The easiest efficiency gain may already be hiding inside a channel that looks healthy.
Its core budget still creates sales. The last 15% reaches more expensive inventory, repeats exposure or stretches into demand that adds almost nothing. Average ROAS blends both layers together, so the channel keeps looking productive while its upper edge quietly burns money.
If that edge is real, you do not need a new channel or a better creative to improve efficiency. You remove the spend that stopped adding sales.
A healthy average can hide a saturated edge
Suppose a channel spends $100k per week and produces 5,020 completed sales. At $85k, the same scope produces 5,000. Ordinary weekly movement is roughly ±90 sales.
If one completed sale contributes $30 before media cost, the observed extra 20 sales represent $600 of contribution against $15,000 of additional spend.
The whole channel may still have a respectable average ROAS because the first $85k works. The question is whether the final $15k does.
Do not look for a bad channel. Look for a flat top.
A low-ROAS channel may be weak at every spend level. That is a triage problem. A saturated channel can be highly productive at its core and weak only at the top.
Business performance remains below its boundary across the observed range.
Spend enters the same upper range while completed sales stop rising with it.
The strongest first clue may already be in the account. Budgets hit caps. Markets launch at different levels. Pacing falls during part of a month. A finance constraint temporarily removes spend. These moments create lower-versus-higher comparisons before anyone builds a model.
Find the comparison the account already created
Name the edge first: channel, current spend, lower range and the completed business result you expect to protect.
- 01
Spend crossed the proposed edge
$97k versus $100k cannot answer a $15k question.
- 02
Sales had time to appear
Compare both ranges after the usual purchase delay.
- 03
The same result is counted
The order, revenue or contribution definition did not change.
- 04
The large outside changes are visible
Price, promotions, releases, inventory and other acquisition moves did not silently become the effect of spend.
- 05
The flat top repeats
Another period, market or reversible reduction challenges the first comparison.
Calculate the opportunity before debating the model
Spend saved − contribution lost = expected contribution improvement.
In the example: $15,000 − $600 = $14,400 observed improvement.
The number is not the entire proof. The 20-sale difference is smaller than ordinary movement, so the next useful action is to challenge the edge with repetition or one reversible reduction.
What did the upper spend layer actually buy?
Compare spend saved with the completed-sales contribution placed at risk.
Turn the candidate edge into one reversible decision
If the pattern repeats across comparable periods or markets, reduce only to the observed lower range. If it is promising but not clean enough, reduce in comparable markets, stagger the change or use planned higher- and lower-spend periods.
Do not change bids, audiences, creatives and the budget edge at the same time. The operation is to learn whether this upper layer matters—not to launch a general campaign optimization.
Response curves come after the local question
You do not need a complete response curve to challenge one suspected 15% edge.
A curve becomes useful when the company wants to compare many possible spend levels, place the freed budget across channels or estimate how far another winner can scale. The next gate is whether your own history can reveal that wider range.
Evidence notes
The mechanism was checked against official Google Ads experiment and geo-lift guidance and Google Meridian and Meta Robyn documentation on marginal return, saturation and response curves. The local decision still depends on the buyer's completed-sales data, economics and comparison quality.