Multi-channel attribution models compared: choosing one that holds up for cross-platform ROAS

What a multi-channel attribution model is

A multi-channel attribution model is the rule that decides how much credit each channel gets for a sale when the buyer touched more than one before buying. Last click gives all the credit to the final click. First click gives it all to the first. Linear splits it evenly across every touch. A data-driven model splits it by how much each touch changed the odds of a purchase, learned from paths that converted and paths that did not, as Google Analytics' attribution guide describes.

Cross-platform ROAS is each ad platform's revenue divided by its spend, set side by side. The model behind the revenue decides which platform looks best. It does not change how many sales happened; it changes which channel is credited with them.

Why Meta, Google and your store all claim the same sale

Meta, Google Ads and a store's own analytics each run a separate attribution model, and each model can credit only the touches it can see. Google's data-driven attribution page says the model looks at interactions with Google's own Search, Shopping, YouTube, Display and Demand Gen ads. Meta's standard attribution, as its attribution settings page describes it, counts conversions that follow an impression of, a link click on or an engagement with a Meta ad, inside the window each ad set uses. Neither sees the other's ads. A buyer who clicked both is one sale in the store and can count as a whole sale in each ad platform.

Shopify names the effect. Its marketing reports help describes an any click model that gives full credit to every channel a buyer clicked, and says it allocates more credit than the orders received. Shopify suggests that model for reconciling what each channel reports. Adding Meta's and Google's own revenue figures together is, in effect, an any click total. Our guide to normalizing cross-platform ad metrics covers the other differences in how each platform counts a conversion.

Meta changed its own definitions in 2026. Meta's March 3, 2026 announcement says click-through attribution for website and in-store conversions now counts link clicks only, and that shares, saves and other non-link clicks moved to engage-through attribution. The same post shortened an engaged video view from 10 seconds to 5. Meta said the change would reach advertisers at different points during the rollout, so a Meta figure from before the switch and one from after it are not like for like.

Which attribution models Google, Meta and Shopify offer in 2026

Google Ads and Google Analytics now offer only data-driven and last click attribution, while Shopify still offers five models and Meta offers standard and incremental attribution. Google's Google Ads attribution models page says first click, linear, time decay and position-based models are no longer supported, and that conversion actions using them were upgraded to data-driven. Google's Analytics attribution guide says the same four have been unavailable there since November 2023. A guide that recommends a time decay or position-based model in Google's tools describes an option that no longer exists.

Meta's incremental attribution page describes a model that predicts whether an ad caused a conversion and optimizes delivery toward the ones it did. That is a different question from the standard windows, which count any conversion that followed an ad.

Two of these tools treat history differently when the model changes. Google Analytics' attribution settings page says a new reporting model applies to historical and future data. The Google Ads models page says a new model changes conversions only from that day on. A team that switches models in Google Ads mid-quarter ends up with one report under two rules.

Five orders credited five ways: a worked example

The same five orders give opposite channel rankings under different models. The orders below are invented to show the method; they are not from any store or ad account. Each order is worth $100, so the store took $500. The brand spent $150 on Meta and $100 on Google Ads in the same period.

The first four models below are Shopify's, as its marketing reports help defines them. The last column is what each ad platform would report for itself, assuming every touch fell inside its window and no ad was only viewed.

First click makes Meta look twice as efficient as Google Ads, 2.00x against 1.00x. Last non-direct click makes Google Ads look four and a half times as efficient as Meta, 3.00x against 0.67x. Nothing about the ads changed between those two columns. The two platforms' own claims add up to $600 against $500 of sales, because both claim order A.

Only the blended figure holds still: $500 of revenue over $250 of ad spend is 2.00x under every model, and it says nothing about which channel earned it. A data-driven model cannot be worked out by hand, because it also learns from the paths that did not end in a purchase. Our Meta vs Shopify ROAS guide walks through the same gap on real report exports.

Attribution without third-party cookies: what actually changed

Third-party cookies are still allowed by default in Chrome, so attribution without them is mostly a Safari, Firefox and iOS question. Google's Privacy Sandbox team said on April 22, 2025 that Chrome would keep its current approach to third-party cookie choice and would not roll out a new standalone prompt. The same post says Chrome's Incognito mode already blocks them by default. On October 17, 2025 Google retired the Attribution Reporting API and several other Privacy Sandbox technologies, citing low adoption, and said it would work on an interoperable attribution standard at the W3C instead.

