Analysis
Ad costs rose 18% in 2025. The profitable accounts spent less.
Average CPC across our account base rose 18% in 2025. The standard response was to spend more to defend rank. The profitable cohort did the opposite — and the gap between the two groups is the clearest margin signal we measured all year.
We split 41 accounts into quartiles by contribution-margin growth. The top quartile cut ad spend 7% year over year. The bottom quartile grew it 31%. Revenue growth between the two groups differed by 4 points; margin growth differed by 19.
The mechanism isn't austerity. It's denominator discipline: the top quartile bid against margin per SKU, so rising CPCs priced them out of exactly the auctions that had stopped being worth winning. The bottom quartile bid against ACOS targets — which don't move when fees or COGS do.
Bar chart: the bottom quartile grew ad spend 31 percent and lost 2 points of margin. The top quartile cut spend 7 percent and gained 17 points.
The uncomfortable version of this finding: at 2025 CPC levels, a mid-margin SKU portfolio managed to a flat ACOS target loses money on roughly 1 in 4 ad-attributed orders. The spreadsheet says on-target. The P&L disagrees.
We expect 2026 to widen the gap. CPCs are still rising; margin-anchored bidding gets more valuable as the auction gets more expensive — restraint compounds.
Rising CPCs punish accounts that don't know their margin per SKU — and quietly subsidize the ones that do.
If you manage to ACOS today, the audit's ad-spend rebuild shows what your last 90 days would have looked like bid against contribution. Median finding across 2025 audits: 23% of spend served SKUs that lose money on the click.
Method — data, definitions, exclusions
- Data
- 41 managed accounts, FY2025, settlement-level margin and Ads API spend. No survey data.
- Definitions
- Contribution margin = revenue − COGS − fees − ads, per SKU. Quartiles by margin growth, not size.
- Exclusions
- 3 accounts onboarded mid-year excluded (partial data). Accounts under $1M revenue excluded from quartile math, reported separately.
- Caveat
- Managed accounts share our bidding method; the cohort is not a random sample of the market. The direction survives the bias; the magnitudes may not.