An 8x ROAS on view-through-heavy remarketing and an 8x on click-only prospecting are not the same number. The first is probably worth closer to 5x once you strip out conversions the ad got credit for but did not cause. The second is close to real. Meta's own Conversion Lift studies put the genuinely causal share of view-through conversions at only 20 to 40 percent, the rest would have happened anyway, which is exactly the credit that inflates the first 8x and not the second. ROAS is not a measurement, it is a measurement run through an attribution model, and the model you pick changes the answer by 15 to 45 percent before any actual performance changes. This is a field guide to reading the inflation, quantifying it in your own account, and running the one report that tells you which 8x you actually have.
The short version: attribution window decides the number, not performance
The single most important thing to understand about ROAS is that the attribution window is a credit-assignment rule, and looser rules hand out more credit. A view-through conversion means someone saw your ad, did nothing, then bought within 24 hours through email, direct, or branded search. Meta counts that as a win for the ad. Sometimes it was. Usually it would have happened anyway. The looser the window, the more of those "would have happened anyway" conversions get folded into your ROAS, and the higher the number reads while the actual business result stays flat.
This is why two accounts can both report 8x and be in completely different shape. The configuration is the story, not the headline.
The attribution inflation hierarchy
Here is the ladder from most trustworthy to least, with the inflation each step adds.
| Attribution setting | What it adds | Inflation vs click-only | Reliability |
|---|---|---|---|
| 1-day click only | Clicked and bought within 24h | Near zero (undercounts 20-40%) | 0.90 |
| 7-day click only | Captures research-to-purchase | Low (10-25% below full standard) | 0.85 |
| 7dc + 1-day engage-through | Adds likes/shares/5s+ video | 5-20% | 0.80 |
| 7dc + 1det + 1-day view (prospecting) | Adds view-through credit | 5-15% | 0.75 |
| 7dc + 1det + 1-day view (remarketing) | Same, high-intent audience | 15-30% | 0.55 |
| 7dc + 1det + 7-day view | Extended view window | 25-45% | 0.50 |
| Remarketing + view-through + frequency 4+ | Compounding 24h windows | 25-45% | 0.35 |
The reliability scores on the right are not vibes. They feed the attribution sub-score inside a formal trust model. The pattern is consistent: view-through is the primary inflation engine, and it is most dangerous on remarketing, where the audience was going to buy with or without the impression.
Why remarketing inflates harder than prospecting
Remarketing audiences have purchase intent independent of the ad. They already know the brand, they are already in the funnel, and many of them are also on the email list. So when a view-through window is open and someone opens a Klaviyo email, clicks through, and buys, the Meta ad that served an impression in the same 24-hour window also claims the sale. You are buying attribution credit for a conversion the email drove. Accounts running Meta retargeting and email against the same audience see the highest view-through inflation of any setup, full stop.
Prospecting is cleaner because a cold non-customer who sees an ad, engages, and then buys was more plausibly influenced by that ad. The view-through credit there is more often real. Same window setting, very different trustworthiness, entirely because of who is in the audience.
Quantifying the inflation: what an 8x is actually worth
Take a purchase campaign reporting a flat 8.0x and adjust for what the window is doing.
| Configuration | What the 8x is really worth | Inflation stripped |
|---|---|---|
| 7dc + 1det + 1dv, prospecting | 6.5x to 7.5x | ~10-20% |
| 7dc + 1det + 1dv, remarketing | 5.0x to 6.8x | ~15-40% |
| 7dc + 1det + 1dv, remarketing, freq 4+ | 4.5x to 6.0x | ~25-45% |
| 7dc only, prospecting | 7.5x to 8.0x | minimal (slight undercount) |
| 7dc + 1det + 7dv, broad audience | 5.5x to 6.5x | ~25-45% |
| 7dc only, no CAPI | unknown | direction of error unknown |
The honest takeaway from this table: a remarketing 8x at high frequency could be a real 4.5x, and a prospecting click-only 8x is basically a real 8x. If you scale the first one thinking you cleared your threshold, you are chasing a phantom.
ROAS is not a measurement, it is a measurement run through an attribution model, and the model you pick changes the answer by 15 to 45 percent before any actual performance changes.
The three inflation sources, sized
View-through (1-day view ON). 5-15% uplift on prospecting, 15-30% on remarketing. Industry lift studies put the actually-causal share of view-through conversions at only 20-40 percent. The rest would have converted anyway. Extending the view window from 1 day to 7 days adds another 15-30% inflation on top, and it is the noisiest signal the platform offers.
Engage-through (1-day engage-through). 5-20% uplift. Captures non-link clicks (likes, comments, shares, saves) and 5-second-plus video views within a day. Minimal on prospecting (5-10%). Higher on remarketing (10-20%) because loyal customers who habitually engage with brand content also habitually buy, so the ad scoops credit for organic loyalty behavior.
Click-window length (7-day vs 1-day click). 20-50% difference for considered-purchase categories. For a high-consideration product, 7-day click legitimately captures the research-to-purchase journey. For impulse buys, 7-day and 1-day click should land within 5-10% of each other. If they diverge a lot on a low-AOV product, your customers are discovering on Meta and converting through branded search, and the ad is taking credit for a halo it created but the search line item is also claiming.
