The error, stated plainly
Adding up ROAS across a marketplace channel, a live commerce channel, a paid social channel, and affiliate, then dividing by four and calling it a blended number, is the single most common measurement error in multi-channel SEA operations. It produces a figure that looks precise to two decimal places and describes nothing real.
Each platform’s ROAS is calculated against its own spend base and its own attribution window. A marketplace’s ROAS counts ad spend against sales the platform itself attributes, on its own click window. A paid social platform runs a different window against a different spend base. Averaging four numbers that were never built on the same footing does not produce a fifth number with any meaning, it produces an average of four incompatible units.
Why the average fails a basic test
Take a channel spending S$2,000 with a reported ROAS of 8 and a second channel spending S$20,000 with a reported ROAS of 3. Averaging the two ROAS figures gives 5.5. The actual blended return, total revenue divided by total spend, is S$76,000 divided by S$22,000, or roughly 3.5. The averaged number overstates the real return by more than half, because it weights the small, high-ROAS channel as if it moved as much money as the large one.
This is not a rounding error. It is the direction most channel-mix decisions get made wrong: a small channel with a flattering ROAS pulls the averaged number up, making the whole system look healthier than the spend-weighted reality, right up until the budget shifts toward the small channel and the return compresses because the audience it was reaching was never as large as the big channel’s.
The only version that holds up
The only blended ROAS that survives scrutiny is total attributed revenue across every channel, divided by total spend across every channel, in the same time window. No per-channel averaging step, no weighting by anything other than the actual dollars each channel spent and returned.
Affiliate needs one adjustment before it goes into that total: its attribution and settlement lag. A channel that reports revenue two to four weeks behind the platforms sitting next to it will understate its own contribution in any same-week blended figure, dragging the total down for reasons that have nothing to do with performance. Lag-adjust affiliate to its own trailing, settled window before folding it into the blended total, or exclude it from the current week’s number entirely and report it on its own settled cadence.
What the blended number actually tells you
Once built this way, the blended figure does something a per-platform number cannot: it tells you whether the whole system is profitable, even when individual channels are pulling in different directions. A marketplace channel running ROAS of 2.5 and a live commerce channel running ROAS of 6 can both be true at once, and neither number alone says whether the combined operation, at its combined spend level, clears the margin the business actually needs.
That combined-spend view is also the only honest input into a channel-mix decision. Shifting S$5,000 out of the weaker channel and into the stronger one changes the blended number by an amount that depends on how much each channel actually spent, not by the gap between their two reported ratios. Running the real math before reallocating budget, instead of eyeballing which ROAS number looks bigger, is the difference between a reallocation that improves the blended return and one that just moves spend around a system that ends up performing the same.
Build blended ROAS from one total revenue figure divided by one total spend figure, lag-adjusting affiliate first. Never from averaging separate per-channel ROAS numbers.