I inherited an account breaking even. It now runs at 2.62× ROAS.
The numbers.
More sales per day. Spending less per day.
Six months, February to July 2026.
The challenge.
I took over the account on 1 February 2026. The brand had just over two months of media running and a structure that worked, but was stuck. The inherited ROAS was 1.78× · once the real product cost was accounted for, the operation was running at break-even. Every unit invested came back, with no margin left.
The inherited CPA sat above the level the business could actually sustain, and nobody knew it. The ad platform said one thing, the checkout said another, and there was no way to tell which was true. Without that reading, any decision to scale was a guess.
Seasonality was against me too. Swimwear in Brazil has a narrow window, and I took over at the end of the peak, exactly when the natural curve starts to fall. The challenge was not to improve ROAS. It was to turn an account that spent into an operation that earned, against the season.
The diagnosis.
Before touching a single campaign, I stopped to understand the whole business, not just the ad account. I cross-checked three sources that had never spoken to each other · the ad platform, the checkout and the customer base.
More than a third of sales came from outside the pixel. Which means any decision taken on platform data alone was systematically wrong.
Then I calculated the maximum viable CPA from the real cost of each product, and compared it with what the account was paying. The inherited CPA was above that ceiling. The operation was selling with no margin and nobody knew. That calculation became the ruler for every scaling decision · ROAS without product cost is fiction.
Without that, I'd have optimised the wrong metric · which is exactly what the account was doing.
The strategy.
A three-layer funnel
The inherited structure mixed objectives with no hierarchy. I rebuilt it in three layers with a defined job · top to feed the pixel with cheap signal and build audiences, middle for remarketing on those audiences, bottom for pure conversion. The order matters · a conversion campaign without a mature audience burns budget.
A short creative testing cycle
Over 40 creatives run across formats · catalogue, carousel, Reels, static and video. The rule is simple · kill what doesn't perform quickly, scale what works, never leave a winner on its own.
Active fatigue management
Frequency monitored per campaign, every week. When remarketing saturated and flattened, I moved the budget to fresh audiences and the campaign that received it took the lead. A campaign doesn't die of old age · it dies of neglect.
Optimisation by hour, market and audience
The hourly breakdown showed hours carrying the result and hours burning budget at break-even · I applied dayparting. In the demographic analysis, the more efficient age band was getting half the budget of the dominant one · I corrected the split.
Carousel and catalogue consistently beat video and static in this operation. It became a production guideline and saved months of trial and error.
The curve.
Monthly ROAS · Dec 2025 to Jul 2026
February was the month of reading and rebuilding · it closed below what I inherited. From March the curve turned and never went back.
The system.
Most agencies hand over a screenshot of the Ads Manager. I built a dashboard of my own that unifies ad platform, checkout and customer base in a single panel, continuously updated, with three layers of reading.
Executive view
For the partners. The health of the business, not of the campaign.
Answers: are we making money?
Paid media
For the operation. Performance by campaign and creative, hour, day, demographics, geography, funnel and frequency.
Answers: where do I adjust today?
Insights
The action plan. Automatic anomaly alerts, prioritised recommendations and new audiences to test.
Answers: what do I do next?
A report shows the past. A system changes the present.
If your media is already running.
If your media is running but the numbers don't add up, the problem is rarely the budget. The Teardown finds where it is.