--- title: 'Meta incremental attribution: explained' description: 'Understand Meta incremental attribution, campaign eligibility and reporting. Compare modeled conversions with confirmed revenue without confusing it with lift.' canonical: 'https://plainrouter.com/library/meta-incremental-attribution' format: Explainer published_at: '2026-10-01' last_updated: '2026-10-01' --- Meta incremental attribution estimates which conversions an ad caused, rather than crediting every eligible conversion within a selected interaction window. It can inform both reporting and ad delivery, but its modeled result is not a controlled lift experiment or a record of settled revenue. To check it, keep three measurements separate: Meta's attributed conversions, your confirmed orders and revenue, and the additional sales an experiment estimates your ads caused. **Plainrouter helps compare the first two. It does not measure incrementality.** ## What is Meta incremental attribution? Meta describes incremental attribution as a machine-learning model that predicts whether an ad caused a conversion. Selecting it for a campaign also tells Meta to optimize delivery toward those predicted incremental outcomes. [Meta's definition]() Standard attribution answers a different question: did a qualifying conversion follow an eligible ad interaction within the chosen window? A purchase can satisfy that rule even if the customer already intended to buy. Incremental attribution tries to distinguish that existing intent from the effect of the ad. For website and in-store conversions, Meta's standard settings currently include one- or seven-day link-click attribution, one-day view-through attribution, and one-day engage-through attribution. Meta notes that some accounts still use earlier click and engagement definitions during rollout. The familiar “seven-day click, one-day view” combination is an example, not the definition of all Meta attribution. [Attribution models and settings]() ## How does Meta estimate incremental conversions? The model asks a counterfactual question: would this conversion have happened without the ad? You cannot observe both outcomes for the same purchase, so the answer is an estimate. Meta's incremental-attribution Help Center pages confirm machine-learning predictions and modeled reporting. They do not document the training-data composition, calibration, or a customer-specific holdout for every campaign. Do not assume that enabling the setting starts a randomized experiment in your account, or that the number identifies precisely which individual orders were caused by advertising. Holdout experiments provide a separate way to investigate that question. In a [Meta Conversion Lift study](), random assignment creates a group eligible to see the tested ads and a control group withheld from them. The treatment group includes people who never actually see an ad. The comparison estimates causal impact and reports uncertainty; it is not simply a before-and-after sales chart. ## Where do you select incremental attribution? Meta's setup documentation, checked October 1, 2026, lists these choices: - Campaign objective: **Sales, Engagement, or Leads**. - Conversion location: **Website** or **Website and app**. - Conversion goal: **Maximize number of conversions** or **Maximize value of conversions**. - Under **Show more options → Attribution model → Options**, select **Incremental**. Meta says bid controls are unavailable with incremental attribution. It also says the attribution model cannot be changed after the campaign is published: create a new campaign to use a different model. Setup screens can vary as Meta rolls out changes. Check the options actually available to your account before planning a test. [Current setup instructions]() This is an ad-delivery choice. Merely adding an incremental reporting column is a separate action. ## How do you compare incremental results without changing delivery? You can inspect modeled incremental results even when your campaign uses another attribution model: 1. Open Ads Manager. 1. Open **Columns: Performance**. 1. Select **Compare attribution models**. 1. Select **Incremental**, then **Apply**. That is the label in Meta's dedicated reporting guide; its general attribution page also refers to **Compare Attribution Settings**. Meta says the incremental column is unavailable for date ranges before April 1, 2025. [Reporting instructions]() Compare campaigns on the same reporting model and period. Comparing one campaign's standard CPA with another campaign's incremental CPA mixes different denominators and cannot establish which performed better. ## Incremental attribution, standard attribution and Conversion Lift | Measurement | What it counts or estimates | Main limitation | Useful for | | --- | --- | --- | --- | | Standard attribution | Conversions credited under selected interaction and time-window rules | An eligible interaction does not prove causation; several channels may claim the same order | Monitoring consistently defined attributed performance | | Incremental attribution | Conversions Meta's model considers caused by its ads | Model uncertainty and assumptions; not your own randomized holdout | Comparing modeled incremental performance across Meta campaigns and selecting delivery optimization | | Conversion Lift | Additional conversions estimated from randomized eligible and withheld groups | Requires sufficient evidence and a sound experiment; interpret the uncertainty interval | Testing causal impact for the population, campaigns and period studied | | Backend orders and settled revenue | Business transactions recorded under your payment and refund rules | Confirms an outcome, not which marketing caused it | Checking whether reported activity corresponds to real business outcomes | A lift study and incremental attribution serve related purposes but are different products. Meta documents separate eligibility and signal-quality requirements for Conversion Lift. [Conversion Lift methodology and requirements]() ## Why can reported conversions fall while sales stay the same? An incremental column uses a different counting rule. Fewer credited conversions can therefore reflect the model's estimate, with no change to the orders your business actually received. With the same spend