Meta Incremental Attribution: The 3-Check Gate Before You Scale
Meta incremental attribution reports and optimizes for conversions Meta's models predict an ad caused. Here is what Meta's docs say, and the three ROAS checks to pass before you scale.

Sections
Meta incremental attribution is an attribution model inside Ads Manager that optimizes delivery, and reports results, based on conversions Meta's models predict an ad caused. Standard click and view windows credit every conversion that follows an ad touch.
The gap between those two numbers is where scaling decisions go wrong. Here is what the feature is, what Meta's own documentation says, and the three checks your ROAS should clear before you add budget.
TL;DR: Platform ROAS counts every conversion inside your click or view window, so it is the number most likely to flatter you. Before any scale step, check three things: platform-reported ROAS against your break-even target, the incremental attribution column in Ads Manager, and 30-day blended ROAS from your store. Scale only when all three clear target. If incremental falls under it, hold budget flat and find out why.
What Meta Incremental Attribution Actually Is
Meta's Help Center describes incremental attribution as an attribution model that optimizes ad delivery for incremental conversions. It "uses machine learning models that predict whether a conversion is caused by an ad." That is the whole official definition. A second Meta page adds that it "enables optimization and reporting based on incremental conversions."
Two jobs hide inside that sentence. One is delivery: the ad auction leans toward people the model thinks your ad actually moved. The other is reporting: you get a results column that only counts conversions Meta considers incremental. You can use the first without caring about the second, and you can read the second without ever running the first.
Here is what I could verify, and where the documentation goes quiet.
- Name in Ads Manager: "Incremental" is an option in the Attribution model dropdown at the ad set level, next to Standard and Custom.
- Availability: setup steps list Sales, Engagement and Leads objectives, with Website or Website and app as the conversion location. Meta's help pages do not publish an eligibility checklist or a rollout schedule.
- Launch date: I could not find one in Meta's official pages. Meta's January 2026 newsroom post says the product reached a multi-billion-dollar annual run-rate seven months after launch, and that a Q4 model rollout drove a 24% increase in incremental conversions versus standard attribution. That is Meta's claim about its own product, not independent data.
- How the model works: not disclosed. Meta says "machine learning models" and stops.
That last bullet matters more than it looks. Everything downstream, including the gate in this post, treats the incremental number as a modeled estimate. Useful, but not a measurement.
Menus move around. I verified these steps against Meta's help pages in November 2026, and you may see slightly different labels in your account.
Meta Incremental Attribution vs 7-Day Click
Meta's attribution model page lists three models: standard, incremental and custom. Inside standard, you pick attribution settings. Those settings apply to standard attribution only, which is the first thing people trip over when they ask about incremental vs 7-day click attribution. They are not two values of the same dropdown.
Standard attribution rules, per Meta, for website and in-store conversions:
- Click-through: events within 1 day or 7 days after a link click on your ad.
- View-through: events within 1 day after an impression.
- Engage-through: events within 1 day after a non-link click, such as a share or save.
- Engage-through: events within 1 day after a non-link click action, such as a share or save.
The logic is rules-based. A purchase lands inside a window after an ad touch, and the ad gets credit. Nothing in that logic asks whether the buyer would have bought anyway. Someone who searched your brand name on Tuesday, saw a retargeting ad on Wednesday and bought on Thursday counts as a 7-day click conversion either way.
Incremental attribution swaps the rule for a prediction. Instead of "did a touch precede the sale," the question becomes "did the touch likely cause it." Meta's own example makes the arithmetic plain: if a campaign has 100 total conversions and Meta considers 70 incremental, the incremental column shows 70 and the standard column shows 100.
Cost per result is recalculated on the same basis.
| Standard (7-day click, 1-day view) | Incremental | |
|---|---|---|
| Question it answers | Did a touch precede the sale? | Did the touch likely cause the sale? |
| Method | Fixed time windows and touch types | Machine learning prediction |
| What it counts | All conversions inside the window | Only conversions modeled as caused |
| Tunable settings | Window and touch type | None, no bid controls |
| Verifiable by you | Yes, from event logs | Only indirectly |
One consequence follows directly. The incremental figure will almost always be lower than the standard figure, because it counts a subset. A drop from 2.12 to 1.71 is not, by itself, proof that anything is broken. The useful question is whether 1.71 still clears the number you need to be profitable. More on that in the gate below.
Meta Attribution Settings in 2026: What Changed
If your numbers shifted this spring and nobody touched the campaign, check the calendar. In a March 3, 2026 announcement, Meta said click-through attribution for website and in-store conversions would count link clicks only.
Shares, saves and other non-link clicks moved into engage-through attribution, which replaces what used to be called engaged-view. Meta framed it as a way to reduce mismatches with tools such as Google Analytics, and said billing does not change.
Rollout varies by account, and Meta's own page notes some accounts may still use prior versions while the change lands. So two ad accounts reading the same event can disagree today for reasons unrelated to creative or audience.
This matters for the gate because your platform ROAS baseline may have been reset. A target you calibrated against last year's click-through numbers is now measured on a narrower definition. If you track Meta ads attribution accurately, re-baseline before comparing to a spring screenshot.
The bigger picture hasn't changed since iOS 14 ATT cut visibility into cross-app behavior. Platforms estimate more, and you verify less. Incremental attribution is the latest step in that direction: a modeled view layered on top of an already modeled attribution window.
The Incremental Attribution Setting in Meta Ads: Where It Lives
Per Meta's setup instructions, you pick the model when you build the ad set. The steps:
- In Ads Manager, click Create and choose Sales, Engagement or Leads as the objective.
- At ad set level, set Conversion location to Website or Website and app.
- Under Conversion goals, choose Maximize number of conversions or Maximize value of conversions.
- Click Show more options.
- Under Attribution model, open the dropdown and select Incremental.
Meta lists two constraints on that page. You cannot use bid controls with incremental attribution. And you cannot change the attribution model after you publish the campaign, so switching means building a new campaign.
Optimizing for it is a separate decision from reading it. For reporting, you do not need to run any campaign on the incremental model. Meta's results page says you can view modeled incremental results even when it isn't your attribution model:
- Open the Columns: Performance dropdown.
- Scroll to Compare attribution models.
- Select Incremental and click Apply.
Meta notes these results are not available for date ranges before April 1, 2025. Another Meta page calls the same feature "Compare Attribution Settings," so expect label drift between help articles and your live account.

