read, and the slowest to set up.

2. A dedicated testing campaign. New ads go into their own campaign with a fixed budget. Winners graduate to the main campaign. This keeps testing spend predictable and stops new ads from stealing budget from proven ones.

3. Straight into the main ad set. New ads launch next to the current winners and Meta decides where the money goes. Fastest, but new ads often get starved because Meta favors what's already working.

Whichever you pick, keep the number of ad sets small. Meta's delivery needs data to settle, and splitting a budget too thin keeps everything stuck in learning. The Facebook ad budget calculator shows how much each ad set needs at your cost per sale.

How much to spend, and for how long

Judge a test by spend, not by days.

My rule from the kill rules: don't judge an ad before it has spent roughly 1 to 1.5x your target CPA. If it reaches 2 to 3x with nothing to show, it's done.

So budget about 2x target CPA for every new ad you test.

Ads you can test per month = monthly test budget ÷ (2 × target CPA)

With a $2,000 testing budget and a $50 target CPA, that's about 20 new ads a month. Halve the budget and you halve the number of shots.

What to measure, in order

Cost per purchase needs volume before it means anything. The creative tells on itself much earlier.

1. Hook rate (video): did people stop? Work it out with the hook rate calculator.

2. Hold rate (video): did they keep watching?

3. Click-through rate: did the promise land hard enough to earn a click? For statics, this is the first read.

4. Cost per result: the final verdict, once there's enough spend behind it.

An ad with a strong hook and click-through rate but a weak early CPA often needs more time or a better landing page. An ad nobody stops for almost never recovers.

Reading the results

Kill concepts, iterate variations. If several versions of a concept all miss, the concept is dead. Stop making more of it.

If one version shows a pulse, the angle may be right and the execution wrong. That's a hook test, not a kill.

When something wins, don't just scale it. Start its replacement. New hooks and formats on a winning concept are nearly free and buy you weeks.

The mistakes I see most

  • Changing five things at once. If the hook, offer and audience all change, you'll never know what worked.
  • Calling a winner on day one. Early numbers swing. Wait for the spend.
  • Testing ten copies of one idea. Near-identical ads reach the same people. They're one test, not ten.
  • Testing only when results drop. By then you're already behind. Test on a schedule.
Pool rule 005Every test has one job.

Quick answers

How long should I run a creative test on Facebook? Until each ad has spent about 1 to 1.5x your target CPA. Days matter less than spend.

How many creatives should I test? As many as your testing budget can fund at about 2x target CPA each, with new ones entering every week.

Should I test creatives in a separate campaign? It's the easiest way to keep testing spend predictable and stop new ads from being starved. Many buyers do. It isn't the only way.

What is a winning ad? One that beats your target cost per result with enough spend behind it to trust the number, not one that had a good first day.

Can I test statics and videos together? You can, but judge them on their own first reads: hook rate for video, click-through rate for statics.

The one-line version

Most ads lose. Testing on a schedule, with enough spend to read the result, is how you find the few that pay for everything.