Set the click window to the AOV
This is the rule a 20-year buyer applies without thinking. Match the click window to how long the purchase decision actually takes.
| Purchase type | Click window | Reasoning |
|---|---|---|
| Considered (>$50 AOV) | 7-day click | Click-only misses 20-35% of true ad-driven sales |
| Mid-impulse ($30-50) | 1-3 day click | Direct-response window covers most journeys |
| Low AOV / impulse (<$30) | 1-day click | Longer windows over-attribute deliberation that does not exist |
| Free lead / trial | 1-day click | Misses only 10-20%, and those are debatable |
| High-LTV subscription | Full stack, then validate against backend LTV |
The test operators actually run: Compare, not Breakdown
The one move that separates operators who trust their numbers from operators who guess is knowing the difference between Compare Attribution Settings and Breakdown by Attribution. They sound interchangeable. They are not, and the gap is the whole ballgame.
Breakdown by Attribution segments the conversions already in your report. It only slices what the ad set is already crediting. It can never show you a conversion that falls outside your current window.
Compare Attribution Settings generates a separate column for each window you select, regardless of what the ad set is set to. It uncovers conversions hiding outside your settings. This is the report that reveals inflation, because it lets you see the same campaign measured two ways side by side.
The view-through inflation test, step by step
- Pull attribution_spec from every active ad set. Confirm they are not running mixed windows (more on that below).
- Open Compare Attribution Settings for the trailing 30 days.
- Run 7dc + 1det + 1dv against 7dc + 1det (view-through off) on the same campaign.
- The delta between those two columns is your view-through inflation percentage for that account. Not a benchmark. Your actual number.
- For the click side, compare 7-day click against 1-day click. Divide 7-day click results by 1-day click results. A ratio above 1.5x means a long consideration cycle, expected on high-ticket. Above 2.5x is suspicious and means investigate the traffic source.
Run this once per account per quarter and you stop arguing about ROAS in the abstract. You know, in dollars, how much of the headline is real.
Reporting nuance
Use Compare when you are investigating, because it uncovers hidden data. Use Breakdown when you are presenting clean breakouts to a client, because it segments without implying a single hidden truth. And always footnote the attribution window on any ROAS you hand a client. There is no single true attribution number on these platforms, so frame alternative windows as "the full picture of conversion timing," never as "here is the real number."
The silent killers: CAPI gaps and mixed windows
Two configurations do not inflate ROAS, they make it untrustworthy, which is worse.
Missing CAPI. Without server-side tracking, Event Match Quality drops from roughly 80% to 40-50%, and 30-40% of conversions cannot be matched to a user. The paradox is brutal: missing CAPI makes your reported ROAS look lower than reality (events go unmatched and credit is missed) while simultaneously emptying your retargeting audiences and degrading future performance. So you underreport today and perform worse tomorrow. The direction of error is unknown, which means you cannot adjust for it. A no-CAPI 8x is not inflated, it is unreliable. Treat it as a number you cannot use until the pixel is fixed.
Mixed windows across ad sets. When ad sets in the same campaign run different attribution windows, comparison-based optimization between them becomes invalid. You cannot tell which ad set is actually better because you are comparing scores kept by different rules. The trust model drops the attribution sub-score to 0.55 for this configuration, the same harsh penalty it gives a remarketing campaign running view-through, precisely because it poisons every decision downstream.
Don't act on a number you can't trust
The reason to do any of this is to gate your decisions. A simple rule that holds up:
- Scale only on metrics with trust at or above 0.70. A budget scale-up triggered by a 0.55-trust ROAS that is 30% inflated by remarketing view-through is the most common self-inflicted wound in paid social.
- For trust 0.50 to 0.69, validate with MER before scaling. Marketing Efficiency Ratio (total ad spend over total revenue) transcends the attribution window entirely, so it is the tiebreaker when platform ROAS is shaky.
- Below 0.40, do not act. Investigate the measurement first.
The same applies to anomaly response. A ROAS drop on a view-through account might be a real decline, or it might be a frequency drop reducing view-through credit, or Meta shifting delivery from exhausted remarketing to prospecting. A high-trust drop is a real signal. A low-trust drop needs an attribution investigation before anyone touches a budget.
FAQ
Is a higher attribution window cheating?
No. It is a legitimate setting, and 7-day click plus engage plus view is a reasonable default for most ecommerce. It only becomes a problem when you forget that the number it produces is inflated relative to a click-only baseline and you make scaling decisions as if it were not. Know your inflation, then the window is just a window.
My remarketing campaign shows 12x and my prospecting shows 4x. Should I move all my budget to remarketing?
Almost certainly not, and this is the exact trap. Remarketing ROAS is inflated by view-through and engage-through on a high-intent audience that was going to buy anyway, and it is capped by audience size. Prospecting at a genuine 4x is filling the funnel that feeds your remarketing. Run the Compare test on the remarketing campaign with view-through off before you believe the 12x.
How do I quickly sanity-check a ROAS number?
Three questions. Is view-through on? Is this a remarketing or prospecting audience? Is frequency above 4? Three yeses means assume the number is inflated 25 to 45 percent and validate with Compare Attribution Settings before you act on it.
What is the difference between Compare and Breakdown again?
Compare creates a column per window and uncovers conversions outside your current settings, so it reveals inflation. Breakdown only segments the conversions already in your report and can never show you a hidden one. Use Compare to investigate, Breakdown to present.
Does MER fix all of this?
MER sidesteps the attribution window because it measures total spend against total revenue, so it is the right backstop when platform ROAS is low-trust. It will not tell you which campaign or audience drove the result, so you still need clean attribution for optimization. Use both: MER to confirm the business is actually working, attribution to decide where to push.