Attribution and Incrementality
Channel-reported ROAS is a self-graded scorecard. This framework shows where attribution breaks, why incrementality is the only ground truth, and how to run both together.
Attribution and Incrementality
A framework for separating credit-assignment (attribution) from true causal lift (incrementality) in paid media measurement.
Attribution and incrementality are two different questions wearing the same costume. Attribution asks: of the conversions that already happened, which touchpoints get credit? Incrementality asks: of the conversions that happened, which ones would NOT have happened without the ad? Channel-reported ROAS — from Meta, Google, TikTok — answers the first question, and answers it generously, because each platform claims credit for users who would have bought anyway.
The framework treats attribution as a directional dashboard and incrementality testing as the ground-truth audit. You use attribution daily for pacing and creative decisions. You use incrementality quarterly to calibrate what attribution is over-counting and to reset budget allocation accordingly.
If your Meta dashboard reports a 4.2x ROAS and your finance team's blended revenue/spend ratio is closer to 1.8x, you are looking at the attribution gap. Every channel reports its own credit using its own rules, and the sums double-count users that all three platforms touched on the path to purchase.
This page is the diagnostic layer underneath every ROAS number you report. It covers how attribution windows inflate credit, how iOS14 broke the signal Meta needs to attribute accurately, and why incrementality testing is the only method that answers the question budget-holders actually care about: what happens if we turn this channel off?
Why channel-reported ROAS overstates contribution
Three structural biases push platform ROAS higher than reality. First, last-click and last-touch models route credit to whichever platform sat closest to the purchase — usually branded search or retargeting, which were going to convert regardless. Second, view-through credit lets a platform claim a sale because someone scrolled past an ad without clicking, often within a 1-day or 7-day window.
Third, attribution windows themselves stack credit. A 7-day-click + 1-day-view window means Meta, Google, and TikTok can all simultaneously claim the same conversion if their pixels each fired during the user's research week. Add them up and you get a blended ROAS that exceeds what your bank account actually saw.
The iOS14 signal-loss shock
Before April 2021, Meta could deterministically tie an ad impression to an in-app event via IDFA. After ATT prompts shipped, opt-in rates settled around 20-30%, which means roughly three-quarters of iOS conversions are now modelled rather than observed. Meta fills the gap with statistical inference — useful, but not the same as a tracked click.
The practical effect on Shopify stores: reported ROAS on iOS-heavy audiences (beauty, apparel, anything skewing female 25-44) became noisier and, on net, lower than the pre-iOS14 baseline — while true contribution moved much less. Stores that cut Meta spend on the strength of the new numbers often saw blended revenue drop more than the savings.
The iOS14 trap
If you reallocated budget away from Meta in 2021-2022 based purely on reported ROAS, run a holdout test before doing it again. Several brands have discovered Meta was 30-50% under-credited post-ATT — meaning the channel they cut was actually their best incremental driver. See iOS14 ROAS Impact for the deeper mechanics.
Incrementality testing as ground truth
Incrementality testing answers the counterfactual: what would have happened without this spend? The two practical methods are geo holdouts (turn the channel off in matched-pair regions for 2-4 weeks and compare revenue) and ghost-bid / conversion-lift tests run inside the ad platform itself. Both produce an incremental ROAS — the lift attributable to the ad, not the credit-shuffled version.
The typical finding: branded search and lower-funnel retargeting show incremental ROAS far below reported ROAS, because most of those buyers were already in-market. Prospecting on Meta and YouTube often shows the opposite — lower reported ROAS but meaningful incremental lift. The reallocation implication is significant: cut harvesting spend, fund discovery.
Reported ROAS vs incremental ROAS by channel (illustrative DTC apparel mix)
Reported ROAS
Incremental ROAS
Attribution and incrementality: common questions
Attribution divides credit for conversions that happened across the touchpoints that preceded them. Incrementality measures which of those conversions would not have happened without the ad. Attribution is bookkeeping; incrementality is causality.
Because Meta counts conversions inside its attribution window regardless of whether other channels also claimed them, and because view-through credit captures users who never clicked. Blended ROAS (total revenue / total spend) is the only sum that has to match reality.
Run a full geo-holdout on your top two paid channels at least quarterly, and any time you're considering a budget shift larger than 20%. Creative refreshes, seasonality, and platform algorithm changes all move incremental lift, so a single test has a shelf life of about 90 days.
Yes. Incrementality tests are slow and expensive — you can't run one on every creative or audience daily. Attribution gives you the high-frequency signal for pacing and optimisation; incrementality recalibrates what that signal means.
There is no best model — every model is wrong in a known direction. Data-driven attribution in GA4 is usually the least bad default for cross-channel views. The right discipline is to pick one model, report it consistently, and use incrementality to correct its biases.
It severed Meta's ability to deterministically tie ad impressions to in-app conversions for the ~70-80% of iOS users who decline tracking. Reported conversions are now heavily modelled, which under-credits Meta in many beauty and apparel verticals while adding noise to creative-level decisions.
7-day-click is the current default and the most defensible single window. Adding 1-day-view inflates reported conversions but rarely changes optimisation outcomes meaningfully. Whatever you choose, hold it constant — switching windows mid-campaign makes period-over-period comparisons useless.
Marketing mix modelling and incrementality are complements, not substitutes. MMM gives you top-down channel contribution from historical spend patterns; incrementality validates the MMM with a real experiment. Most mid-market stores get more value from quarterly geo-holdouts than from a full MMM build.
On branded search and retargeting, reported ROAS is often 3-8x higher than incremental — most of those buyers were already converting. On prospecting and upper-funnel video, the gap is much smaller and sometimes reversed, especially post-iOS14.
ROAS as a number is only as honest as the attribution model behind it. Treat channel-reported ROAS as a leading indicator, blended ROAS as the financial truth, and incremental ROAS as the strategic truth that should drive budget allocation.
Track CAC, channels, and funnel conversion in one place
Metricuno connects ad spend, funnel events, and revenue so you can see CAC by channel, cohort, and campaign — without stitching together five tools.