ROAS is not the problem.
Return on ad spend is one of the most useful compression metrics in performance marketing. It tells you how much attributed revenue a platform reports relative to media spend. That is valuable. The mistake is treating the number as a complete explanation of what the advertising caused.
A high ROAS campaign can be genuinely efficient. It can also be harvesting branded demand, retargeting customers who were already likely to buy, benefiting from an organic journey that happened earlier or receiving credit because the platform happened to be the final measurable touch.
Attribution and incrementality are different questions.
Attribution asks which touchpoints should receive credit for an observed conversion. Incrementality asks whether the conversion would have happened without the intervention. They are related, but they are not interchangeable.
This matters most when channels have very different jobs. Brand search often captures existing intent. Demand-generation media may create consideration that closes through search or direct. Retargeting may help, but it may also claim customers who needed very little persuasion.
Recent ROAS may not be mature ROAS.
Some businesses have meaningful conversion lag. Revenue and conversions continue to backfill after the reporting date, particularly across higher-consideration products or cross-device journeys.
If the team reacts to yesterday or the last few days as though attribution is final, it can repeatedly reduce campaigns just before the missing conversions arrive. A simple maturity rule — for example, treating recent windows as provisional — is often more useful than adding another dashboard.
Blended numbers can hide the decision.
An account-level ROAS can remain stable while one category is scaling efficiently and another is deteriorating. Conversely, a new test can look weak in isolation while total acquisition improves because it is introducing customers who later convert elsewhere.
The right reporting level depends on the decision: account, channel, campaign, category, product, new customer or geography. One number cannot answer all of them.
A better decision framework.
We still use ROAS. We just place it inside a wider evidence set: category economics, new-customer mix, brand versus non-brand demand, conversion lag, assisted journeys, search behaviour, geographic variation and the performance of the wider account.
Then the question becomes more useful: is the next rupee likely to create acceptable incremental value? That is a better scaling question than “Which campaign has the highest ROAS?”
The takeaway.
ROAS is a useful instrument. It becomes dangerous only when it is promoted from a metric to a theory of how growth happened.