Is Meta ROAS Inflated? What the Numbers Can and Can't Say
Short answer: platform ROAS is generous by design, and nobody can tell you by how much without an experiment. It credits conversions that follow an ad touch. Some of those buyers were already on their way.
Retargeting and branded-search-adjacent audiences are the usual suspects, since the people in them were warm before the ad showed up. That's why retargeting often posts the prettiest ROAS in the account.
The meta incremental attribution column is an attempt to discount that. It is also a model, and Meta itself advises against reading it casually.
Its help page recommends comparing within the same attribution model rather than across models, such as incremental against standard. A second page says results can't be compared in the Campaign Overview table across ad sets using different models.
So how do you use a number its author hedges on? As a stress test, not a verdict. If your platform ROAS clears target by a mile and the incremental column still clears it, you have two independent-ish reasons for confidence.
If the incremental figure drops below break-even, you have a warning that deserves a real investigation. Neither result is ground truth.
The third number, blended ROAS, plays a different role. It ignores attribution altogether. You divide total store revenue by total ad spend, which makes it hard to game and impossible to double count. The MER guide and our deeper take on blended ROAS cover the arithmetic.
Its weakness runs the other way: it includes revenue from email, organic and returning customers, so it can look healthy while paid does little. Funnel ratios give a third read, and the add to cart rate benchmark shows where a real lift would show up first.
The Three-Check Gate: When to Scale Meta Ads
Before any scale step, run three checks in order. Each one covers a different blind spot, and a pass on one tells you nothing about the others.
Check 1: platform-reported ROAS against your target. Use your standard attribution setting, the one you normally report on. The target is your break-even ROAS, plus whatever margin of safety you want.
Compute it with the break-even ROAS calculator and cross-check it in the ROAS calculator, and if you need the formula, how to calculate ROAS walks through it. Pass means platform ROAS is above target.
Check 2: incremental attribution ROAS against the same target. Open Compare attribution models, select Incremental and read the meta incremental attribution ROAS for the same campaign and date range. Pass means the incremental figure is also above target. This is the check that asks whether the ad appears to cause the revenue.
Check 3: 30-day blended ROAS from the store. Take total sales for the last 30 days and divide by total ad spend across every platform.
Shopify's Performance by marketing channel report defaults to the last 30 days and includes ad spend and ROAS, but only for channels that send cost data to Shopify. Add spend from any platform that doesn't report there. Pass means blended ROAS is above target too.
Why 30 days? Short windows are noisy, and purchase lag blurs attribution at the edges. A month smooths order timing, promotions and weekday swings. It also keeps the check slow on purpose: platform ROAS can move daily, and that check should not be the one overruled by a good Tuesday.
A rule of thumb that works: scale when all three clear target. Hold when any one fails. If you want to size the step once you've passed, the sibling post on how much to increase Facebook ad budget covers that, and this post stays on the question of whether to move at all.
Worked Example: 2.12 Platform ROAS vs 1.71 Incremental
Everything below is an illustration with made-up numbers, not data from a real store.
A store sells with roughly 53% contribution margin before ad spend. That puts break-even ROAS at about 1.9. The team spends €30,000 a month on Meta and €20,000 on other channels.
| Check | Reading | Target | Gap | Result |
|---|---|---|---|---|
| 1. Platform ROAS (7-day click, 1-day view) | 2.12 | 1.9 | about 12% above | Pass |
| 2. Incremental ROAS (compare attribution models) | 1.71 | 1.9 | about 10% below | Fail |
| 3. Blended ROAS, 30 days (€104,500 sales / €50,000 spend) | 2.09 | 1.9 | about 10% above | Pass |
Read left to right and the story seems fine. Meta says 2.12, the store says 2.09, and both beat 1.9. That's the trap. Two numbers that agree don't confirm causation, because blended includes every sale the store made, caused by ads or not.
