Your ROAS Is Lying to You

Last quarter, a DTC brand running paid social hit a 6x ROAS on Meta. The marketing team celebrated. The CFO signed off on a budget increase. Three months later, they paused all Meta spend for two weeks to run a test.

Revenue barely moved.

That story is not unusual. It is, in fact, exactly what happens when you run your marketing org on reported ROAS and nothing else. The number looked real. The cause and effect did not hold up.

This is not a story about bad marketing. It is a story about a metric that was never designed to do what most companies are asking it to do.

Return on ad spend is calculated simply: revenue attributed to ads divided by the cost of those ads. If you spent $50,000 on Google and the platform attributed $300,000 in revenue to those campaigns, your ROAS is 6x.

The problem is in that word: attributed.

Attribution is a modeling exercise. It is a platform’s best guess at connecting a sale to an ad touchpoint, using whatever data it can see. Google’s attribution model favors Google touchpoints. Meta’s favors Meta touchpoints. Neither platform has full visibility into your customer’s actual path to purchase, and neither has any incentive to undercount its own contribution.

What you are measuring with ROAS is not impact. You are measuring credit assignment. Those are fundamentally different things.

Here is the scenario that breaks ROAS as a performance metric.

Say 10,000 people were going to buy your product this month regardless of whether they saw an ad. They typed your brand name into Google. They clicked a retargeting ad on their way to checkout. They had already decided to buy. The ad was present. The sale was recorded. ROAS goes up.

This is called baseline revenue, and every business has it. Customers who were already in the funnel. Loyal buyers who repurchase on a schedule. Brand searchers who would have found you organically. When your ad platform tags these conversions, it is inflating your reported ROAS with sales that would have happened anyway.

The gap between what your ad platform reports and what your ads actually caused is often significant. For mature brands with strong organic demand, that gap can be enormous. Some companies find, when they actually test it, that 40 to 60 percent of attributed revenue falls into this category.

That money was not at risk. The ad did not win it. The attribution model just found it.

Modern attribution has gotten more sophisticated in some ways and more self-serving in others. Broad match keywords, automated bidding, and value-based optimization all expand campaign reach. They also expand the pool of conversions a platform can claim credit for.

When Google switched heavily to Performance Max, advertisers found their reported ROAS went up in many cases. What also went up, when people tested it carefully, was the percentage of conversions that were brand searches and existing customers. The algorithm found the easiest conversions to claim. The ROAS looked better. The actual new customer acquisition often did not.

Meta has a similar dynamic with its advantage+ campaigns. Broader targeting means more overlap with people who would have converted anyway. The algorithm optimizes for conversions it can attribute, not necessarily for the conversions your business actually needed to generate.

This is not conspiracy. It is just how these systems are built. They are optimized for reported performance. Reported performance and real-world impact are related, but they are not the same number.

The question ROAS answers is: how much revenue did my ad platform associate with this spend?

The question that actually matters is: how much revenue would I have lost if I had not spent this money?

That second question is called incrementality, and it requires a different measurement approach entirely.

Incremental revenue is the revenue that only happened because of your advertising. Not the revenue your platform claimed. Not the revenue that came in while your ads were running. The revenue that would not have existed without the ad.

When you divide incremental revenue by ad spend, you get incremental ROAS, sometimes written as iROAS. This number tends to be lower than reported ROAS. Significantly lower in many cases. But it is an honest number. It reflects what your advertising is actually worth to the business.

A campaign with a 7x reported ROAS and a 1.8x iROAS is a campaign that looks great on paper and barely covers its own cost in reality. That difference is the lie.

Imagine you run a coffee shop. Every morning, 200 regulars walk in and order without any prompting. You put up a sandwich board outside. That day, 230 people come in. The sandwich board gets credit for 230 customers because they all passed it on the way in.

But 200 of those customers were coming regardless. The sign influenced 30 people. That is what the sign actually did.

If you measured the sign’s ROI using the full 230 customers, you would massively overinvest in sandwich boards. If you measure only the 30 incremental customers, you have an honest picture of what the sign is worth.

Digital ad attribution is largely measuring 230 customers and crediting the sign.

The most reliable way to measure whether your advertising is creating real revenue is through controlled experiments. There are a few methods that work in practice.

Geo holdout testing is the gold standard for most brands running national or regional campaigns. You split geographic markets into two groups, run advertising in one group and not the other, then compare revenue outcomes. If the markets are well-matched going in, the difference in outcome is your incremental revenue lift. This approach works across channels and is not dependent on cookies, pixels, or any platform’s attribution model.

