Why Your Facebook ROAS Is Lying to You (And What to Measure Instead)

If you’re running Facebook ads and watching your ROAS climb inside Meta Ads Manager, it’s tempting to read that number as proof the campaigns are working. We get it — it’s a clean metric, it sits right there on the dashboard, and it feels like accountability. But facebook ads attribution inside Meta’s own platform is one of the most systematically misleading numbers in digital marketing. At Choco Media, we’ve spent enough time inside client accounts to know: the gap between what Meta reports and what’s actually happening can be enormous. This post is about understanding why, and what to track instead.

This is written for marketing managers, founders, and agency folk who are either scaling Facebook spend or considering pulling back because results “look fine on paper” but revenue isn’t moving. If you’ve ever presented a 4x ROAS to a client only to see them shrug because nothing felt different in the business, this is for you.

We’ll cover how Meta’s attribution model inflates reported returns, why last-click and view-through attribution create false confidence, how incrementality testing works in practice, and three alternative measurement approaches that actually tell you whether paid media is doing something real.

How Meta Attributes Conversions (and Why It Flatters Itself)

Meta’s default attribution window is 7-day click and 1-day view. That means any purchase made within seven days of someone clicking your ad — or within one day of someone simply seeing it — gets credited to your campaign. On the surface this seems reasonable. In practice, it has a few consequences that quietly inflate your numbers.

First, the 1-day view-through window. If someone scrolls past your ad on their phone, doesn’t click, but then buys your product later that day through Google search or direct navigation, Meta claims that conversion. It saw them. It gets credit. This is attribution by proximity rather than causality.

• A user sees your Facebook ad on Monday morning during their commute.
• They don’t click. They continue scrolling.
• That afternoon, they Google your brand name and buy.
• Google Analytics records this as organic search. Meta records it as a Facebook conversion.
• Both platforms are technically correct by their own rules. Neither is telling you the full truth.

Second, the overlap problem. Facebook, Google, and your email platform are all likely running at the same time. Each one will claim credit for every conversion where there’s any overlap in the attribution window. It’s common to find that the sum of attributed conversions across all platforms is two to three times the actual number of purchases. The money was only spent once, but each channel reports it as a win.

The Last-Click Problem Hasn’t Gone Away

Many teams have moved away from last-click attribution in GA4 in favour of data-driven or linear models. That’s a step in the right direction. But Facebook’s internal attribution is still largely operating on a last-touch-within-window logic, and even GA4’s data-driven model doesn’t fully account for the top-of-funnel awareness that Facebook prospecting creates.

Consider a standard customer journey for a mid-price direct-to-consumer product:

1. User sees a Facebook Reel ad on a Wednesday. Doesn’t click, but registers the brand.
2. Two days later they see a retargeting ad after visiting the website. Still doesn’t buy.
3. A week later they receive a promotional email. They click through and purchase.

In this journey, email gets the conversion in most attribution models. But Facebook arguably started the chain. If you optimise purely on last-click ROAS, you’d cut Facebook prospecting and see your email revenue crater over the following quarter. We’ve seen this happen. The lag between cutting top-of-funnel spend and the downstream impact on email conversion is typically 6–12 weeks, which means by the time you notice, you’ve already damaged the pipeline.

“The metric that’s easiest to report is rarely the metric that’s most accurate. ROAS looks like a clean number until you ask where it came from.”

What Multi-Touch Attribution Models Actually Tell You

Multi-touch attribution (MTA) distributes credit across touchpoints in the conversion journey. There are several models — linear (equal credit across all touches), time-decay (more credit to recent touches), position-based (40% first, 40% last, 20% middle), and data-driven (algorithmic). Each is more useful than single-touch, but all share a core limitation: they’re based on observed data only.

If a customer saw your billboard, heard a podcast ad, and talked to a friend about your product before clicking a Facebook ad and converting, none of that invisible influence appears in your attribution model. MTA can only measure what it can track. And with iOS 14+ signal loss, what Meta can track dropped significantly — Meta’s own estimates suggest modelled conversions (events that can’t be directly verified) now account for a substantial share of reported results.