The incremental figure says Meta's model credited roughly 81% of the revenue the standard window credited (1.71 divided by 2.12). If that holds, each extra euro going in returns less than a euro of margin after covering the ad itself. You'd be paying to grow a campaign whose caused return sits under break-even.
Verdict in this example: don't scale. Hold budget flat for now. Meta's model can be wrong, so this is a flag, not a ruling. The next section covers how to settle it.
Reading Mixed Signals Without Freezing
The three-check gate produces eight possible patterns. Four matter in practice.
| Platform | Incremental | Blended | Reading | Action |
|---|---|---|---|---|
| Pass | Pass | Pass | Clean | Scale |
| Pass | Fail | Pass | Credit inflated, store fine | Hold, test for lift |
| Pass | Pass | Fail | Paid fine, business isn't | Check margin, returns, other channels |
| Fail | Any | Any | Below target | Fix before scaling |
The second row is the example above: platform and blended look healthy, and only the causal estimate disagrees. Some operators scale a little anyway when the incremental miss is small, reasoning that the model carries error bars. That's a judgment call about risk tolerance, and it's cheaper to make once you've seen the gap persist across several weeks.
The third row deserves attention too. Strong platform and incremental numbers with a weak blended figure usually point away from ads. Look at refunds, shipping costs, discount depth, or a collapse in repeat purchases. Also check whether you are comparing a Meta-only spend figure against a store total that includes other channels' spend, which skews the ratio.
What about post-purchase surveys? A "how did you hear about us" question won't settle the argument, but it adds a fourth lens that depends on no pixel and no model. When survey share and incremental share point the same way, trust rises.
Don't let any single failed check turn into paralysis. The gate exists to stop one specific mistake: scaling off the most flattering number. It doesn't require perfect agreement. When platform and incremental numbers agree but revenue lags, look past the ad at the page with landing page testing for Facebook ads.
How Incremental Attribution Relates to Lift Tests and Holdouts
A lift test or holdout is the only way in this toolkit to measure incrementality directly. You withhold ads from a randomly chosen group, compare outcomes, and the difference is what the ads caused. The incremental attribution column predicts that difference without withholding anything.
That makes them complementary, not interchangeable. Meta calls incrementality tests the "gold standard" of measurement in its March 2026 announcement, and says fully adopting incrementality measurement "can take time." Meta incremental attribution is its bridge for advertisers who haven't run one. Our Conversion Lift glossary entry covers the Meta-native tool.
Use the experiment to audit the model. A practical sequence:
- Run a lift test or holdout test on one meaningful campaign.
- Compute the real ratio of caused conversions to standard-attributed conversions.
- Compare it with the share Meta's incremental column shows for that campaign.
- If they land close, lean on the column between experiments. If they diverge, trust the experiment.
For the theory behind all of this, start with incrementality and what is incrementality testing, then the glossary entry for incrementality. I won't re-teach it here. The point for this post is narrower: the column is a cheap daily proxy, and the experiment is the periodic calibration.
Limitations of Meta Incremental Attribution
Nothing about the column removes the need for judgment. Plan around these.
- It is modeled. Meta doesn't disclose the model's features or error range. Treat results as modeled conversions, the same way you treat other estimates inside the platform.
- History is short. Meta states the comparison results don't exist for ranges before April 1, 2025. You cannot back-test last Black Friday.
- Low volume is noisy. A campaign with a handful of daily purchases gives a ratio that swings wildly. Wait for enough conversions before trusting a gap.
- Cross-model comparison is discouraged. Meta recommends comparing within the same model, which is awkward when the whole gate is comparing standard to incremental. Use the gap as a signal and your target as the anchor.
- Locked setup. No bid controls, and no changing the model after publishing. Optimizing for it commits you to a new campaign structure.
