Short answer: Mobile app attribution is highly accurate where a deterministic signal links a click to an install, and it is an estimate where that signal is missing. No vendor publishes a single accuracy figure that holds across apps, so the useful question is which of your installs were matched deterministically and which were modelled.
What is mobile app attribution?
Mobile app attribution is the process of assigning each app install and in-app event to the marketing source that caused it, usually by matching an ad click or view to the first app open. A mobile measurement partner (MMP) does the matching and picks one source per install.
How accurate is mobile app attribution?
It is exact for installs that carry a platform-provided signal and uncertain for the rest, and the share of each differs by platform, channel and consent. Three things decide the answer for your app:
- Matching method: deterministic or probabilistic.
- Platform rules: what Apple and Google allow you to see.
- Windows and rules: how long after a click or view an install can still be credited.
Treat any single accuracy percentage as a claim until it names its method and data.
Deterministic or probabilistic matching: what is the difference?
Deterministic matching: a match made on an identifier or token that is unique to the click and the install, so the result is a yes or no. Probabilistic matching: a match made on shared signals such as IP address and device model, so the result is a likelihood.
- Deterministic examples: the Google Play install referrer on Android, a click ID passed through a deep link, or a device identifier the user has allowed.
- Probabilistic examples: matching on IP and user agent when no token exists, for instance on iOS when a user has not consented to tracking.
- Where it breaks: shared Wi-Fi, VPNs and carrier networks put many devices behind one IP.
Methods differ by vendor, and the vendor's documentation is the place to check. Linkrunner states it is deterministic-first, using IP and user agent matching only as a fallback when no deterministic signal exists. Ask any vendor you are comparing for the same split, because it tells you how much of your data rests on estimates.
What does Apple's SKAdNetwork limit on iOS?
SKAdNetwork (SKAN) limits iOS attribution to aggregated, delayed and privacy-thresholded data, so you cannot trace an individual user. Apple's documentation says the signed postback "doesn't include user- or device-specific data" and that values are included only if they meet its privacy threshold.
- Up to three postbacks: from iOS 16.1, an advertised app can update its conversion value during up to three conversion windows, which gives up to three postbacks per install.
- Postback data tier: Apple's documentation says the device assigns a tier that determines how much detail the postback can contain, which helps keep installs anonymous within a crowd.
- Consent not required for SKAN: apps can call these APIs regardless of tracking authorisation status.
For you this means small campaigns can return little detail, and iOS reporting reads as a trend rather than a user-level record. Compare SKAN with MMP-measured installs from consented users, and do not expect an exact match.
How reliable is install referrer on Android?
Android is generally the more precise platform because Google Play passes referral data to the app. Google's documentation says the Play Install Referrer API "securely" retrieves referral content from Google Play, that the data is available for 90 days and does not change unless the app is reinstalled.
- What it carries: the referrer string, the click timestamp and the install-begin timestamp.
- Limit: it only covers installs from Google Play, so other app stores and sideloaded installs need another method.
- One call: Google advises reading it once, on first launch after install.
Why do click and view windows change the numbers?
An attribution window is the period after a click or view during which an install can be credited to that ad. Short windows credit fewer installs and long windows credit more, so the same campaign gives different totals under different settings.
- Click-through: an install credited after a tap.
- View-through: an install credited after an ad impression with no tap. It is the most debated type, because the user may have installed anyway.
- Last-click: the usual rule, which gives the install to the most recent qualifying click, so other networks lose installs they also touched.
Linkrunner supports configurable last-click attribution and view-through, which lets you change the window and see the effect on your own installs.
Why do network and MMP numbers differ?
Ad networks count every install they touched within their own windows, including modelled iOS conversions, while an MMP deduplicates across all channels and gives each install to one source. The full list of causes and fixes is in why Meta and your MMP report different installs.
How to sanity-check your attribution
- Compare windows first. Put the network and MMP on the same click and view windows and the same time zone.
- Split iOS and Android. A large gap on iOS is expected, a large gap on Android usually points to a setup error.
- Add up the sources. Paid, organic and referral installs from your MMP should match total store installs within the lag you expect.
- Look for sudden shifts. A step change in one week usually follows a release, an SDK change or a fraud pattern. Linkrunner's anomaly alerts flag sudden ROAS drops and install spikes with a suspected cause, and any vendor with a comparable alert helps here.
- Test on your own device. Click a campaign link, install and check the source.
For tool comparisons see the MMP pricing page and the methodology page.
FAQ
Is mobile attribution 100 percent accurate? No. It is exact where a deterministic signal exists and estimated where it does not, and the mix depends on platform, channel and consent.
Is Android attribution more accurate than iOS? Usually, because the Play install referrer gives a direct signal, while iOS relies on SKAdNetwork's delayed and aggregated postbacks.
Why does my MMP show fewer installs than the ad network? The network counts every install it touched, while the MMP credits one source per install. Windows, view-through and modelled conversions add to the gap.
What attribution window should I use? Use the window your decision depends on, and keep it the same when you compare networks. A longer window credits more installs but also more that might have happened anyway.
Can I test whether my attribution works? Yes. Click a campaign link on a test device, install the app and confirm the install lands under the expected source.