ICDL Digital Marketing (ICDL): Blended ROAS is hiding the problem

Blended ROAS is an average across three channels with completely different jobs, and averages hide the decision you need to make.

Problems:

1. It mixes brand and non-brand search. Brand searches would largely arrive anyway. Counting that revenue as advertising return inflates search ROAS and makes every other channel look weak by comparison - which is precisely the marketing lead's point.
2. It ignores margin. Boots are 55% of search revenue at 22% margin; tents at 45%. A euro of search revenue is therefore worth far less than a euro of tent revenue. A ROAS target rewards selling low-margin volume.
3. It ignores assisted contribution. Display at ROAS 1.1 looks like the worst channel, but if exposure lifts later search conversion by 30%, its value is partly booked in the search column.

Better headline measure: contribution margin after ad spend (or profit ROAS) by product group, with brand search reported separately from non-brand.

A workable floor: gross margin per order must exceed cost per acquisition for each product group, measured on non-brand traffic. That number drives real decisions; 3.2 does not.

Step 1 - fix the reporting split, before moving money. Separate brand from non-brand search so each channel's true incremental return is visible. Without this, any reallocation is guesswork.

Step 2 - treat the three channels by role, not by ROAS ranking.

• Non-brand search: scale where CPA is under the margin ceiling for that product group. Sell tents harder than boots in ads; boots at 22% margin cannot survive the same CPA.
• Social: retains budget if it produces assisted conversions and new-customer acquisition, not just last-click sales.
• Display: reduce to a test budget, but not to zero - it is the only channel plausibly creating the lift the marketing lead describes, and killing it removes the evidence.

Indicative split for one quarter: search €2,300 (with brand capped and non-brand weighted to high-margin lines), social €1,100, display €600 held as a deliberate test with a defined hypothesis and end date.

Step 3 - cap brand search spend at the level needed to defend the results page, and stop counting it as growth.

Step 4 - review at 30, 60 and 90 days against CPA by product group, not blended ROAS.

Finance is right that correlation is not proof - people who see display ads may already be further down the funnel. A defensible test needs a control group.

Design:

• Geography or audience holdout. Split the market into matched regions (or matched audience segments) and switch display off in half for 8–10 weeks. Random assignment, not self-selection, is what makes it a test.
• Measure downstream, not clicks. Primary metric: non-brand search conversion rate and total revenue per region, not display clicks. Secondary: new-customer share.
• Keep everything else constant - same search bids, same social, same seasonality window across both groups.
• Pre-register the decision rule. For example: "if the exposed group's total revenue per capita exceeds the holdout by more than the display cost, display stays." Agreeing this before the data arrives stops the result being argued away.

Minimum viable version if a full holdout is impossible: pause display for four weeks in one region and compare its revenue trend with matched regions, accepting weaker evidence.

What would change the conclusion either way: if the holdout performs the same as the exposed group, the 30% figure was selection, and display budget should be cut.