- Signal quality still rules. A model fed weak events guesses worse. Check your Conversions API setup and match quality first, and see the attribution guide if your data is shaky.
- Reporting access. Meta's Marketing API attribution documentation does not mention incremental attribution at all, so check before you build a dashboard that assumes you can export the column.
- Other windows are in motion. The March 2026 link-click change shifts standard numbers, so mid-year comparisons need care.
Meta's claim of a 24% gain in incremental conversions comes from its own newsroom. I'd run a small split before believing it for your account.
Last, high order values slow the loop. If you sell a €500 product, a campaign might produce a few purchases a day, and any 30-day read leans on a thin sample.
The average order value of your catalog changes how fast any of these checks settles. Small accounts hit the same thin-sample problem when they test ads, which creative testing on a small budget works around.
Outside-In Proof: Using Competitor Ad Longevity as a Sanity Check
Your own attribution is noisy. Competitors' ad behavior is a different kind of evidence, because it doesn't pass through your pixel at all.
An ad that has run continuously for months, then gets iterated into new variants, suggests someone is seeing returns they like. Media buyers rarely keep funding a loser for a quarter. The ad timeline analysis in adlibrary shows when each ad first appeared and how long it stayed live, so you can separate a flash test from a long-running workhorse.
Here is the honest limit. Longevity is a proxy, not proof. We can't see a competitor's ROAS, their margin or their attribution settings, and some brands let weak ads coast. A long-running ad tells you where to look, not what to conclude.
Where it earns its place is as a tiebreaker when the gate gives mixed signals. If your incremental figure is soft on a creative angle and three competitors have run that same angle for six months, the problem may lie in your execution rather than the idea.
Search by category with unified ad search, save the long-runners to saved ads, and follow the competitor ad research workflow to compare angles against yours.
For a broader view of scaling mechanics, the spend-scaling roadmap and our piece on scaling ad spend confidently pair well with the gate. Long-running statics are the clearest version of that longevity signal, and Meta static ads covers what makes them last.
Frequently Asked Questions
What is Meta incremental attribution?
Meta incremental attribution is an attribution model that optimizes delivery for, and reports, conversions that Meta's machine learning models predict were caused by an ad. Standard attribution instead credits every conversion inside a click or view window. In Ads Manager it appears as the Incremental option under Attribution model.
How is incremental attribution different from 7-day click attribution?
7-day click is a rule that credits any purchase within seven days of a link click. Incremental attribution is a prediction of whether the ad caused the purchase. The incremental count is therefore a subset of what standard attribution reports, so incremental ROAS usually comes out lower.
Is Meta ROAS inflated?
Platform ROAS counts conversions that follow an ad touch whether or not the ad caused them, so it tends to overstate causal return, especially in retargeting. How much it overstates is unknown without a lift test or holdout. Check it against the incremental column and your store's blended ROAS.
When should I scale Meta ads?
Scale when platform ROAS, incremental attribution ROAS and 30-day blended ROAS all clear your break-even target. If any one fails, hold budget flat and investigate. Sizing the increase is a separate question, covered in our budget post.
Do I need to run campaigns on meta incremental attribution to see the column?
No. Meta's help page says you can view modeled incremental results through Compare attribution models even when incremental isn't your campaign's model. Results aren't available for date ranges before April 1, 2025.
The cheapest scale decision is the one a second number talked you out of, and the costliest is the one nobody checked. Run the gate before the budget edit, treat meta incremental attribution as the stress test, and keep a holdout in your back pocket for the day the numbers disagree.
Which of your three checks fails first? Starter at €29 a month covers focused manual research, and Pro at €179 gives power users and small teams the volume. Compare them on pricing.
Further Reading
Related Articles

