One number: ROAS 2.1 → 3.4

The account we walked into

A subscription-friendly refill brand with real repeat economics and a Shopping setup assembled in a hurry two years earlier: one campaign, every SKU, internal product names in the titles, and a Performance Max add-on running with no brand exclusions. Monthly spend near £9,500. The board number was 2.1, and because the field average is 2.87, the internal narrative had hardened into "Shopping underperforms for us." Nobody had asked what the 2.1 was made of.

Why did the reported number fall first?

Because the old 2.1 was partly fiction. The account ran PMax with no brand exclusions, so people searching the brand's own name were counted into the return. Excluding brand in week one dropped the reported figure to 1.8, which is what prospecting was genuinely earning. Uncomfortable, and the necessary starting point: you cannot improve a number that is lying to you. We set the target from the margin instead, 33% contribution meaning break-even near 3.0, and wrote 3.2 into the engagement.

The PMax channel report, standard in every account since November 2025, is what made the argument short. It showed the campaign's conversions skewing to Search placements, meaning brand queries, while its Display and video impressions burned quietly. One report tab replaced what used to be a week of inference, and the same tab later became the emissions ledger's data source.

What did the 90 days consist of?

The kit-versus-refill split did the heaviest lifting. Kits recruit new customers and can carry a lower first-order return because refills repeat; refills bought by searchers are usually existing customers and get a stricter target. One campaign pretending both are the same product is the most common Shopping structure mistake we see in DTC accounts. The title work carried its published weight too: optimized titles typically lift impressions 15 to 30% and click-through 10 to 20%, and this feed started from internal range names, the worst-case baseline where those ranges actually land.

What we deliberately did not do

No discount codes in ad copy, despite two months of pressure when the number sat at 1.8. Refill economics live on full-price repeat orders; training searchers to expect a code would have bought the target and sold the margin. We also declined to reinstate uncapped PMax when a strong week made it tempting: the cap is the reason the return held, and the emissions column, roughly a third lower than peak after the cap, would have given the trade away anyway.

Is 3.4 actually good?

Against the published field, yes with context: the 2025 ecommerce average of 2.87 and the 2024 median of 2.04 put 3.4 comfortably in the upper band, while the common 4:1 "healthy" bar remains top-quartile territory. Against this brand's economics, it means Shopping now funds its own growth: above the 3.0 break-even with repeat refill revenue on top. The write-up's number excludes brand demand entirely, which is why it reads lower than the vanity version and why we trust it.

What about the carbon column?

The channel mix changed the estimate meaningfully. Standard Shopping tiles sit at the light end of the 1–5g CO2e per-impression range; the PMax display and video placements weigh several times more per impression. Capping PMax at 20% of budget cut the account's estimated media emissions by roughly a third from its peak while the return rose, the kind of trade the report exists to make visible, and the channel report turned the estimate into simple arithmetic. Method and framework in the carbon method; routine details in the Shopping and PMax service.