ROAS Is a Rearview Mirror: The Metric That Lies to Ecommerce Brands

Your ROAS looks great. Your business is stagnating. That contradiction is more common than most ecommerce brands want to admit, and it points to a fundamental problem with how the industry has been measuring marketing effectiveness.

ROAS does not measure what most people think it measures. As the measurement firm Measured puts it in their piece "Your ROAS Is Lying to You: Here's What to Measure Instead", what you are measuring with ROAS is not impact. It is credit assignment. Those are not the same thing. One tells you which channel's hand was nearest the register when the sale happened. The other tells you whether the ad actually caused the sale.

That distinction is not a technical footnote. It is the entire problem.

The Credit Assignment Trap

At its core, the metric answers one question: how much revenue did the ad platform claim credit for, divided by what you spent? It does not ask whether that revenue would have happened anyway. It does not ask whether the customer was already going to buy and simply clicked your retargeting ad on the way. It does not ask whether another channel created the demand your bottom-funnel campaign harvested.

Measured calls this the difference between credit assignment and impact. Attribution models are built to assign credit. They are not built to prove causation.

For ecommerce brands, the channels that are easiest to measure, Google Search and retargeting, are also the channels most prone to overclaiming. They intercept customers who were already in motion. They show up at the end of a consideration process they did not create. ROAS numbers look extraordinary because conversion rates at the bottom of the funnel are high. But strip out the demand that upper-funnel brand activity and organic intent created, and the incremental contribution of those final-click channels is often much lower.

Measured confirms this directly: incremental ROAS is "significantly lower in many cases" than reported ROAS.

When you sum reported ROAS across platforms, the distortion compounds further. Meta claims credit. Google claims credit. Each platform attributes the same conversion to itself using its own attribution window. As Data Analytics Stack documents, summed platform ROAS across channels can exceed your actual total store revenue. You are double and triple-counting the same sale.

MediaPost published a piece titled "Why ROAS Is the Most Dangerous Metric In Marketing". Dangerous is the right word. Not because the math is wrong, but because it is incomplete in a way that systematically misleads budget decisions.

What Neil Patel and the IPA DataBank Agree On

Neil Patel has argued that attribution metrics reward demand capture over demand generation. Channels that capture existing intent look efficient in an attribution model. Channels that build future intent, the ones that make customers want your product before they even search for it, are invisible to that same model.

This is not a new observation. Les Binet and Peter Field documented the same structural problem in Marketing in the Era of Accountability, their analysis of campaigns from the IPA Effectiveness Awards DataBank. Their finding was stark: campaigns optimized for ROI-style metrics tend to reward short-term efficiency at the cost of long-term growth. IPA DataBank data showed that campaigns built around sales activation showed quick revenue spikes followed by rapid declines, while campaigns aimed at building brand preference and market share delivered slower but more durable business effects.

Binet and Field also found that ROI optimization specifically, what we would now call ROAS targeting, encourages budget cutting that boosts apparent ROI by reducing investment, often at the cost of long-term profitability. The metric improves. The business erodes.

Accountability becomes a trap. You become very good at explaining short-term numbers while quietly undermining the conditions that generate long-term growth.

Patrick Gilbert covers this tension at length in Never Always, Never Never, drawing on Binet and Field's framework to argue that measurement has been misused: we have taken tools designed for learning and turned them into verdicts. Pressure to demonstrate accountability has narrowed marketing's field of vision to whatever is easiest to track, not whatever actually matters.

Why iOS and Modeled Conversions Make It Worse

Meta's ROAS numbers have a specific inflation problem that has gotten worse, not better. Attribution window settings, view-through credit, and modeled conversions all inflate reported performance. iOS privacy changes removed a significant portion of the signal Meta's attribution relied on, and the platform responded by modeling conversions it can no longer observe directly.

Modeled conversions are educated estimates. They are not observed events. When your Meta ROAS is built partly on modeled data, you are optimizing against a reconstruction of reality, not reality itself.

None of this means Meta advertising does not work. It means the reported ROAS number is an even less reliable measure of actual causal impact than it was before. The gap between attributed ROAS and incremental ROAS, already wide, has likely grown.

Some practitioners now advocate for new-customer ROAS as a partial fix, arguing that the key distortion is blending repeat purchasers with new acquisition in the same ROAS figure. A returning customer who buys again with minimal ad influence inflates the number without reflecting genuine marketing effectiveness. Separating prospecting from retention performance gives a clearer picture, though it still does not solve the incrementality problem.