ROAS in 2026: The Number Every Operator Argues About
ROAS = revenue ÷ ad spend, but platform numbers are modeled, not deterministic. Formula, D7/D30 benchmarks by category, and breakeven math.

Tracking Meta Ads Attribution Accurately: The Complete 2026 Guide
Fix your Meta ads attribution gaps: pixel + CAPI deduplication, attribution windows, UTM cross-referencing, modeled conversions, and creative-level measurement in one guide.

iOS 14 ATT: Five Years On — What We Know Now About the Shockwave That Rewired Ad Measurement
ATT opt-in rates settled at ~25%. Meta rebuilt its measurement stack. MMM came back. Five years after iOS 14, here is what the data actually shows.

Meta Static Ads: Why Volume Beats Polish After Andromeda
Meta static ads win after Andromeda because volume beats polish. Native formats, AI variants from one winner, when video wins, and refresh cadence.

How Much to Increase Facebook Ad Budget: Size the Move by ROAS Headroom
How much to increase Facebook ad budget: a ROAS-headroom ladder for move size, a cadence table by spend and AOV, learning-phase limits, and retract rules.

Add to Cart Rate Benchmark: Find the Weakest Funnel Step Before You Scale
Compare your add to cart rate to Baymard and Littledata data, fix the weakest funnel step, then track the cost ladder after you scale ad spend.

Facebook Ads Creative Testing Small Budget: Force a Fair Read
Facebook ads creative testing small budget playbook: force Meta to read every ad with impression caps or one-ad ABO, then score, kill and scale winners.

Landing Page Testing Facebook Ads: Test the URL, Not the Ad
Landing page testing facebook ads, done right: one ad, one ad set per URL, equal budget. Sample size, headline mirroring and a results table.