In short
- Is: a statistical estimate of conversions, or of their link to an ad, that a platform can't observe. Google and Meta report it together with the conversions they observed, not as a separate number
- Causes: visitors who decline cookies, browsers that block or cap cookies such as Safari, Apple's App Tracking Transparency since iOS 14.5, and journeys that switch devices
- Google Ads reports modeled and observed conversions together in one column, and the modeled ones take up to 5 days to settle
- Meta estimates missing or partial data from similar users it can measure and distributes conversions among campaigns, ad sets and ads
- No order ID: a modeled conversion is a platform's estimate of how many of your real orders its ads brought, not an extra order. adtribute counts your shop's real orders and adds no estimates
What modeled conversions are
A modeled conversion is a conversion an ad platform reports without having observed it. Google and Meta see ad interactions on their side and learn of conversions through their tags or your uploads. When the link between the two is missing, they estimate it from users they can measure and add it to the same total.
The link breaks for four reasons:
- Consent. Tags that respect a declined cookie banner don’t follow that visitor. Your CMP collects the choice, Consent Mode passes it to Google’s tags.
- Browsers. Safari blocks cross-site cookies and limits how long cookies set by scripts last (ITP), so a click today and a purchase next week may never meet.
- Apple’s App Tracking Transparency. Since iOS 14.5, an app needs permission before it links a user’s data with other companies’ apps and websites for advertising (Apple). For people who decline, Meta says performance reporting for web conversions may be limited (Meta).
- Devices. A click on a phone, a purchase on a laptop.
What gets estimated matters. Google’s modeling “determines whether a Google ad interaction led to the online conversion. It doesn’t determine whether or not a conversion happened”: in most cases Google receives both the interaction and the conversion and misses only the linkage (Google Ads Help). A modeled conversion is an estimate about your real orders, not an additional order.
A question customers keep sending us: why do Google or Meta show conversions we can’t find among our orders? Modeled conversions are part of the answer, and why the search fails.
How Google models conversions
- Google Ads reports modeled and observed conversions together in the Conversions column, at the same level of detail. It models where browsers block third-party cookies or limit first-party ones, for visitors who decline consent, for iOS traffic under ATT and across devices, and the estimates take up to 5 days to settle. Google checks its models against observed conversions it holds back and says it avoids systematically over-reporting (Google Ads Help).
- Consent Mode lets Google Ads estimate the conversions of visitors who declined the banner from those who accepted it, once an account is large enough, as the Consent Mode v2 entry’s section on what it does to your Google Ads and GA4 numbers explains.
- GA4 models key events for the same reasons, and Google Ads conversions built on GA4 key events include that modeling (GA4 Help). Its BigQuery export leaves the modeled data out (GA4 Help).
How Meta models conversions
Meta models “conversions for missing or partial data”. It pairs users whose conversion data may be missing or partial, for example because of their privacy preferences, with measurable users who behave alike by browser, device, demographics and geography, and estimates the missing conversions from the measurable ones. It also models how conversions distribute among campaigns, ad sets and ads, and checks its models against ground truth such as hold-out groups (Meta). Ads Manager’s results include the estimates.
For people on iOS 14.5 and later, Meta adds a protocol of its own, Aggregated Event Measurement, to recover limited measurement of their website conversions (Meta), as the ATT entry’s section on what ATT changed for Meta and Google explains.
The more conversions Meta can match itself, through the pixel and server events with hashed customer data (CAPI), the less it estimates.
What modeling means when you compare with your shop
- No order ID. The platforms report modeled conversions as part of a total, never as records with an order or click ID, so nothing lines them up with your orders.
- Each platform models its own share. Google and Meta each estimate how many of the orders they can’t see came from their ads, and neither knows what the other claims. With views and each platform’s own windows, that is why platform totals exceed your orders (platform-reported conversions, data discrepancy).
- Recent days move. Modeled conversions settle over up to 5 days, and Google Ads reports a conversion on the date of the ad click (docs: Google Ads Conversion Upload).