Other browsers moved years earlier. WebKit announced on March 24, 2020 that Safari blocks cookies for cross-site resources by default. Mozilla made Total Cookie Protection the default for Firefox desktop users worldwide on June 14, 2022, which keeps each cookie in a separate jar for the site that set it. On iPhones, Apple's user privacy page says that from iOS 14.5 an app needs the user's permission to track them, and it counts advertising measurement as tracking. Without that permission, the device's advertising identifier reads as all zeros.

Store-side models have a quieter limit. Shopify's marketing reports help says sessions and visitors are counted with cookies on the shopper's device, so a buyer who clicks an ad on a phone and buys on a laptop can appear as two visitors with two paths. Meta's Conversion Lift page recommends its Conversions API as a data source because it depends less on the web browser. Marketing mix models and lift tests avoid the problem altogether, because neither follows a person from site to site.

Marketing mix models and lift tests: measuring without following people

A marketing mix model estimates each channel's contribution from total spend and sales over time, and a lift test measures it by keeping a random group away from the ads. Google's Meridian is an open-source marketing mix model that Google says is free for anyone to use. Meta's Robyn is an experimental open-source one from Meta Marketing Science. Both work on totals per week, not on individual paths, and Robyn's guide calls the method privacy-friendly.

A lift test is an experiment. Meta's Conversion Lift splits an audience into a test group that can see the ads and a control group that cannot, and reads the difference in conversions as the lift. Google's Conversion Lift works the same way, split by users or by geography. Meta's March 2026 post calls incrementality experiments such as Conversion Lift the gold standard in measurement.

The minimums come from Google's data-driven attribution page, the Analytics attribution settings, Meridian's data guide, Robyn's guide and Meta's Conversion Lift page. Meta prorates its lift guide for campaigns longer than 90 days: a 180-day campaign needs USD 10,000 and 1,000 conversions. Its signal quality bar is a Conversions API setup with an event match quality score above 5, or a supported data source sending a lower-funnel or app install event.

How to pick an attribution model for your data and spend

The right attribution method is the one your data can support today; add the next one when you clear its entry cost. The order below is this article's recommendation, built on the requirements in the table above.

1. Use each ad platform's numbers inside that platform only. Compare Meta ad sets with each other and Google campaigns with each other. Meta's attribution page warns that results cannot be compared across ad sets with different attribution models.
2. Choose one store-side model for cross-channel reports, and name it on every report. Shopify's default, last non-direct click, is a reasonable start, because it credits a direct visit only when nothing else touched the order.
3. Track the blended ratio every week. Total revenue over total ad spend needs no model, so it is the check on every figure above it.
4. Run a lift test once one campaign qualifies. Meta's guide is a campaign started in the past year with USD 5,000 of spend and 500 conversions. Use the result to see how far Meta's own figure sits from what the ads caused.
5. Consider a marketing mix model once you hold two years of weekly spend and sales by channel. Robyn's guide says to allow at least four weeks just for collecting the data.

A brand that advertises on one platform needs only the first step and the blended ratio. Attribution choices start to matter once two or more paid channels compete for the same buyer.

Where overads fits

overads is an AI marketing agent for small teams: it drafts posts, reports and recommendations from your connected accounts and waits for a person to approve them. On the ad side, overads reads Meta, Google, X and Snapchat ad accounts read-only. It stores each platform's figures as that platform attributed them, and it does not re-attribute them across channels. The table records what its sync took from each platform in the code on October 17, 2026; X Ads, read since October 23, 2026, is not in it.

What overads does not do:

• It has no multi-touch or data-driven model of its own, so it cannot split one sale between Meta and Google Ads.
• It does not read store orders, so the blended ratio stays a spreadsheet job.
• It does not run lift tests or marketing mix models.
• It cannot change ads. Access is read-only, for the reasons in our guide to why many ad tools can read campaigns but not change them, and any change it suggests is applied by a person in the ad platform.
• It does not read LinkedIn, Pinterest or TikTok ad accounts.

A team that needs one attribution model across every channel is better served by a dedicated attribution tool, or by a marketing mix model such as Meridian or Robyn fed from a data warehouse; our guide to whether you need an attribution tool sets out what the vendors publish on price and where one starts to pay. The overads analytics API returns spend and results per day, split by account, platform and currency, with a missing figure as null rather than 0. That covers the media side of a weekly model input for Meta, Google, X and Snapchat; the sales side still comes from your store. For how reporting tools compare on Meta, Google and Snapchat, see cross-platform ad reporting tools compared.