• Linear model: Good for understanding breadth of influence. Undercredits high-impact touchpoints.
• Time-decay: Logical for short consideration cycles, misleading for high-consideration purchases.
• Position-based: Works well when first-touch awareness matters as much as last-touch conversion.
• Data-driven: Most accurate of the in-platform options, but requires volume and still relies on Meta’s signal set.

Our recommendation is to use data-driven attribution in GA4 as your primary source, treat Meta’s reported ROAS as a directional signal rather than a ground truth, and then layer in the approaches below for anything above a few thousand euros in monthly spend.

Incrementality Testing: The Only Way to Know What’s Actually Working

Incrementality testing asks a specific question: if we turned this campaign off, how many of those conversions would still have happened? The conversions that would have happened anyway are called baseline conversions. Only the lift above that baseline — the incremental conversions — should be attributed to your advertising spend.

The simplest version is a geo holdout test. You split your target market into two regions, run ads in one and not the other, hold everything else constant, and compare conversion rates after 2–4 weeks. The difference between the treated and holdout regions is your incremental impact.

Meta’s own Conversion Lift tool automates a version of this at the campaign level. It’s not perfect — it requires volume to achieve statistical significance, and the holdout audiences need to be genuinely similar — but it’s the closest thing to a controlled experiment available within the platform. We run these quarterly on client accounts to reality-check reported performance, and the findings are consistently sobering.

• A typical e-commerce account we audit shows 30–60% of Meta-reported ROAS is non-incremental.
• Retargeting campaigns are the worst offenders — these often show spectacular ROAS because they’re reaching people who were going to buy anyway.
• Cold audience prospecting often looks inefficient on ROAS but shows the strongest incrementality lift.

This last point is critical. If you’re optimising Facebook spend based on ROAS, you’ll systematically over-invest in retargeting (which looks great) and under-invest in prospecting (which looks expensive). The result is a shrinking top of funnel and declining long-term performance. We covered the other side of this in our post on retargeting — specifically the audience segments where spend actually compounds rather than cannibalises.

Marketing Mix Modelling: Overkill for Some, Essential for Others

Marketing mix modelling (MMM) is a statistical approach that correlates business outcomes (revenue, leads, signups) with marketing inputs (spend by channel, seasonality, price changes, external factors) over time. Unlike attribution, it doesn’t require individual user-level tracking. It works from aggregated data, which makes it both privacy-safe and more robust to signal loss from iOS changes.

MMM was historically expensive and slow — a consulting engagement, six-figure budgets, months of analysis. A new wave of open-source and lightweight tools has changed this. Meta’s open-source Robyn and Google’s Meridian are both free and can produce useful results for smaller businesses. Tools like Northbeam, Triple Whale, and Rockerbox offer managed MMM-style approaches at more accessible price points.

When MMM makes sense

• Monthly digital ad spend above €15,000 across multiple channels.
• Mix includes both online and offline channels (radio, out-of-home, events).
• Business has at least 12–18 months of clean historical data.
• Team has the capacity to interpret and act on model outputs, not just generate them.

When it’s overkill

• Single-channel or two-channel advertiser.
• Monthly spend under €5,000.
• Less than a year of consistent historical data.

For smaller accounts, the combination of geo holdout tests and a disciplined spend-down experiment — cutting a channel entirely for one month and watching revenue — gives you most of the insight without the model complexity.

The Simple Tests We Actually Use With Clients

Most businesses don’t need sophisticated modelling to get a clearer picture than Meta Ads Manager provides. Here are three practical approaches we use as part of our paid media work with clients:

1. The spend-down test

Cut Facebook spend by 50% for four weeks on a segment or geo where you have good data. Watch what happens to total revenue, new customer acquisition, and organic or direct traffic. If revenue falls proportionally, the spend was working. If revenue barely moves, you were paying for conversions that would have happened anyway. This is blunt but effective, and it requires no tools beyond a spreadsheet.

