Meta tests on a small budget: one ad set with many creatives, or one per audience?

At $30 a day I'd skip the audience split, because it mostly burns money. Meta's learning phase wants roughly 50 optimization events a week per ad set, and a few hundred clicks spread across three ad sets means none of them gets there. Put the three creatives inside one broad ad set and let delivery sort out the winners. Creative does most of the targeting now anyway.

The catch is you have to read results at the ad level, not the ad set average, or a good angle can hide behind a weak one. Where do you draw the line for when a split actually makes sense?

Agree on the broad ad set. One test before you split: would you actually write a different ad for this audience? If retargeting visitors need a different offer than cold users, a split makes sense, because one creative can't speak to both. If the only difference is age or interest, let delivery sort it out. Then do the math. At $30 a day with a $15 cost per result, you get about 60 results a week, which is enough to fund one ad set's learning phase, not three. Only split when each group can clear that bar on its own.

The math holds, but the 50-event guideline is per ad set, and a retargeting ad set often can't hit it because the pool is so small. Take 2,000 site visitors a month. Even with a good cost per result, that group falls well short of 50 a week, so it never really leaves learning. With small pools, I'd only split when the offer genuinely differs, and even then it's better to widen the audience than to squeeze the budget.

Agreed on the pool problem, but the optimization event matters just as much. A retargeting ad set optimizing for purchase at 8 a week will sit in learning no matter how wide the audience is. Switch the same audience to add to cart and Meta may get enough volume to exit learning. The catch is you then have to check that purchases still follow, so compare the ad set's purchases in GA4 against what Meta reports. A cheap event that doesn't convert will look great and tell you nothing.

Switching to add to cart isn't the fix, I'd say. Meta will find cheap add-to-carts easily, but with a small retargeting pool that often means you're optimizing toward people who fill a cart and leave. Eight purchases a week is a thin signal, but it's the one that actually matches what you're paying for.

At that size I'd widen the source first. Try 180 days of visitors instead of 30, or add engaged video viewers. Your GA4 check is a good safeguard, but all it tells you is whether the cheap event is lying. It doesn't tell you whether it's the right goal.

I'm not sure the 180-day widening holds up. For most stores, a visitor from six months ago is a different buyer than one from last week. They've already decided or moved on, so the purchase rate per visitor drops, and Meta ends up paying to reach fairly cold people under a retargeting label. You get more volume, but it's weaker volume. If the purchase pool is too thin, I'd keep the 30-day window and put that cash into a broader prospecting ad set, which has a pool big enough to learn on its own.

Partly agree, and it really comes down to how long your buyers take to decide. If most purchases happen within a week or two of the first visit, Iris is right: a 180-day visitor is mostly someone who already moved on. But for a $900 sofa, or a course people sit on for a month, a 30-day window misses people who were still deciding.

The easy way to settle it is to look at the lag between first visit and purchase in GA4 over the last few months. If the typical gap is under a week, keep the window short. If it's several weeks, the longer window may be catching buyers you'd otherwise lose.