Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Mobile & App Marketing

Mobile Measurement

Measuring an app means answering two questions the web answers more easily: where did this user come from, and what did they do afterwards. The first runs through mobile measurement partners and the privacy frameworks of each operating system; the second runs through deep links and an in-app event taxonomy. This hub explains why apps differ from the web, sets the three parts side by side and introduces the core guides that cover each, with further guides on finding the activation event, reconciling install numbers, testing deep links, checking app events on every release and setting an app's lapse threshold.
SectionintermediateUpdated

In this section

Why app measurement differs from web

On the web, a click carries its source with it. The landing page sees the referrer and the campaign parameters, a cookie or session links later behaviour to that visit, and the analytics tool can credit the conversion to the click. An app breaks that chain twice. The click goes to a store, which does not pass campaign parameters to the app it installs, and the first open happens later, in a different environment, with no memory of the ad.

Privacy rules then limit how the chain can be rebuilt. On iOS, linking a user's activity across apps owned by different companies requires the person's permission through the App Tracking Transparency prompt, and Apple provides SKAdNetwork as a privacy-preserving alternative that reports installs to ad networks in aggregate, with delays and limited detail. On Android, the Play install referrer passes some campaign information to the app on first open, and the platform has its own privacy changes under way. The result is that install Attribution on mobile is less precise than web attribution and has to be read with that in mind.

The other half, what users do once inside, is in some ways easier: the app controls every screen and can record events reliably. The difficulty there is design rather than access. Without a deliberate event taxonomy, an app's analytics fills with inconsistent events that nobody can use to compute retention or value.

App measurement has three parts, each with its own tools and its own failure modes.

PartQuestion it answersMain toolsTypical failure
Install attributionWhich campaign or source produced this install, and which events followedMobile measurement partner (MMP), SKAdNetwork, Play install referrer, network SDKsEach network claims its own installs; iOS data delayed and aggregated; no one source of truth
Deep linksDid the tap land on the right screen, and can the app tell which link brought the personUniversal links, Android App Links, deferred deep links through an MMP or link serviceLinks open the home screen or the browser; new installs lose the context of the link
App analyticsWhat do users do after install, and do they come backFirebase or Google Analytics for apps, product analytics tools, the MMP's event feedEvents named inconsistently; no activation event defined; retention not reported by cohort

The parts depend on each other. Attribution needs in-app events to tell networks which installs mattered; links need analytics to show what happened after the tap; analytics needs attribution to split behaviour by source. Planning them together, before launch, saves a painful retrofit later.

The core guides

  1. App attribution: MMPs and SKAdNetwork explains what a mobile measurement partner does, how install and event postbacks reach ad networks, what SKAdNetwork and the iOS tracking permission allow and withhold, how the Android install referrer works, how modelled results and incrementality tests fill the gaps, and how to choose and configure an MMP.
  2. Deep links and app analytics events covers universal links and app links, deferred deep links for people who install first, an event taxonomy designed for apps, user properties and audiences, and retention and cohort reporting.

General attribution theory, including how to choose between rules-based, data-driven and experimental approaches, belongs to the analytics section; its guide to choosing an attribution approach is the place to start if the question is which model to believe. The mobile guides apply that thinking to the constraints of apps.

Further guides

Further guides are tutorials for jobs the core guides assume have been done:

  • Finding your app's activation event — choosing the event from the app's own data: candidate actions, retention by cohort, timing, volume for campaigns, and one shared definition.
  • Reconciling app install numbers across reports — comparing install counts across store consoles, a measurement partner, ad networks and analytics: definitions, a monthly table and a tolerance.
  • Testing app deep links before a campaign — a route table, a matrix by platform and context, logged-out and missing-item cases, deferred links, and campaign parameters.
  • Checking app events on every release — a tutorial for the release build: the events a release touches, clean test devices, the debug view, postbacks to the partner and networks, tracking and consent states, the change log and version adoption.
  • Setting an app's lapse threshold — a tutorial for the number that re-engagement, paid audiences and churn reports share: the core action that counts, the silences between actions, how many come back after each length of silence, checks by season, rhythm and platform, and one written rule.