and fewer counted results, calculated cost per result rises mechanically. Consider an illustrative week with €1,000 in spend, 100 standard-attributed purchases and 60 modeled incremental purchases. The two CPAs are €10 and about €16.67. If your order ledger contains 120 confirmed purchases, neither Meta column changes that ledger. Nor does subtracting 60 from 120 tell you how many sales were organic: the measures have different populations and meanings. An audience with strong existing purchase intent could show a larger gap between attributed and incremental results. That is a reason to investigate retargeting, not proof that retargeting always performs worse or that the gap is wasted spend. Changing the campaign's optimization model can also change delivery and actual sales. Keep that effect separate from changing a reporting column. Do not conclude “sales fell” or “nothing changed” from the attributed count alone. ## How to check Meta's numbers against your own revenue Use this reconciliation workflow before deciding a reporting difference is a tracking failure or a reason to change budgets. ### 1\. Define the outcome and population Start with the same business, store and conversion event. Compare purchase counts with purchase counts, and monetary values with monetary values. Separate completed purchases from leads, trials, canceled checkouts and test events. Total business revenue includes customers who may never have encountered a Meta ad. ### 2\. Align dates, currencies and reporting rules Write down the ad-account timezone, selected dates, attribution model and windows. Check whether the report places a result on the interaction date, conversion date, or settlement date. The same calendar range is not enough if those rules differ. Choose a consistent revenue basis: gross or net, treatment of tax and shipping, and whether refunds are included. Keep currencies separate unless you explicitly apply the same conversion method to both reports. ### 3\. Allow the reporting period to mature Avoid comparing today's partial platform total with a complete payment export. Record the data's last-read time and recheck after the relevant attribution window and reporting delays. Keep missing platform data marked unavailable; do not convert it to zero. ### 4\. Investigate event integrity For a suspicious order, follow its order reference, event ID, payment state and delivery evidence. Check for repeated Purchase events, mismatched values and browser/server duplicates. A request accepted by Meta proves receipt, not that Meta credited an ad or that the ad caused the purchase. See [how Meta Conversions API works](). ### 5\. Record what the comparison can establish A reconciliation can show that your event stream is incomplete, duplicated, stale or measured on a different basis when supporting evidence confirms the cause. A numerical gap alone cannot establish any of those explanations. Even perfectly matching totals cannot demonstrate incrementality. If the decision is whether advertising creates additional sales, use an appropriate controlled experiment. If the decision is whether your reporting reflects real paid orders, start with the transaction and delivery checks above. ## Where Plainrouter fits [Plainrouter Signals]() keeps arrival counts, eligible verified revenue and stored Meta reporting as separate measurements. Its [passive denominator]() provides arrival context; arrivals are not unique buyers or evidence of ad-caused sales. With Stripe Checkout connected, Plainrouter records eligible paid settlement evidence and handles supported refund events. [Verified revenue]() describes qualifying observed evidence, not modeled uplift. Revenue is grouped by currency and is distinct from accepted conversion deliveries. [Stripe setup and settlement rules]() The MCP tool ` get_performance ` reads stored Plainrouter-versus-Meta reconciliation. Its separate Purchase comparison shows Meta-attributed purchases and eligible accepted server-side Purchase deliveries, with their own period and attribution details. It does not make a fresh Meta API request, and those delivery counts are not interchangeable with settled revenue. [MCP tool reference]() Use this as an independent collection and revenue baseline when investigating Meta's claims. **Plainrouter does not measure incrementality**, expose Meta's counterfactual model, or certify that an attributed purchase was caused by an ad. Read the [purchase comparison limits]() before interpreting a difference. ## Frequently asked questions ### Should I optimize for incremental conversions? Consider testing it when your conversion measurement is reliable and the business question is additional outcomes from ads. First compare existing campaigns using the same incremental reporting model, then evaluate an optimization test against real business results. A lower modeled CPA alone is not proof of higher profit or causal lift. ### Does incremental attribution replace lift studies? No. Incremental attribution supplies modeled estimates and a delivery option. Conversion Lift uses a randomized comparison to estimate impact for a particular study; its uncertainty and eligibility still matter. ### Does it work with the Conversions API? Conversions API supplies conversion signals; incremental attribution is a modeling and optimization choice applied to eligible campaigns. They have different jobs. Sending server events does not make them incremental, and browser/server duplicates still need deduplication. Meta also documents CAPI as a supported, recommended data source for Conversion Lift, which is a separate experiment. ### Why don't the numbers match GA4 or Stripe? They can differ in collection coverage, event definitions, timezones, attribution rules and treatment of settled or refunded payments. Stripe settlement establishes that money was paid under its payment records; an analytics attribution report assigns credit. Reconcile those definitions before treating either number as wrong. ### Can I calculate incrementality by subtracting Meta purchases from total orders? No. That subtraction compares attributed activity with a business total; it does not estimate what would have happened without advertising. A causal comparison needs a valid counterfactual, not just a second dashboard.