The Fix: Incrementality First, Attribution Second

Measured's recommendation is direct: start with geo holdout tests on your highest-spend channel. These tests, which Measured says can be designed and launched in a few weeks, compare conversion behavior in regions that see your advertising against regions that do not. The difference is your incremental lift. That number reflects causation, not correlation.

Running a holdout test frequently produces uncomfortable results. Brands discover that a meaningful portion of the revenue their ads claimed credit for would have happened anyway. That is not a reason to abandon paid media. It is a reason to understand what you are actually buying.

Incrementality vs attribution is not a debate about which tool is better. It is a debate about what question you are asking. Attribution asks: which channel touched the conversion? Incrementality asks: which channel caused it? Budget decisions should be driven by the second question.

Marketing mix modeling adds a third layer. Where incrementality tests individual channels with experimental precision, MMM looks across your full marketing investment over time and asks which combinations of spend tend to produce the strongest business outcomes. It is slower and requires more historical data, but it accounts for factors attribution cannot see: brand-building effects, seasonality, competitive activity, and the compounding value of sustained market presence.

A practical measurement stack for an ecommerce brand is not complicated in concept, even if it requires discipline to execute. Use attribution for within-channel optimization: which creative, which audience, which keyword. Use incrementality to pressure-test whether your highest-spend channels are actually driving behavior that would not have happened otherwise. Use MMM to set allocation strategy across channels over longer time horizons. And use blended ROAS or Marketing Efficiency Ratio, total revenue divided by total ad spend, as a business-level sanity check that cuts across platform-reported numbers.

At AdVenture Media, the teams working on ecommerce measurement have consistently found that brands optimizing solely toward platform ROAS end up over-investing in bottom-funnel channels at the expense of the upper-funnel activity that feeds them.

The Scoreboard Problem

In Never Always, Never Never, the chapter on measurement uses the analogy of NFL film rooms and scoreboards. Attribution data, like a box score, tells you who scored. It does not tell you why the team won, or whether it will keep winning. Offensive linemen who created the conditions for every touchdown do not appear on the scoreboard. A brand campaign that seeded purchase intent weeks before the search click does not appear in your ROAS dashboard.

Here lies the structural flaw. We built incentive systems around the metrics that are easiest to see, not the ones that best explain performance. Agencies whose survival depends on demonstrating short-term attributable ROAS will, rationally, tilt work toward whatever is easiest to measure and easiest to claim credit for. That is not a failure of character. It is a failure of framing.

The brand vs performance marketing split is real, and the ROAS obsession sits at the center of it. Brands that measure only what attribution can see end up systematically underinvesting in the brand activity that makes performance marketing work. ROAS numbers stay healthy right up until they do not, and by then the brand equity that sustained demand has been quietly depleted.

The 95-5 rule also matters for how you interpret ROAS data. At any given moment, roughly 95% of your potential buyers are not in-market. ROAS only captures the 5% who are. Optimizing exclusively toward that number means you are measuring your success at harvesting demand, not your success at building it. Those are different businesses.

What to Do Instead

ROAS is not useless. Treat it as a channel-level diagnostic, not a business scorecard. Here is what a more honest measurement approach looks like:

• Run a geo holdout test on your highest-spend channel. Measured recommends starting here. The test takes a few weeks to design and launch. The result will tell you your incremental ROAS, which is the number that actually justifies the spend.
• Calculate blended ROAS or MER across your entire ad budget. Total revenue divided by total ad spend. This cuts through cross-platform credit overlap and gives you a business-level view that no single platform can manipulate.
• Separate new-customer and returning-customer performance. Blended ROAS that mixes repeat purchasers with new acquisition is measuring two different things. Acquisition efficiency and retention efficiency deserve separate metrics.
• Build toward MMM for allocation decisions. Attribution tells you what to optimize within a channel. MMM tells you how to allocate across channels. Using attribution for allocation decisions is the wrong tool for the job. The marketing mix modeling vs attribution distinction is one of the most practically important in measurement.
• Hold the film room and the scoreboard separate. Measurement tools are for learning and improvement. Business outcomes, revenue growth, profit, market share, brand demand, are the scorecard. Confusing them is how agencies end up optimizing for metrics instead of results.

A clear trend is emerging across the industry. More sophisticated brands are demoting ROAS from north-star metric to optimization metric, and building measurement systems that include incrementality, MER, and MMM alongside attribution. The tools are more accessible than they have ever been. The mental shift is harder.

A rearview mirror is a useful instrument. You should not drive by looking only at it.

For a deeper framework on how these measurement tools work together, the Scoreboards vs. Film Room framework covers the full model. And if you want to understand how the attribution problem connects to the broader brand vs performance budget debate, the analysis of how Airbnb cut performance marketing spend and grew is worth reading alongside this one.