ROAS Is the Easiest Number to Fake. Here's How Agencies Do It.

Quick Answer

ROAS, return on ad spend, is the most-quoted metric in performance marketing and the easiest one to manipulate without breaking any rules. Agencies inflate ROAS by claiming credit for branded search, picking long attribution windows, including view-through conversions, and reporting platform-attributed revenue instead of incremental revenue.

A 5x ROAS report can sit on top of a business that has not grown all year. If you want to know whether your ad spend is making you money, ask for incremental revenue and contribution margin. ROAS alone will not tell you.

The Number That Sells Sales Calls

Every agency pitch deck has a slide with a big ROAS number. 4x. 7x. "We took this client from a 2.1 to a 5.6 in 90 days." The number is supposed to mean the ad spend is working.

Sometimes it does. Often it means the agency knows how to dress up a report.

I have seen this a hundred times. A brand spending $80K/month gets a slide saying ROAS is 5.2. They feel good. They renew. Six months later they look at their P&L and the business is flat. The slide was not lying, exactly. It was just answering a question they were not asking. The question that the founder thought they were getting answered was "is the spend making us money?" The question the agency was answering was "are the platforms attributing revenue to the ad sets at a healthy multiple?" Those are wildly different questions, and the gap between them is where most of the trouble lives.

Here are the four most common ways the number gets inflated. Each one is technically legal. Stacked together, they can turn a 2.0 incremental ROAS account into a 5.5 platform-reported ROAS account without anything actually changing about the business.

1. Branded Search Inflation

The cheapest, highest-ROAS ad spend is on people typing your brand name into Google.

These are customers who already decided to buy. They are not new revenue. They are revenue you would have captured anyway through organic search, except now you are paying Google for it instead of letting the organic listing capture the click for free.

Most agency reports include branded search in the topline ROAS. That is how you get a 12x return on a campaign that contributed nothing to growth. Strip it out, and the real number drops by 30-60% for most brands we audit.

Here is what that looks like with real numbers. Say your account spends $50K/month on Google Ads. Of that, $8K is on branded keywords with a 15x ROAS. The remaining $42K runs at 3.2x ROAS on non-branded. Your blended ROAS is 5.1x. The number on the slide.

But the honest comparison is the non-branded number, because the branded revenue was effectively organic with a tax on it. Pull branded out and your real performance is 3.2x, not 5.1x.

The defense the agency will offer is "but branded search protects against competitor poaching." Sometimes true. Run a 30-day test where you pause branded ads and watch what happens to organic traffic for the same brand terms. In about 70% of accounts we have tested, organic recovers almost all of the lost clicks within a week. In the other 30%, there is real competitor pressure and branded ads earn their keep. The point is to know which case you are in, not to pay for branded by default.

2. Long Attribution Windows

Default attribution windows are getting longer. 7-day click. 28-day view. Sometimes longer. The longer the window, the more sales the platform claims credit for, including sales that would have happened with no paid touch at all.

A 28-day view-through window means a customer who saw your ad three weeks ago and bought from an email today gets counted as a paid conversion. The platform reports it. The agency reports it. The ROAS goes up. The business growth does not.

The fix is to standardize on a short window for the headline number. We default to 7-day click as the primary reporting window across both Meta and Google. View-through conversions appear in a separate column with their own attribution rate, and we usually discount them by 50-70% before they hit any blended metric. The longer-window numbers still exist in the appendix as directional information, but they do not drive decisions.

Running the same campaign report at 7-day click vs 28-day click and view can produce a 30-40% gap in the ROAS number. That gap is not performance. It is reporting choice. Founders who do not know which window their agency is using are usually looking at the most generous version, because that is the default the platforms ship and the version that flatters the agency.

3. View-Through Conversions

This is the same problem in a smaller form. A view-through conversion happens when someone is served an ad, does not click, and later converts through another path. The platform takes credit for the impression.

For some campaigns this is fair. Upper-funnel video ads do influence purchase decisions even when the user does not click. For most direct-response campaigns, view-through credit is a fiction. Someone scrolled past a product ad on Instagram three weeks ago and now they are searching for the product on Google. The Google ad clicks them in.

Both platforms claim the conversion. Both can be wrong at the same time.

