Meta Advantage+ creative enhancements: leave them on or test them against originals?

Meta's creative enhancements are on by default for a lot of ad setups now, and the list keeps growing: music, text tweaks, brightness, crops, and so on. The question is whether to leave them running or switch them off. I'd be careful with the ones that change how the product looks or add words to the image. If a skincare brand's shade gets rendered slightly off in an auto-enhanced version, that's a return waiting to happen, and a higher CTR won't make up for it. Crops and minor brightness changes are usually low risk, so I'd let those run as a test against the untouched original. Keep the originals as your control, otherwise you can't tell whether an enhancement actually earned anything. Has anyone seen enhancements lift CTR without moving cost per purchase?

CTR is a weak yardstick for this. Brightness and crop changes can pull in curious clicks that never buy, so I'd judge the split on cost per purchase and let it run until each variant has enough conversions to mean something. Are you comparing at ad level or running a proper split test?

Ad-level comparisons usually mislead here, because delivery drifts toward whichever version gets early clicks. Use Meta's A/B test tool and don't call a winner until each side has about 50 purchases. If cost per purchase is within 10 to 15 percent at that point, call it a tie and keep the original.

50 purchases makes sense in principle, but a skincare brand doing maybe 20 a week would need a month or more to get there, and by then the creative might be tired. In that case I'd test one enhancement at a time against the original instead of the whole bundle. If the bundle wins, you can't tell which change did it. A crop or brightness tweak could be doing all the work while the text overlay quietly hurts.

Fair, I was too strict with the 50. It's a rough floor for catching small gaps, not a magic line. At 20 purchases a week you can only reliably see a difference of around 30% or more in cost per purchase, so one-at-a-time tests only make sense for changes you expect to be big. For smaller ones, use add to cart as an early read and confirm on purchases. I'd still run the whole bundle first, because if it doesn't move cost per purchase, you've saved yourself from splitting it.

I'd test the bundle against the original first, then split out only the changes that touch the product or add words to the image. Those are the ones that can drive returns, and returns don't show up in cost per purchase until weeks later, so they deserve their own check even when the sample is thin.

Agreed on splitting out anything that touches the product. I wouldn't trust the split until you've looked at the rendered creative in each placement, though. Enhancements can differ between Feed, Stories and Reels, so a shade shift on one placement can get averaged out in the totals. Pull a few previews per placement, then judge returns only on orders from the enhanced ad IDs over a 30-day window, since that's roughly how long returns take to show up for skincare.

Honestly it comes down to whether your orders carry the creative identifier at all. Meta's ad ID often never makes it onto the order record, so you end up matching on UTM parameters, and returns get logged against the order, not the click. If the store can't pass the ad or creative name through to the order, returns just roll up by campaign, and that blends enhanced and original ads together. Confirm that link exists before you trust any returns split.

Meta's URL tags can pass {{ad.id}} and {{ad.name}} into your UTMs, so the ad identifier reaches the order without depending on the store. The catch is that if the enhancements sit inside a single ad, the enhanced and original versions share that ID, so you can't split returns between them. To get a clean read, run them as separate ads or as separate test arms.