Most Meta ad accounts don’t fail because the creative is bad. They fail because budget keeps flowing to ads that stopped working weeks ago. The fix isn’t a bigger budget or a new agency, it’s the discipline to read the numbers honestly, cut the losers, and pour fuel on the winners.

Here’s exactly how we did that inside a live account for a research e-commerce client we manage, using real numbers from a single 30-day window, so you can run the same teardown on yours.

The account at a glance

This is a research e-commerce brand that scales real budget: $40,000+ in spend at a roughly 3.7x blended ROAS, returning $150,000+ in tracked revenue. That’s a healthy blended number, but the blended number is a trap. It hides the fact that some ads were carrying the account while others quietly bled money.

When you only look at account-level ROAS, you’re averaging winners and losers together and calling it “fine.” The entire job of a media buyer is to break that average apart and act on what’s underneath. So we pulled a single 30-day window and lined the ads up side by side.

Three ads, three completely different stories

Inside the same account, over the same 30 days, sat three ads with wildly different outcomes:

• The loser: 0.68x ROAS. It returned just $47 in revenue for more than it spent, money poured straight down the drain. At sub-1x, every dollar it touched came back as less than a dollar.
• The winner: 28.9x ROAS. It generated $2,005 in revenue on just $69.39 in spend, at a $17.35 cost per purchase. That is not a typo, one ad turned about seventy dollars into two thousand.
• The workhorse: 5.10x ROAS driving $3,500 in revenue across 44 purchases at a $15.60 cost per purchase. Not the flashiest multiple, but the biggest absolute contributor and the ad built to scale.

Three ads, one account, one window. If you only looked at the blended 3.7x, you would never know any of them existed.

One ad turned $69.39 into $2,005 at a 28.9x ROAS. Another beside it returned $47 at 0.68x. Same account, same window, opposite outcomes, that spread is the whole game.

How to actually read the data

Cost per purchase and ROAS tell you whether an ad works. The upstream metrics tell you why, and whether a struggling ad is fixable or just done. When we audit an account, we read the funnel top to bottom:

1. CPC and CTR (the top of the funnel)

Cheap clicks and a healthy click-through rate mean the creative and audience are resonating, people see the ad and click. If your CTR is under ~1% and CPC is climbing, the problem is creative or targeting, and no amount of budget will fix it. Diagnose here before you touch spend.

2. Add-to-cart rate (the middle)

This is where intent shows up. A strong add-to-cart rate says the traffic is qualified, not just curious. A high CTR with a weak add-to-cart usually means a landing-page or offer mismatch, not an ad problem, and it tells you to fix the page rather than kill the ad.

3. Cost per purchase and ROAS (the bottom)

These are your verdict metrics. But you only trust them once you’ve confirmed the funnel above them is healthy and you have enough conversions for the number to be real. The 5.10x workhorse earning across 44 purchases is durable. A good ROAS built on one or two conversions is noise, which is exactly why the 28.9x spike, real as it is, gets read as a creative signal to clone rather than the whole budget’s new home.

Killing the loser, scaling the workhorse

Once the data was clear, the moves were simple. We killed the 0.68x loser outright, it was returning $47 against more than it cost, with no funnel signal worth saving. Then we concentrated budget behind the 5.10x workhorse at its $15.60 cost per purchase, the ad already proving it could convert at volume. This is the single highest-leverage action in paid social, and most advertisers never take it because pausing a running ad feels like a loss.

The 28.9x winner plays a different role. A multiple that high on that little spend is a signal, not a scaling target, push $5,000 through it overnight and the efficiency almost never holds. So we treat it as a creative tell: what angle, hook, and audience produced it, and how do we clone that into the workhorse’s budget.

A few rules we follow so we don’t fumble the reallocation:

• Shift budget in steps, not leaps. Scaling a winning ad set 20-30% at a time keeps Meta’s optimization stable. Doubling the budget overnight resets the learning phase and can crater performance temporarily.
• Watch cost per purchase as you scale, not just ROAS. As spend increases you reach a broader, colder slice of the audience, so expect some efficiency drift. Scale until cost per purchase approaches your break-even ceiling, then hold.
• Don’t kill on one bad day. We pause on a trend across a meaningful window and enough conversions, not a single-day dip. Daily volatility is normal; a sustained sub-1x result like the 0.68x loser is a decision.
• Clone your spikes, don’t just admire them. A 28.9x ad is a lesson in what your audience responds to. We keep a small test budget running new creative built off that angle so there’s always a next winner in the pipeline.

Why this only works inside a real system

A teardown like this depends on trustworthy data. If your pixel is misfiring, your Conversions API isn’t set up, or purchases aren’t attributing correctly, every number above is fiction, and you’ll scale the wrong ad with total confidence. That’s why we treat tracking and paid ads as one connected system, not separate line items. Clean GA4/pixel/CAPI tracking feeds honest reporting, honest reporting drives budget decisions, and budget decisions compound month over month.

The winner-and-loser teardown isn’t a one-time cleanup. It’s the weekly rhythm of a well-run account: read the funnel, cut what’s dragging the average down, and concentrate spend where the data says it belongs. Do that consistently and a 3.7x blended account climbs toward 5x, without spending an extra dollar.

Want us to run this teardown on your account? Get in touch and we’ll show you where your budget is leaking.