The honest fix is to report click-through conversions only as the headline, model view-through credit at a steep discount (we usually use 25-40% of the platform's claimed view-through revenue), and run periodic holdout tests to validate the discount rate. If you have never seen a view-through discount applied to your numbers, you are looking at inflated revenue.

4. Platform-Reported vs Actual Revenue

Every platform reports its own revenue numbers. Google says it drove $100K. Meta says it drove $80K. Add them up and the total exceeds what the brand actually sold, because both platforms are claiming credit for the same customers.

The fix is to anchor every report to Shopify, Stripe, or whatever the actual record of revenue is, and then allocate that revenue across channels based on a model the agency can explain. Most agencies do not do this. They paste platform numbers into a deck and call it a report.

The math gets ugly fast. Take a brand doing $400K/month in actual revenue. Meta claims $180K, Google claims $220K, TikTok claims $60K. Add them up and you get $460K of "attributed" revenue against $400K of actual revenue. The overlap is $60K, or 15% of total revenue, that two or more platforms are both claiming. Without an anchored view, the agency reports the sum of platform attributions, and the ROAS calculation gets inflated by that 15% overlap automatically.

Tools like Triple Whale, Northbeam, and Polar Analytics solve this with either first-party tracking or post-purchase surveys that ask the customer "how did you hear about us?" None of those solutions are perfect. All of them are better than letting the platforms grade their own homework and adding up the results.

The Cumulative Effect

These four problems compound. Take a brand whose honest, incremental, contribution-margin-positive performance is a 2.4x return on ad spend. Apply branded search inflation (lifts to 3.5x), apply 28-day attribution windows (lifts to 4.2x), include full-credit view-through conversions (lifts to 4.8x), and use the sum of platform-attributed revenue without anchoring to actual sales (lifts to 5.4x). The agency has done nothing wrong by any platform's rules. The number on the slide is more than double the honest performance.

This is why ROAS as a single-number scorecard is dangerous. It is too easy to walk the number up without walking the business up. Every additional layer of inflation makes the next conversation harder, because once you have shown a client a 5.4x, walking them back to 2.4x is brutal even if the 2.4x is the truth.

What Honest ROAS Reporting Looks Like

We strip branded search out of paid search ROAS by default and report it as its own line. We use 1-day click and 1-day view windows for the headline number and add 7-day and 28-day windows as supporting context in the appendix. We anchor revenue to the actual sales platform via post-purchase survey data or first-party tracking. And we report incremental revenue, our best estimate of what would not have happened without the ad spend, separately from total attributed revenue.

The result is uglier numbers and clearer decisions. A "5.2 ROAS" becomes a 2.8 incremental ROAS, which sounds worse but is the actual answer to "is this making us money?" The first month of moving a client to this reporting is uncomfortable, because the headline numbers drop and there is a moment of "wait, are we doing worse?" The honest answer is no, the reporting is doing better. Everything else stays the same. What changes is what the founder is looking at, and once they have seen the honest version, the prior version stops being useful.

How to Audit Your Own Reports

Three questions to ask your agency this week:

1. Is branded search included in our overall ROAS? If yes, can we see a version with it stripped out?

2. What attribution windows are we using, and what does the same report look like on 1-day click?

3. How much of our reported revenue would the platforms double-count if we added them up, and how are we anchoring to actual sales?

If the answers are vague or come back the next day with phrases like "directionally" and "platform-attributed," you have your answer. If they ship a clean answer within a week with the math shown, you have a partner who is willing to be measured honestly. That is the kind of agency to keep.

FAQ

Is ROAS a useless metric? No. It is a useful diagnostic when reported honestly and paired with incremental revenue. The problem is using it as the only number on the report.

What is incremental revenue? The portion of revenue that would not have happened without the ad spend. Estimated through holdout tests, geo experiments, or media mix modeling. Almost always lower than platform-attributed revenue.

Should I trust platform-reported numbers at all? Trust them as inputs, not as outputs. The platform is grading its own homework. Use the platform numbers to understand campaign movement, but anchor your final reporting to your actual sales data.

Is branded search always bad to run? No. It is sometimes the right call, especially when competitors are bidding on your terms. The problem is including it in the blended ROAS without breaking it out, which flatters the non-branded performance unfairly.