ROAS is not ROI, and the difference costs money

Last July, one of our clients had a month where return on ad spend dropped by 26%.

Revenue went up 11.6% in the same period. Orders were up. Average order value hit a record. By every measure the business cared about, it was the best month on record.

If we had managed that account on ROAS alone, we would have cut the budget.

This is the most expensive misunderstanding in performance marketing, and it survives because three different numbers get used as if they were one.

The three numbers, plainly

ROAS is revenue attributed to ads, divided by ad spend. It measures the ads. Nothing else.

MER, the marketing efficiency ratio, is total revenue divided by total marketing spend. Everything you sold, divided by everything you spent to sell it. Content, agency fees, creator payments, ads, all of it.

ROI is profit, not revenue, relative to the total investment. It is the only one of the three your accountant recognises.

A 16x ROAS and a 6x MER can describe the same month. They are not in conflict. They answer different questions, and only one of them is the question a founder actually has.

The problem with the dashboard

Every advertising platform is also the judge of its own performance.

Meta counts a conversion if someone saw an ad, then bought within the attribution window. It does not know or care that the buyer first heard about the product through a creator video, searched the brand name on Google two days later, and only then saw the ad. Meta claims that sale. Google Analytics, using a different model, may claim it too.

Add up what every channel claims and you routinely get more revenue than the business actually made. That is not a bug. Each platform measures correctly within its own frame, and no platform can see the others.

The consequence is predictable. The channel that closes the sale gets the credit. The channel that created the demand gets defunded, because on paper it produced nothing.

What it looks like with real numbers

Here is a three month picture from a clean haircare brand we work with, launching a new range on paid social.

Look at month three in isolation and the story is a decline. ROAS fell from 23x to 16.9x, a drop of 26%. On a weekly report, that is the number that gets circled in red.

Now look at the same month against the business. Revenue reached its highest point. Orders rose. Average order value hit a record. Cost per order was still half what it had been in month one.

Three things explain the ROAS drop, and none of them is a performance problem.

Ad budget increased 51%. Scaling spend almost always lowers ROAS. You exhaust the cheapest audience first, then pay more for the next one. A 16.9x return on a larger base can produce more absolute profit than 23x on a smaller one.

The site went down for maintenance. Traffic arrived and could not convert. The spend still counted.

Campaigns re-entered the learning phase after being reactivated. The first 24 to 48 hours after a restart produce spend without sales. That is the algorithm recalibrating, not the campaign failing.

The number that actually held

The line worth watching in that table is not ROAS. It is the marketing share of revenue.

In month one, generating 100 in revenue required 49 in marketing spend. By month three it required 16, while absolute spend was higher than at the start.

ROAS never showed that. MER did, and so did the simple ratio of marketing to revenue.

Why MER catches what ROAS misses

In that account, roughly 58 to 65% of orders came from social, 25 to 29% from organic Google search, and about 16% direct.

Organic search and direct traffic were growing month over month. Both are, in practice, downstream of the content: creator video, product testing formats, and people searching the brand name after seeing it. None of that appears in a ROAS calculation, because none of it is ad spend.

So the content was creating demand, search and direct were carrying it, and the ads were capturing it at the bottom of the funnel and taking credit for the whole journey.

Cut the content on the basis of ROAS, and the ads would have kept their multiple for a few weeks, then quietly declined as the demand feeding them dried up. By the time the dashboard showed it, the cause would have been three months in the past.

MER catches this because it does not care which channel gets the credit. Total revenue, total spend. It cannot be gamed by attribution.

How to set this up

Track total marketing spend, not just media. Ads, agency fees, creator payments, content production, tools. If it was spent to generate revenue, it belongs in the denominator. Most teams only count media, which flatters MER by a wide margin.

Report ROAS and MER side by side, every week. ROAS tells you which campaign to optimise. MER tells you whether the whole system is working. Neither answers the other's question.

Add the marketing to revenue ratio. It is MER expressed as a percentage, and it is the version a non-marketer understands immediately. "We spend 16 to make 100" needs no explanation in a board meeting.

Track cost per order or per qualified lead in absolute terms. Ratios move when revenue moves. Absolute cost per outcome is harder to argue with.

Give any change a 30 day window before you act on it. Budget increases, creative refreshes and campaign restarts all depress short term ROAS by design. Reacting inside two weeks means reacting to the learning phase, not to performance.

When ROAS is still the right number

None of this makes ROAS useless. It is the correct tool for comparing two campaigns, two audiences or two creatives inside the same platform, over the same window, at the same budget level. That is a controlled comparison, and ROAS handles it well.

It becomes misleading the moment you use it to judge one channel against another, or one month against another when something structural changed.

The short version

If your ROAS is falling and your revenue is rising, you probably do not have a problem. You have a measurement frame too narrow for what you are doing.

Find the number that describes the business rather than the platform, and check it before you cut anything.