Most ad accounts don’t have a creative problem. They have a creative testing problem. Somebody makes six ads, launches all six in one ad set, watches Meta spend most of the budget on one of them inside two days, and calls the survivor the winner. Nothing was learned, because nothing was isolated. Next month the process repeats with six new guesses.
Creative is now the biggest lever in the account. Meta’s Advantage+ products have absorbed most of the targeting decisions you used to make by hand, so two brands running the same budget and the same audience settings win or lose on what’s in the ad. That makes creative testing the operating rhythm that decides your cost per lead, and that rhythm needs a written framework.
Here’s the one we run. Five parts: a hierarchy of what to test, a structure that isolates it, a spend rule that says when a result is real, a scorecard that reads the right metrics in the right order, and a loop that turns winners into the next round.
Test concepts before you test colours
The most common mistake in creative testing is testing small things first. A brand launches four versions of the same ad with different button colours or headline tweaks and learns, at best, that one tweak edged the others by a few percent. The angle underneath all four was never questioned.
Order matters. Start at the top:
- Concept: the core idea or angle. Price, speed, proof, urgency, a founder story, a product demo. Two concepts can differ in cost per result by two or three times. Nothing below this level moves the number that much.
- Format: the same concept as a UGC video, a static image, a carousel, or a text-on-screen edit. Format changes hook rate more than anything except the concept.
- Hook: the first three seconds of a video or the first line of a static. Same concept, same format, different opening.
- Execution: copy edits, colours, music, length. Cheap to make, low ceiling.
Run the first round at concept level with wide gaps between the ideas. Once a concept wins, spend the next rounds finding its best format and hook. Execution tweaks come last, if at all.
Isolate one variable per test
A test only teaches you something if you know what changed. Put six ads in one ad set and let Meta’s delivery pick, and you’ve learned which ad the algorithm preferred in its first 48 hours, which is a different question from which ad converts best over a month.
Our structure for a testing campaign:
- One testing campaign, separate from the scaling campaign, so the tests don’t compete for budget with proven ads.
- One ad set per concept in the first round, identical audience and budget, so the only difference between them is the creative. Meta’s built-in A/B test tool does this for you and stops the ad sets overlapping.
- Two to four ads per ad set at most.
- A fixed budget on each. Don’t let the algorithm reallocate mid-test.
The scaling campaign runs your proven winners with broad targeting and most of the money, the way we set it out in the Meta ads system post. Testing sits beside it, usually with 10 to 20% of total spend. Winners graduate across and losers get switched off, so the scaling campaign never sees an ad that hasn’t earned its place.
Decide when a result counts before you launch
The second mistake is calling a winner too early. An ad with three leads at $12 each looks better than one with two at $18, and neither number means anything yet. At that volume the difference is noise.
Set the threshold before the test starts and write it down. Two rules we use:
- Enough conversions per ad. Meta’s own learning phase needs around 50 optimisation events per ad set within seven days before delivery stabilises. That’s a fair stand-in for “enough signal”. If your budget can’t get each ad near that number, test fewer ads, or optimise for a higher-volume event further up the funnel (landing page views, add to carts) and validate the winner on leads afterwards.
- A minimum spend per ad before judgment. Our rule of thumb is two to three times your target cost per lead, per ad, before anything is switched off. A $30 target means every ad gets at least $60 to $90 to prove itself. Cut earlier and the eventual winner may never have received its budget.
One calculation keeps the budget honest. Say the target CPL is $30 and you want four concepts tested. Four ads at $90 each is $360 as an absolute floor, and closer to $1,500 a week if you want each ad near the 50-event line. If that number is uncomfortable, test two concepts. A clean test of two beats a mess of six.
Read the scorecard in order
Cost per lead is the verdict, but it’s a slow verdict and it can’t tell you why an ad lost. Four metrics read together explain the loss, and they should be read in this order, because each one is a gate for the next.
- Hook rate: three-second video plays divided by impressions. It tells you whether the opening stopped the scroll. A low hook rate with a strong landing page still means a dead ad, because nobody watched long enough to hear the offer.
- Hold rate: ThruPlays (or 15-second views) divided by three-second plays. This is whether the middle of the ad kept them. Strong hook and weak hold usually means the opening promised something the body didn’t deliver.
- Outbound click-through rate: whether the ad and the offer earned a click. Good hold and weak CTR usually points at the offer or the call to action.
- Landing page conversion rate: whether the page kept the promise the ad made. Strong CTR and a weak conversion rate means the ad did its job. That’s a landing page problem, and swapping the creative won’t fix it.
Reading in sequence tells you where to work. An ad failing at the hook needs a new opening; its concept may be fine. An ad winning everything up to the page needs the page fixed before you spend another cent on video.
Benchmarks vary too much by industry and format to publish one number. The useful comparison is your own account: rank every ad in the round on each of the four, and the pattern is usually obvious within a minute.
Turn one winner into the next round
A winning concept is where the work starts. Meta creative fatigues: frequency creeps up, CPMs rise, and an ad that held a $25 CPL in month one does $40 by month three. The framework only beats guessing if it produces a steady supply of new winners.
Iterate on the winner in the same hierarchy:
- Same concept, new hooks. Three new openings on the winning body is the cheapest test in the account and often the highest-return one.
- Same concept, new format. If a UGC video won, cut a static from its strongest frame and a carousel from its three main points.
- Same concept, new messenger. Different creator, different age, different setting. People who scrolled past the first face may stop for the second.
- Then a fresh concept round, so there’s always something in the pipeline when the current winner fades.
Cadence depends on spend. An account spending under $5,000 a month can run one test round every two weeks. Above $20,000 it’s usually a weekly rhythm with four to six new pieces of creative, because fatigue arrives faster at that volume. Either way, put a date on the next round the day this one launches.
Keep a testing log. One row per ad: concept, format, hook, launch date, spend, the four scorecard metrics, CPL, and the decision. After three months that log is worth more than any benchmark report, because it tells you what your audience stops for, in your account, with your offer.
What the framework changes
Guessing produces one winner every few months by chance, and the account lives off it until it fatigues. A framework produces a steadier stream of winners on a schedule, with a written reason each time, and the cost per lead stops swinging with the mood of the algorithm.
It also changes what you ask a creative team for. Instead of “make us some ads”, the brief becomes “four concepts, one format each, three hooks per concept, by Friday”. That’s a brief a team can actually deliver against, and it’s what turns creative from a monthly bill into a system you run.
Want your creative tested on a schedule instead of a hunch?
Creative testing is one of the systems we build and operate alongside paid media. See how we run creative and brand as a performance system, or book a call and we’ll map a testing structure to your current spend.