# Blended ROAS Fell. Your Site Did Not Get Worse. Your Traffic Mix Changed.

Source: https://attributionplaybook.com/blog/2026-11-16-blended-roas-traffic-mix-not-site-problem.html

November 16, 2026

*Blended ROAS dropped from 4.1 to 3.3 in a month, and the meeting immediately turned to the site: the new theme, the checkout, the offer. Nobody opened the traffic mix, where prospecting spend had doubled and branded search, the cheapest revenue a store ever gets, had shrunk as a share of the whole.*

## Blended ROAS is a weighted average, and weights move

A blended figure divides total revenue by total spend. Its components have wildly different returns: branded search and retargeting convert warm audiences at high rates, cold prospecting buys strangers at low ones. Change the proportions and the blend moves even if every component performs exactly as before. Doubling prospecting from 20 to 40 percent of spend will pull a blended ROAS down while prospecting itself is working, because you bought a larger share of the hardest revenue. That is not deterioration. That is a decision showing up in an average that cannot explain itself.

The operator habit that prevents the misread is to never quote a blended number without its mix next to it: spend share by audience temperature, revenue share by new versus returning, device split. Three lines. With them, a falling blend during a prospecting push reads as planned. Without them, the same number triggers a site audit, a theme rollback discussion, and a nervous call with the agency, all aimed at a store that did nothing wrong. The metric did not fail. The reading did.

## What to watch instead, and in what order

First, component returns: ROAS and cost per acquisition by campaign class, branded, prospecting, retargeting, each against its own history, never against each other. Second, revenue per session split by source and device, which separates a traffic quality change from a site change, because a site problem shows up inside every source at once while a mix problem shows up as a shift between sources. Third, the blend itself, demoted to a finance view: useful for cash planning, useless for diagnosis.

New versus returning is the split that catches the most misreads after mix. A store that grows its new buyer share will watch conversion rate and blended return fall while doing everything right, because new buyers convert worse than returning ones on any site ever built. Report the blend with the new buyer share attached, or expect to re-litigate the same panic quarterly. This is the discipline the [first party data rebuild](2026-11-15-first-party-data-attribution-rebuild-cookie-loss.md) was really about: owning numbers you can decompose, instead of renting a single number you cannot.

## The meeting rule that ends the argument

Adopt one rule: no blended metric is discussed until its decomposition is on the screen. Mix, source, device, new versus returning, in that order, takes four minutes and settles most arguments before they start. When every component is stable and the blend moved, the answer is mix, and the correct response is to judge the mix decision on its own terms: was the prospecting spend bought at an acceptable cost per new customer, with a payback the business accepts.

Judge prospecting as buying future returning revenue, because that is what it is. A cold campaign that breaks even on first order and feeds a returning pool that converts at three times the rate is a good campaign that blended ROAS will call mediocre forever. Kill it on the blend's say so and the returning pool starves a quarter later, at which point the blend falls again and the site gets blamed a second time. Averages are patient liars. Decompose them.

## Frequently asked questions

### Why did blended ROAS fall when our campaigns did not change?

Because the mix changed. Blended ROAS is a weighted average, and shifting spend toward cold prospecting, which returns less per dollar than branded or retargeting traffic, pulls the average down even when every campaign performs exactly as before.

### How do I tell a traffic mix problem from a site problem?

Split revenue per session by source and device. A site problem depresses every source at once. A mix problem leaves each source's own rate stable while the shares between sources move. The pattern of the fall tells you where it lives.

### Should we stop reporting blended ROAS?

No, demote it. It is a fine finance number for cash planning. It is a bad diagnostic. Report it with spend mix and new versus returning share attached, and never let it open a discussion about site performance on its own.

### How should cold prospecting be judged if not by blended ROAS?

By cost per new customer and payback period. Prospecting buys future returning buyers, who convert far better than strangers. A campaign near break even on first order can be strongly profitable once the returning revenue it creates is counted.
