Marginal ROAS: How to Know When to Stop Scaling a Channel

Your blended ROAS says the channel is working. Your marginal ROAS — the return on the last dollar you spent, not the average of all of them — may be telling you to stop scaling it three weeks ago.

Most e-commerce teams make budget decisions on averages. A channel returns 4x, so it gets more budget; a channel returns 2x, so it gets cut. The problem is that averages describe the past performance of your entire spend, while budget decisions are always about the next dollar. Those are different questions, and on any channel with an auction behind it, they have different answers. As spend rises, you exhaust the cheapest high-intent audiences first and pay progressively more for progressively weaker demand.

What Marginal ROAS Actually Measures

Marginal ROAS is the revenue generated by an incremental unit of spend. If moving a channel from €50k to €60k a month adds €15k in revenue, your marginal ROAS on that step is 1.5x — even if the blended figure the dashboard shows you is still a comfortable 3.5x. The average is being propped up by the first €50k, which keeps performing while the newest €10k quietly loses money.

This is why channels rarely fail loudly. They fail at the margin, while the aggregate number stays green long enough for everyone to keep increasing the budget.

Why Averages Hide the Problem

Diminishing returns are built into how paid channels work, and a blended ROAS is structurally unable to show them:

• Auction dynamics: more budget means bidding deeper into the auction, paying higher CPMs for audiences with weaker purchase intent.
• Frequency saturation: past a certain spend level you stop reaching new buyers and start re-hitting the same ones, which inflates attributed revenue without adding incremental revenue.
• Demand capture limits: branded search and retargeting have a hard ceiling — the existing demand they harvest. Spend past that ceiling buys impressions, not customers.
• Reporting lag: attribution windows smear revenue across weeks, so the damage from over-scaling shows up long after the budget decision was made.

Average ROAS tells you whether a channel worked. Marginal ROAS tells you whether the next euro will. Only one of those is a budgeting question.

How to Measure Marginal ROAS Without a Data Science Team

You do not need a full marketing mix model to see your response curve. A few practical methods get you most of the way:

• Spend-step tests: raise budget on a channel in controlled increments (10–20%) and hold everything else steady. Compare the revenue delta to the spend delta. That ratio is your marginal ROAS.
• Geo splits: scale spend in a subset of comparable regions and hold the rest flat. The difference in revenue growth between the two groups isolates the return on the incremental spend.
• Spend-tier analysis: plot weekly spend against weekly new-customer revenue over the past six months. If the curve flattens above a certain spend level, you have found your saturation point without running a single test.

Whichever method you use, measure against new-customer revenue rather than total attributed revenue. A marginal euro that harvests an existing customer who was buying anyway is not a return — it is a rebate on demand you already owned.

When to Cut, When to Redistribute

A declining marginal ROAS is not automatically a reason to abandon a channel. It is a reason to stop feeding it. The right move is usually redistribution: pull the last, least productive tier of spend and move it to a channel that is still early on its response curve — which for most brands means underinvested channels like affiliate partnerships, commerce content, or contextual placements on the open web, where competition for high-intent audiences is thinner than in the big social auctions.

Set a floor before you scale: decide the marginal ROAS at which incremental spend no longer clears your contribution margin, and treat it as a hard stop. If your margin structure means you break even at 2x, then a channel scaling at 1.5x marginal is losing money on every additional euro, no matter what the blended number says.

This is the same discipline that underpins incrementality testing: the question is never whether a channel drives revenue, but whether the next unit of spend drives revenue that would not have happened anyway. Brands that budget on marginal returns grow slower on paper and faster in reality — because every euro they deploy is buying new demand, not repurchasing their own averages.