App data in the measurement plan

An app is one part of a business's measurement, not a separate world. The measurement plan that governs the website and paid media should also name the app's activation event, its retention definition and the events that represent value, so that the same objectives are read the same way on every surface. Where a person uses both the website and the app, a shared user identifier at login is what lets the two be joined. The activation event tutorial shows how to choose that event from the app's own data and write one definition the whole plan uses.

Two numbers decide whether app marketing works: customer acquisition cost (CAC) by source, and the lifetime value (LTV) of users from each source. Attribution supplies the first, imperfectly; analytics supplies the second, by cohort. Neither is trustworthy alone. Because install attribution under privacy constraints is noisy, the plan should also schedule tests that measure Incrementality directly, such as geographic holdouts or paused-spend tests, and read network reports as optimisation signals rather than as the truth. The attribution models framework explains how rules-based credit and experiments relate, and where the store consoles, the measurement partner, the networks and analytics report different install totals, the reconciliation tutorial records what each one counts and the difference the plan should expect.

Finally, the plan should say who owns each tool. The MMP, the analytics SDK and every ad network SDK collect data on users, and each needs to be listed in the app's privacy disclosures and consent flow. The broader mobile and app marketing guide shows where measurement sits in the growth loop.

Illustrative example — not a real result

Ferncastle Bank joins app and web measurement

Ferncastle Bank, an invented challenger bank, reports app installs from its MMP and account openings from web analytics, and the two never meet. It defines one activation event, "account funded", fires it in both the app and the website with the same user ID, and sends it to the MMP as a postback. Over the next quarter it sees that paid social drives 38% of installs but only 21% of funded accounts, while search drives 24% of installs and 41% of funded accounts. Cost per funded account is £96 on paid social and £58 on search, and the bank moves £30,000 of monthly budget towards search before testing paid social again with a regional holdout.

Common mobile measurement problems

SymptomUsual causeGuide
Networks together claim more installs than the store reportsEach network attributes under its own rules; no MMP to deduplicateApp attribution: MMPs and SKAdNetwork, Reconciling app install numbers
iOS campaign results look thin and lateSKAdNetwork postbacks are delayed and aggregated by designApp attribution: MMPs and SKAdNetwork
Event-optimised campaigns never leave the learning phaseIn-app events not sent to networks as postbacks, or sent too lateApp attribution: MMPs and SKAdNetwork
Links in emails and ads open the browser or home screenUniversal links or App Links misconfigured; no fallback routesDeep links and app analytics events, Testing app deep links
New installs from a campaign land on a generic onboardingNo deferred deep linkingDeep links and app analytics events, Testing app deep links
Analytics full of events nobody usesNo event taxonomy or naming conventionDeep links and app analytics events
Retention reported as one blended figureNo cohort reporting by install date and sourceDeep links and app analytics events
Unsure which attribution model to trustTheory question, not a mobile oneChoosing an attribution approach
Each team counts a different activation eventEvent chosen by opinion; no single definition shared by the partner, networks, product and financeFinding your app's activation event
Install totals from the store, the partner and analytics never agreeEach source counts something different; platforms combined; no tolerance set from the app's own historyReconciling app install numbers
A metric steps up or down after an app release with no campaign changeAn event renamed, dropped or sent without its parameters in the new buildChecking app events on every release
Re-engagement contacts users who were only between sessions, or reaches users whose habit has already goneA lapse line borrowed as a number of days, not read from the app's own return dataSetting an app's lapse threshold, Re-engaging lapsed app users

When one of these figures moves and the cause is not yet known, the app marketing troubleshooting section starts from the symptom, checks the measurement first and routes each cause to the guide that fixes it.

Terms used on this page

Frameworks

Sources

  1. SKAdNetwork — Apple Developer
  2. Firebase documentation — Google

Related