2. The new-customer ROAS filter

Strip your reported ROAS calculation down to first-time buyers only. Repeat purchasers who convert through a Facebook retargeting ad likely would have returned anyway — they’re loyal customers, not paid-acquisition wins. New customer acquisition cost (nCAC) is a harder but more honest metric. If your blended ROAS looks good but nCAC is rising, you’re spending more and more to bring in the same number of new buyers.

3. The 7-day lag analysis

Pull weekly spend and weekly revenue for the past 12 months. Shift the revenue column forward by 7 days — or 14 days for longer consideration cycles — and look for correlation. If there’s a visible relationship between spend and lagged revenue, that’s evidence your advertising is doing real work. If the correlation disappears, that’s a signal worth investigating before you scale further.

• Use GA4’s exploration reports to segment new versus returning customers by acquisition source.
• Compare Meta-reported conversions against GA4’s conversion count for the same period — gaps above 30% warrant investigation.
• Set up a simple spreadsheet with weekly spend and new customer revenue as a minimum viable attribution sanity check.

What to Put on Your Dashboard Instead of ROAS

We’re not suggesting you ignore ROAS — it’s still a useful directional signal and a reasonable shorthand for efficiency conversations. But it shouldn’t be the primary metric you optimise against. In client work we’ve found the following set more reliable for understanding whether spend is actually producing growth:

• nCAC (new customer acquisition cost): Total Meta spend divided by number of verified first-time buyers from paid-social source, using GA4 data rather than Meta’s own reporting.
• Incremental ROAS (iROAS): ROAS calculated from a holdout test. Tells you what the spend actually produced, not what Meta claimed it produced.
• Blended MER (marketing efficiency ratio): Total revenue divided by total advertising spend across all channels. A coarser metric but immune to attribution manipulation — the business either grew or it didn’t.
• 7-day new-user sessions from paid social: A leading indicator. If awareness spend is working, new user traffic should trend up before revenue follows.
• Brand search volume: A downstream signal that prospecting is building awareness. Track via Google Search Console or Google Trends over a rolling 90-day window.

For most accounts we manage, switching from ROAS to blended MER as the primary metric immediately reduces the incentive to over-invest in retargeting. The business stops optimising the number that’s easiest to game and starts optimising total growth.

A Note on Meta’s Modelled Conversions Post-iOS 14

Since Apple’s App Tracking Transparency framework rolled out in 2021, Meta has been filling data gaps with modelled conversions — statistical estimates of conversions that happened but couldn’t be directly measured. Meta is transparent about this in its reporting (there’s a “modelled” indicator in the breakdown columns), but it’s easy to miss when you’re reading dashboard summaries.

In practice, this means a portion of your reported conversions are estimates rather than verified events. Meta’s modelling is reasonably accurate at the aggregate level, but it’s affected by changes in your audience mix, creative performance, and broader market conditions. Treating modelled conversions as verified events overstates certainty in ways that compound over time.

• Enable Conversions API (CAPI) to send server-side events to Meta — this directly improves signal quality and reduces reliance on modelled data.
• Review the “Data Source” breakdown in Meta’s Events Manager to see what share of your reported events are matched versus modelled.
• Implement the Meta pixel and CAPI in parallel (event deduplication handles the overlap) — dual-source matching gives Meta more verified signal to work with.

Taking Action

The goal here isn’t to make you distrust every number your ad platform produces — it’s to give you enough context to hold those numbers appropriately. Meta’s ROAS reporting is a useful starting point, not an ending point. The businesses that grow efficiently on paid social tend to be the ones who treat platform reporting as one signal among several, invest in periodic incrementality testing, and make decisions based on business outcomes rather than platform-reported metrics. If you want a framework for thinking about the whole paid media picture, our paid media playbook covers the broader multi-channel approach we recommend in 2026.

If you’re finding it hard to tell whether your Facebook spend is actually driving growth, or you’re looking at a reporting stack that contradicts itself across platforms, that’s a solvable problem. Get in touch and we’ll walk through your current setup and measurement approach together.