In one line
Plot how conversions respond as spend rises and you get a response curve — typically S-shaped: efficient at first, then flattening, then nearly level.
Why it matters
Once you're in the flat part of the curve, more spend barely moves conversions. Estimating this curve lets you judge whether a channel still has room to scale or is already saturated.
Where it's used
Once you have a channel's response curve, you can derive its marginal CPA and use that to decide where to allocate more budget.
Run the numbers
Say a channel produces 500 conversions on $10,000 of spend. Average CPA is $20. Double spend to $20,000 and conversions reach 700, putting average CPA at $28.57.
The number that matters is not that average but the increment. The added $10,000 bought 200 conversions, so marginal CPA is $50 — more than twice the original average. Whether to keep scaling should be decided against that $50 and your target CPA, not against the $20 average. This is how a channel that still looks cheap on average turns out to already be in expensive territory.
What to watch out for
A response curve is estimated from past data, so predictions far beyond any spend level you've actually tried are unreliable. It's safer to scale gradually near the range you've already tested.
Go deeper
Using response curves to optimize budget allocation is covered in Marketing Budget Allocation.