Conversion lift studies are offered by most major platforms. They involve randomly splitting your audience and withholding ads from a holdout group. The gap in conversion rate between exposed and unexposed users is your lift. These studies are better than nothing, but they have limitations around audience selection and the fact that the platform is running the test on its own performance.

Media mix modeling takes a different angle. Rather than running a controlled experiment, MMM uses statistical analysis across historical spend and revenue data to estimate what portion of your results each channel is actually driving. It can tell you, at a portfolio level, which channels are generating incremental returns and which ones are mostly capturing demand you already had.

The common thread in all of these methods is that they measure what would have happened without the advertising. ROAS does not do that.

This is not an argument to throw out ROAS entirely. It has legitimate uses.

For campaign management at the tactical level, ROAS is a fast signal. If one creative is converting at 8x and another at 2x on the same audience, that is useful directional information. If a channel is consistently producing 10x ROAS while another produces 2x, you have a reason to investigate.

ROAS becomes dangerous when it drives budget decisions without any check on whether those attributed conversions were actually incremental. That is when it stops being a useful tool and starts being a number that justifies spending money on impressions that look efficient but are not driving growth.

Smart marketing measurement teams use ROAS as an operational metric and incrementality as a strategic one. The day-to-day optimization runs on ROAS. The budget allocation decisions run on iROAS.

If you are allocating budget based on reported ROAS alone, you are almost certainly overspending on retargeting and brand search while underspending on prospecting and upper funnel. Retargeting reaches people already in the purchase funnel, so it claims high ROAS. Prospecting reaches cold audiences, so it claims lower ROAS. But prospecting is often what is actually building your customer base.

The result, over time, is that you harvest existing demand very efficiently while slowly starving the top of the funnel. Revenue holds flat or declines. You spend more trying to fix it. The ROAS on your retargeting still looks great.

This is one of the more common ways that fast-growing brands stall out. The attribution model tells a story of efficiency. The actual growth trajectory tells a different one.

The fix is not complicated to describe, even if it takes work to implement.

Start by running an incrementality test on your highest-spend channel. Geo holdout tests can typically be designed and launched in a few weeks. You will likely find that your incremental ROAS is lower than reported ROAS. That is normal and expected. The goal is to know the real number so you can make real decisions.

Once you have an iROAS baseline, you can start using it as a benchmark for budget allocation. Channels and campaigns that clear your incrementality threshold get investment. Channels and campaigns that look efficient on ROAS but underperform on iROAS get scrutinized.

Over time, the goal is to build a measurement system where every major budget decision is grounded in causal data, not attributed data. That means regular holdout testing, ideally combined with a media mix model that gives you a portfolio-level view of where your spend is actually generating returns.

ROAS will still be on your dashboard. It should just not be the final word.

What is a good ROAS? There is no universal answer. A healthy ROAS depends on your margins, business model, and how much of your attributed revenue is actually incremental. A 5x reported ROAS with 40 percent baseline revenue overlap is functionally worse than a 3x reported ROAS with high incrementality. Focus on your incremental ROAS relative to your contribution margin target, not on a benchmark number.

What is iROAS? Incremental ROAS (iROAS) measures the revenue your advertising actually caused, divided by what you spent. Unlike reported ROAS, it excludes sales that would have happened regardless of whether the ad ran. It is calculated using controlled experiments, such as geo holdout tests or conversion lift studies, rather than attribution models.

Why does ROAS vary so much between platforms? Each ad platform attributes conversions using its own methodology, attribution window, and data access. Because platforms can only see the touchpoints that happen on their own network, they tend to overcount their own contribution. A single purchase can be claimed by Google, Meta, and a third-party platform simultaneously if a customer touched all three before converting.

How do I measure incrementality? The most common methods are geo holdout testing, platform-level conversion lift studies, and media mix modeling. Geo holdout tests are generally considered the most reliable because they are not run by the platform being measured and do not rely on user-level tracking.

Should I stop using ROAS altogether? No. ROAS is a useful operational metric for campaign management and creative testing. The problem is using it as the primary basis for budget allocation decisions. Pair it with incrementality data and you have a much more complete picture of what your advertising is actually doing for your business.

Measured helps brands run geo holdout tests and measure true media incrementality across every channel. If you want to know what your advertising is actually worth, start there.