Your ROAS dropped and the first instinct is to blame Meta. It's rarely Meta. In almost every account we audit, the real cause is one of four things, all of them fixable, and none of them require waiting for the algorithm to "fix itself."

Here's the honest breakdown of where ads actually break, and what to check for each one.

One ad is probably carrying your whole account

Open Ads Manager, sort by conversions at the ad level over the last 30 days, and look at your top ad. If it's responsible for more than 15% of total conversions, you have concentration risk. Above 25%, one ad is holding the account together on its own.

Now check its frequency. Above 3.0 and it's already fatiguing, whether your CPA has noticed yet or not. Meta will keep feeding that ad to the same high-converting audience because that's the highest probability play right now, not because it's good for you long term. It performs well for a while, then one morning CPA doubles overnight and nothing in the account settings has changed. The ad just died of natural causes, and because it was carrying a huge share of conversions, the whole account feels it at once.

Most founders assume they're diversified because they're running ten ad sets. Then they check and eight of those ten are pushing the exact same two ads.

Volume isn't the same as diversity

More ads doesn't mean more ideas. Twenty crops of the same image are one idea wearing twenty outfits, and the algorithm can tell. Real diversity comes from changing the actual angle, the reason someone should care, not just the format or the thumbnail. Test different reasons to buy before you test different ways to package the same reason.

A simple gut check, scroll your live ads quickly the way a customer would. If you can't tell them apart at a glance, Meta can't either, and no amount of separate ad IDs will fix that.

The failing ad might be dying at a different stage than you think

ROAS tells you an ad isn't working. It never tells you why, and three separate metrics answer three separate questions.

Thumbstop ratio shows whether people stop scrolling at all, your earliest signal of relevance. Hook score shows whether the opening holds attention once they've stopped, telling you if the angle is actually landing. Click to purchase rate shows whether that attention turns into buyers, and when it's low, the honest answer is usually audience mismatch, not spend.

Match the fix to the exact stage that's weak. A scroll problem needs a new opener. A hold problem needs a new angle. A conversion problem needs a new audience entirely, since more budget just buys more of the wrong click.

You might be picking the format before you know the job

"We need more UGC" is not a strategy, it's a format with no reason attached. Every format exists to do a specific job. Proving something needs a demo. Explaining something needs room to build the argument. Convincing someone who's still on the fence needs a longer form, a VSL or advertorial style creative. Skepticism needs someone else doing the talking, a real testimonial. One strong claim needs almost nothing, a static will do it. A real scenario needs a skit or POV that can actually show it happening.

Name the job before the format gets chosen, and the format stops being a guess.

The actual fix

Run the concentration check this week. Test three to five genuinely different concepts, not variations, every week. Diagnose a failing ad by stage before you touch the budget. And ask what job the message needs to do before anyone opens Canva.

None of this is the algorithm punishing you. It's usually just the account running thinner than it looks on the surface, and every one of these checks takes minutes, not a media buyer overhaul.