A month of spend, zero sales. The hard part was deciding which zeros were real.
AI pulled thirty days of ad receipts for a furniture brand in an hour. The judgement was sorting real zeros from scheduled zeros from broken-measurement zeros before cutting the budget.
A furniture brand I look after has been watching its blended return on ad spend sit between 2.5 and 3 for weeks. The target is 4. The instruction was simple: take real money out of the budget without touching anything that works.
The pull took AI about an hour. With the ad accounts connected over MCP, I had thirty days of performance across Meta and Google in one view — every campaign, its daily spend, its purchases, its cost per initiated checkout. Five years ago this was two days of exports and a pivot table nobody trusted.
What the pull surfaced: several campaigns had been spending every single day for a month with zero recorded purchases. Not low return. Zero.
Not all zeros mean the same thing
Here is where the hour of AI work stops and the twenty years start. A zero in the purchases column is not an instruction. It is a question. And in this account there were three different kinds of zero wearing the same face.
Real zeros. A set of top-of-funnel geo and audience tests that had run long enough to prove themselves and hadn't. Checkouts were being initiated at a cost no furniture margin can carry, and none of them closed. These I paused the same afternoon — the largest single cut in the batch. Tests are allowed to fail; they are not allowed to fail indefinitely on the company card.
Scheduled zeros. A seasonal World-Cup-themed burst with an end date days away. Weak numbers, but killing it early saves almost nothing and burns the learning. It retires on its date, not in a panic.
Fake zeros. The most dangerous kind. On one platform, almost no campaign was reporting purchase events at all — while the blended dashboard clearly showed revenue moving. That is not a performance problem, that is a measurement problem. Meta's own answer to browser signal loss is the Conversions API, sending events server-side instead of trusting the pixel — and an account without it will under-report its way into bad decisions. You don't kill a channel on a broken meter. I wrote about dashboards lying by omission before; this was the paid-media version.
The dashboard hands everyone the same zeros. Experience is knowing which ones are lying.
What stayed untouched
Brand search, one shopping setup, and a handful of country campaigns paying for themselves three to five times over. Nothing sacred about them either — they simply cleared the bar. The watchlist got the in-betweens: campaigns above zero but below target, each with a date on which they either recover or follow the tests out.
If you think your account doesn't have this kind of waste in it, the industry data says otherwise. The ANA's programmatic transparency study put waste at roughly 23 percent of open-web programmatic spend — about one dollar in four. Waste is not the exception in paid media. It is the default state you have to actively cut against.
Why furniture makes this harder
Furniture buying is a long-consideration category — people spend weeks choosing a sofa. That makes top-of-funnel spend endlessly defensible: any test can claim it is "building the pipeline." Which is exactly why the zero-purchases test needs a hard review date. In a long-journey category, patience and denial produce identical dashboards for a surprisingly long time.
One more habit from the same account: after pausing, verify the pause. I've been burned before by "off" not meaning off, and by AI handing me a number that was wrong in the most convincing way. The follow-up check is part of the cut, not an optional extra.
The expected result: blended return moving from the high twos toward 4 within ten days, if revenue holds while the dead spend comes out. If it doesn't recover, the watchlist campaigns are next. The AI will pull those receipts too. Deciding what they mean stays my job.
The cut took an afternoon. Knowing which zeros were real took twenty years.
Sources & further reading
External:
ANA — Programmatic Media Supply Chain Transparency Study (2023)
Meta Business Help Center — About the Conversions API
Related posts:
People spend weeks choosing a sofa. Most furniture brands sell like it's one visit.
Your analytics dashboard is lying by omission
Off had to mean off
The AI handed me a number, wrong in the most convincing way