Multi-touch attribution in 2026: what it can and cannot see
Multi-touch tools credit only what they can see: Safari, Apple's tracking prompt and declined consent each remove visits, and credit along a path is not cause. Measure your own gap between Shopify orders and GA4 purchases.
€99 once, excl. VAT. Full refund within 30 days, no questions asked. You keep the read.
By Joris van Huët, Founder & CEOPublished 4 min read
Run the numbers for your store: the free customer journey credit calculator.
Multi-touch attribution can credit only what it can see, and the limits are documented: Safari deletes cookies created in JavaScript after 7 days of no user interaction with the site, Apple has required apps to ask permission before tracking since iOS 14.5, and GA4 links no persistent identifier to events from visitors who decline consent. Credit along a path is also not cause: across 15 Facebook experiments, observational methods often failed to reproduce the randomised result (Marketing Science, 2019).
What do Safari, Apple and consent take away from a multi-touch tool?
WebKit says its tracking prevention detects link decoration, such as click IDs added to URL parameters, and caps the expiry of cookies created in JavaScript on the landing page to 24 hours. It also deletes all cookies created in JavaScript, and all other script-writeable storage, after 7 days of no user interaction with the site. A tag that keeps its visitor ID in such a cookie can't link a visit to one made after the gap, so the path restarts.
Apple's documentation says that starting in iOS 14.5 apps must ask permission before tracking activity across other companies' apps and websites, and that if a user chooses Ask App Not to Track the developer can't access the system advertising identifier (IDFA). Tools that match on that identifier have nothing to match for those users.
Google says that when users don't grant consent, events are not associated with a persistent user identifier: if GA4 collects 10 page views, it can't report whether that is 10 users or 1. How much of your traffic is consented is partly a design choice: in a field experiment with 40 participants, removing the opt-out button from the first page of a consent notice raised consent by 22 to 23 percentage points (Nouwens and colleagues, CHI 2020). GA4 can model the missing behavior only when a property meets criteria that include at least 1,000 events a day with analytics_storage='denied' for at least 7 days and at least 1,000 daily users with it granted on at least 7 of the previous 28 days. Modelled data also isn't available in data export such as BigQuery.
What survives is what your own store records with the visit: orders in Shopify, UTM-tagged landing pages and sessions where consent was given. Within those limits, path reports are a fair way to see which channels appear early or late in the journeys you can observe.
What do experiments say about observational attribution?
Gordon and colleagues compared 15 US advertising experiments at Facebook, with 500 million user-experiment observations and 1.6 billion ad impressions, against observational models, and report that the observational methods often fail to produce the same effects as the randomised experiments, even after conditioning on extensive demographic and behavioural variables (Marketing Science, 2019). A larger follow-up with 663 experiments found median experimental lifts of 29%, 18% and 5% for upper, middle and lower funnel outcomes, against 83%, 58% and 24% from double/debiased machine learning using over 5,000 user-level features (Marketing Science, 2023).
Both studies cover Facebook advertising and test statistical models, not any vendor's product, and this page cites no published test of a named multi-touch platform against experiments. They show what can go wrong when credit along a path is read as cause, and they are a reason to check any platform with a holdout.
How do you measure your own blind spot for free?
- Count orders twice. For one full week that ended at least a week ago, note Shopify orders and GA4 purchases for the same dates and time zone.
- Explain the gap. Subtract cancellations, test orders and time-zone differences.
- Read what is left as orders your browser-side tracking never saw.
- Check modelling. In GA4, Admin, Data display, Reporting identity shows whether Blended is selected, which includes modelled data.
Pass: add the unexplained gap to each channel in turn, and the ranking of your channels doesn't change, so the paths are safe to rank on. Fail: the ranking flips, so use the tool's numbers to describe journeys, not to move budget, and confirm with a holdout.
Sources, 30 September 2026: Tracking Prevention in WebKit (WebKit, Apple); If an app asks to track your activity (Apple Support); Behavioral modeling for consent mode (Google, Analytics Help); Dark Patterns after the GDPR (Nouwens and colleagues, CHI 2020, peer-reviewed); A Comparison of Approaches to Advertising Measurement (Gordon, Zettelmeyer, Bhargava and Chapsky, Marketing Science, 2019, peer-reviewed); Close Enough? (Gordon, Moakler and Zettelmeyer, Marketing Science, 2023, peer-reviewed).
Related answers
Frequently asked questions
Does multi-touch attribution still work after Apple's tracking changes?
It works for what it can observe. Under Apple's App Tracking Transparency rules, apps must ask before tracking, and users who choose Ask App Not to Track withhold the advertising identifier. Compare a tool's orders with your store's orders to see how much it misses.Does Google Consent Mode fill in the missing conversions?
Partly. Google says GA4 models the behavior of users who decline cookies, but only when a property meets data thresholds, and modelled data isn't available in exports such as BigQuery. Tools built on those exports don't get the modelled users.Is multi-touch attribution accurate?
Credit along a path is not a measured effect. In published Facebook experiments, observational methods often failed to match randomised results, even with rich data. This page cites no test of a named platform, so check any tool against a holdout.
Go deeper: Incrementality testing, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Keep reading
Terms in this article
- Ad ImpressionAd Impression is a single instance of an advertisement displaying on a webpage. Impressions are a key input for models measuring the causal impact of ad exposure on user behavior.
- Attribution ModelAn Attribution Model defines how credit for conversions is assigned to marketing touchpoints. It dictates how marketing channels receive credit for sales.
- Dark PatternsDark patterns are user interfaces designed to trick users into unintended actions. These deceptive practices should be avoided.
- IncrementalityIncrementality measures the true causal impact of a marketing campaign. It quantifies the additional conversions or revenue directly from that activity.
- Incrementality TestingIncrementality Testing measures the additional impact of a marketing campaign. It compares exposed and control groups to determine causal effect.
- Landing PageLanding Page: A single web page that appears after clicking a search result, marketing promotion, email, or online advertisement.
- Machine LearningMachine Learning involves computer algorithms that improve automatically through experience and data. It applies to tasks like customer segmentation and churn prediction.
- Multi-Touch AttributionMulti-Touch Attribution assigns credit to multiple marketing touchpoints across the customer journey. It provides a comprehensive view of channel impact on conversions.