Spending more on our best-ROAS channel would have bought nothing

The problem

Brand paid search was the best line on the ROAS report: about 11 attributed, about 4 after incrementality. Then clicks fell 27% year over year in one week. Leadership's question was reasonable: would spending more have helped? Everyone's reflex was yes. Best channel, clicks down, fund it.

The method

Check the constraint first. Budget-lost impression share for brand, by month: between 0.0% and 0.4% for twelve straight months, 0.0% the week in question. Rank-lost ran 12% to 39%. The core trademark campaign already held 96% of its available impressions. At 100% share on every brand campaign, the ceiling was about 152K clicks. Last year had 160K. You could not buy back to last year at any price.
Measure the elasticity. 6,000 campaign-days, 28 campaigns, 18 months. The naive slope of clicks on cost per click is +0.79, which is what a quick read produces and why the reflex feels right. With campaign fixed effects and a demand control, it is +0.08 with a standard error of 0.11. Zero. Bidding up wins impressions at a worse click-through rate and nets no clicks.
Do the marginal math. The cost of one more click is not the CPC; it is CPC times (1 + 1/elasticity). Even granting an elasticity of 0.8, an incremental brand click costs about $11 against $42 of attributed revenue: a 4x marginal ROAS that needs better than 60% incrementality at a 40% margin. Our own geo holdout, brand cut 61% in test markets, put brand at about two-thirds incremental. Organic brand rose 9% where paid was cut. Shopping, local, and non-brand did not move.
Find the clicks. Over 80% of the decline was one campaign: the phrase-match trademark campaign, 31K clicks last year, 70 this year, turned off in a restructure. Its money had moved to a geo campaign at a 67% higher CPC. A restructure, read as a budget problem.

What it showed

You cannot buy back to last year at any price

Brand paid-search clicks, one fiscal week · last year vs. this year vs. the ceiling at 100% impression share on every brand campaign · figures perturbed

Budget was never the constraint. The channel was already capturing 96% of the impressions it was eligible for. The remaining 4% would not have closed the gap, and the elasticity says it could not have been bought anyway.

Where brand impressions were actually lost

Share of eligible brand impressions lost to budget vs. lost to rank, by month, 18 months · figures perturbed

What the report called a budget problem was a rank and restructure problem. Money does not fix rank, and the restructure was one switch.

The best number on the ROAS report belonged to the one channel that could not absorb another dollar.

Six weeks later, the opposite mistake. Bids were cut 31%. First-slot share on searches for our own brand fell from 80% to 44%, the worst in two and a half years, and no-name resellers took most of the gap. You cannot buy more brand clicks, but you can give them away. The tell that it was us and not a competitor: being outbid pushes your CPC up. Ours fell 34% in ten days, on our own bid calendar.

What changed

Brand search was not funded up to close a click gap. The money went to channels with headroom.
Brand bids are now sized as position insurance, enough to hold the top slot on our own name, not as a volume lever.
The restructure was reversed where it made sense: the switched-off campaign was the cheap one.
A bug found on the way: the brand-classification regex matched "Unbranded" too, mis-tagging spend in two pipelines. Fixed.

Technical notes, for the people who will ask

Impression share. Budget-lost and rank-lost share pulled from the campaign cross-device statistics table, where those fields actually live, weighted by eligible impressions. The ceiling calculation scales each brand campaign's clicks to 100% share and sums.

Elasticity. Panel of 5,981 campaign-days across 28 campaigns, January 2025 to July 2026. log(clicks) on log(CPC) with campaign fixed effects and eligible impressions as the demand control: +0.075 (se 0.107). Current fiscal year only: +0.03. First differences: -0.10. Uncontrolled: +0.79. Mechanism split: impression share responds to CPC (+0.60 on the logit) while click-through rate falls (-0.12), so clicks net to zero.

Marginal cost. With elasticity ε, the marginal cost of a click is CPC × (1 + 1/ε). At the week's $4.82 CPC and $43.61 attributed revenue per click, ε = 0.8 implies $10.84 per incremental click and a 4.0x marginal ROAS, which requires more than 62% incrementality to clear a 40% margin. Measured elasticity near zero makes the marginal cost effectively unbounded.

Holdout. July 2025 geo test: brand paid search reduced 61% in test DMAs. Organic brand +9% in test vs. control; shopping, local inventory, and non-brand flat (difference-in-differences p-values 0.63 to 0.86). Read: roughly 65% of brand paid clicks incremental, 35% substituted by organic.

The other case studies: the organic decline, the mart, and the dashboard suite. Try the planning model that puts the ceiling and the elasticities to work. Or back to all work.

Click counts, shares, and dollar figures are perturbed; campaign names are generalized. The tests, the elasticity specification, the holdout, and the decisions are real.