What Is Modeled Conversion Data in Meta Ads?

Meta can’t observe every conversion that follows an ad. When some of the data is missing or only partly there, which Meta says is likely due to industry or regulatory changes, Meta fills the gap with a statistical estimate instead of leaving it blank. Meta says those estimates give a more complete view of reported results, so it helps to know when they are in play and how much weight to put on them.

How does Meta model conversions?

Meta describes three steps. For the first, it judges similarity from on-platform signals such as browser, device, demographics and geography; missing data can come from things like privacy preferences.

1. Model the missing data. Meta pairs people whose conversions it can measure with similar people whose data is missing or partial, then models the missing conversions from the measurable ones.
2. Distribute reported conversions. Where a reporting level lacks data, Meta models how likely each measured conversion came from a click or impression in a given campaign, ad set or ad.
3. Validate against ground truth. Meta compares its models with aggregated data representative of the population it can’t fully measure, such as historical data or holdout groups, then calibrates.

What does Meta use modeling for?

Use	What Meta says it does
Reported results	Estimates conversions where data is missing or partial, for a more complete view of results
Reporting by level	Spreads conversions across campaigns, ad sets and ads so you can see which ones influenced them
Delivery	Lets the delivery system learn from more representative data

Where will you see modeled numbers?

• Breakdowns. Meta says statistical modeling may be used in breakdowns when conversion data is missing or partial.
• iOS app campaigns using SKAdNetwork. Apple reports results at campaign level, so Meta may model ad set and ad results. It won’t model when the campaign has a single ad set or ad.
• Metrics labelled “estimated”. Meta marks a metric as estimated when it comes from sampling or modeling rather than a straight count. Store visits is Meta’s example of a modeled, estimated metric.

Are modeled conversions the same as estimated metrics?

Estimated is the wider label: Meta says estimated metrics come from sampling or modeling. Meta also says its modeled metrics, like store visits, don’t include attributed conversions such as online purchases that it can count. Separately, where it can’t measure conversions directly because data is partial or missing, it may use modeling to account for some of them. So attributed purchases are counted where Meta can measure them, and some may be modeled where data is partial or missing.

How should you use estimated numbers?

Meta says businesses can use estimated metrics as directional guidance for planning and business decisions, to gauge ad performance, and to measure conversions where data may be missing or incomplete. It also states plainly that no modeling solution is perfect.

FAQ

Is a modeled number the same as a sampled one?

No. Meta says estimated metrics come from either sampling or modeling. Sampling looks at a portion of data that represents a larger population, and metrics that count Meta accounts are sampled. Modeling uses data from several sources to measure activity that’s hard to count directly.

Why might Meta’s purchase count differ from my store’s orders?

Meta’s count can include conversions it accounts for with statistical modeling where data is partial or missing, and an order list contains no modeled entries. Meta describes estimated metrics as directional guidance, so read its purchase count as partly estimated rather than as a list of orders.

Does modeling affect which ads get delivered?

Yes. Meta says modeling for delivery lets its systems learn from more representative data. It adds that its models adapt to factors such as browser, device, demographics and ad creative, and that it uses different models depending on the conversions being measured.