Blended ROAS Is Not Incrementality: The Limits

Blended ROAS feels like the honest number. You take total revenue, divide by total spend, and the messy attribution debate disappears. The problem is that it answers a different question than the one you think it answers. Blended ROAS tells you whether the whole machine is efficient. It never tells you whether a specific ad, campaign, or channel actually caused a sale that would not have happened anyway.

That gap, between efficiency and causation, is where most budget decisions quietly go wrong.

For the neighboring economics, compare The Budget Scale-Down Test: DIY Incrementality and use Modeled Conversions: How Much ROAS Is Estimated? to validate the measurement decision.

Two questions that look like one

There are two distinct questions an operator asks about spend, and they get collapsed into a single metric far too frequently.

The first is an accounting question: across everything I spent, how much revenue came back? That is blended ROAS. Total revenue over total cost. It is a scorecard for the business as a system.

The second is a causal question: if I turned this specific spend off, how much of that revenue would disappear? That is incrementality. It isolates the lift a channel produced beyond what would have happened with no ad at all.

The whole confusion in the blended roas vs incremental roas debate comes from treating the first number as if it answers the second. It does not, and it cannot, because of how the numerator gets built.

Why blended ROAS hides the thing you care about

The revenue in your blended numerator is a pile of sales with very different origins mixed together:

• People who saw an ad and bought because of it (true incremental conversions)
• People who were already going to buy and happened to click an ad on the way (harvested demand)
• People who came from organic, email, referral, or repeat purchase and never touched paid at all
• People who bought because of brand momentum built over months

Blended ROAS sums all of these and divides by paid spend. So the metric can move for reasons that have nothing to do with your ads working better. A strong email send, a wave of repeat buyers, a seasonal demand spike, a creator mention, all push blended ROAS up while your actual ad-driven lift is flat or falling. The number looks like ad performance. It is mostly demand you did not create.

This is why a brand can scale paid spend, watch blended ROAS hold steady, and still be lighting money on fire. The incremental contribution of the new spend can be near zero even as the blended average looks calm, because the denominator grew but the numerator was being filled by demand the ads merely intercepted.

The retargeting trap, in plain mechanics

The clearest example is retargeting. Take a prospecting-plus-retargeting setup. The retargeting campaign almost always reports a gorgeous platform ROAS, because it shows ads to people who already added to cart or visited a product page. Those people convert at high rates. Of course they do, they already raised their hand.

Blended ROAS happily counts that revenue. But strip the retargeting out and a large share of those buyers come back anyway through direct, email, or organic search for your brand name. The ad did not create the purchase. It took credit for intercepting it. The honest incremental ROAS on that retargeting line can be a fraction of its reported figure, sometimes low enough that the spend is pure margin leakage dressed up as your best-performing campaign.

You cannot see this in blended ROAS. The metric structurally rewards demand harvesting and structurally hides whether you created any new demand.

What blended ROAS is genuinely good for

This is not an argument to throw the number out. Blended ROAS, paired with contribution margin, is the right efficiency gauge for the business as a whole. It is hard to game at the total level, it ties directly to cash, and it keeps you honest about whether the entire acquisition engine is paying for itself.

Use it as a guardrail. If blended ROAS against your margin structure means you are losing contribution on every order, no amount of channel-level optimization fixes that. It is also the correct altitude for a board conversation or a monthly P&L review, because it maps cleanly to money in and money out.

The error is altitude confusion: using a total-business efficiency metric to make a per-channel causal decision. That is like judging which player to bench by looking at the final score.

Closing the gap: how to actually read incrementality

You will never derive incrementality from the attribution table, because attribution assigns credit, it does not test causation. Closing the gap means designing a comparison where some buyers could have been exposed and some could not.

A few approaches that hold up:

1. Geo or audience holdouts. Suppress a channel for a matched slice of your audience, keep it on for the rest, and compare conversion rates between the exposed and held-out groups. The difference is your lift. This is the cleanest read available to many brands.
2. Scaled on/off tests. Turn a suspect line item, frequently retargeting or branded search defense, fully off for a defined window and watch total orders, not platform-reported orders. If total demand barely moves, the spend was harvesting.
3. The marginal-spend test. Incrementality lives at the margin, not the average. Push spend up in a controlled step and measure the change in total revenue against the change in total spend. Marginal ROAS on new budget is almost always worse than blended average ROAS, and that delta is the real signal for where the ceiling is.
4. Triangulation, not a single source. Read platform-reported conversions, blended ROAS against margin, and a periodic lift test together. When platform ROAS is high but a holdout shows little lift, trust the holdout.

These tests cost you some clean revenue during the window and they take patience. That is the price of a causal answer. The alternative is reallocating budget on a number that was never designed to carry that weight.

A note on tooling and honesty

This is exactly the kind of read that is easy to get wrong at a glance and tedious to do by hand across every campaign. An intelligence layer like Bach AI can flag where reported ROAS and incremental contribution are likely diverging, surface the high-harvest line items worth holdout-testing, and quantify the contribution at stake, so you decide what to test rather than trusting a flattering average. The judgment stays yours. The point is to make the gap visible, not to paper over it with one tidy number.

The takeaway

Hold three numbers, not one. Keep blended ROAS against contribution margin as your business-level efficiency guardrail. Watch marginal ROAS on new spend as your scaling ceiling. And run periodic holdout tests as your only real source of causal truth.

When a campaign reports a ROAS that looks too good, treat it as a hypothesis to test, not a result to celebrate. Blended ROAS will tell you the machine is running. Only incrementality tells you which parts are pulling their weight, and that is the difference between scaling what works and subsidizing demand you already had.