- Modeling doesn’t always make the platform’s number bigger. Google Ads gives brand search much of the credit, so non-brand campaigns can look weaker in Google’s own reporting than on click-based attribution, modeled conversions included (Google Ads benchmarks).
A worked example: one month of Google Ads, reconciled
In October a shop receives 1,200 orders at €75 net of VAT, €90,000 in total, and spends €6,000 on Google Ads. 80% of visitors accept its cookie banner, Google’s tags run in advanced Consent Mode, and adtribute’s script loads only after consent. Buyers decline as often as other visitors, and Google and adtribute credit the consenting buyers alike.
| Conversions or orders | Revenue | ROAS | |
|---|---|---|---|
| Google Ads, observed: consenting buyers linked to a click | 180 | €13,500 | 2.25 |
| Google Ads, modeled: its estimate for buyers who declined | 40 | €3,000 | |
| Google Ads, Conversions column | 220 | €16,500 | 2.75 |
| adtribute, orders with a tracked Google Ads click | 180 | €13,500 | 2.25 |
| adtribute, orders without a tracked touchpoint | 240 | €18,000 |
Look for Google’s 220 conversions among your orders. 180 match orders with a tracked Google click. The other 40 match nothing: they are Google’s estimate of how many of the 240 untracked orders its ads brought. The truth lies between 180, what both tools observed, and 420, if every untracked order came from Google, and Google’s model says 220. No one can check that order by order, only in aggregate with a holdout test.
What you can check is the sum: set everything the platforms claim beyond your click-based attribution against the orders your tracking couldn’t attribute. Claims within that pool are plausible; claims far beyond it mean the platforms also take orders your tracking gives other channels.
How adtribute treats modeled conversions
The e-commerce attribution platform adtribute counts the orders your shop received and adds no estimated conversions to them.
- Orders come from your shop, server to server, whether or not a visit was tracked (compliance). An order without any touchpoint, no tracked visit and no matched discount code, stays unattributed and shows as
No matching Touchpoints found(docs: Discount codes in adtribute). - Declined visitors stay a visible gap. Your consent tool decides which visits the script tracks, so the orders of visitors who declined arrive without touchpoints: what Google and Meta estimate for them, adtribute shows as unattributed. How the script, the Shopify pixel and the uploads to Google Ads and Meta handle consent is in the Consent Mode v2 entry’s section on how adtribute handles consent.
- Uploads don’t make the counts equal. Google shows only the uploaded conversions it can tie to a Google Ads click, so its count won’t equal your orders (docs: Google Ads Conversion Upload). Sharing your orders with Google or Meta needs a legal basis under the GDPR, which you decide as the controller (compliance page).
Our view on modeled conversions
Our view: modeled conversions are an honest attempt to count what privacy rules hide, and inside a platform they’re the right input for bidding. But each is the platform’s estimate of its own share, validated by the platform itself. Decide between channels on numbers that add up to your orders, and judge each platform’s surplus against the orders you couldn’t attribute. When a channel looks better right after a consent or tracking change, ask whether it sold more or was modeled more.
FAQ about Modeled Conversions
Are modeled conversions made up?
No. They estimate real conversions, and Google and Meta check their models against data they hold back. But each platform estimates its own share, which you can check only in aggregate or with a test, never order by order.
Can I see which conversions are modeled?
Not one by one. Google reports modeled and observed conversions together in the Conversions column at the same level of detail, and Meta’s results include its estimates. Compare the platform’s total with what you can match, such as the orders your tracking ties to its clicks.
Does adtribute model conversions?
No. adtribute counts the orders your shop received and attributes each to the touchpoints it recorded, such as tracked clicks, discount codes and post-purchase survey answers. An order with none of them stays unattributed instead of being estimated.
Does uploading our orders to Google and Meta replace modeling?
It gives them observed conversions: real orders with a click ID or hashed customer data to match. Both platforms still estimate what they can’t match and credit by their own rules.