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

Mobile & App Marketing › Mobile Measurement

Deep Links and App Analytics Events

How deep links and app analytics work together: universal and app links, deferred deep links, an event taxonomy for apps, and retention reporting.
Guideintermediate10 min readUpdated

On the web, a link is an address and a page view is recorded by default. In an app, neither is automatic. A link has to be registered with the operating system before it opens the app instead of a browser, the app has to know how to route it to the right screen, and every action worth measuring has to be named and sent deliberately. Teams that treat these as engineering details discover the cost later, when campaigns cannot be compared and retention cannot be broken down.

Links matter because every marketing channel that reaches an app user ends in one: the email button, the push notification, the social ad, the QR code in a shop. If the link opens the right screen with the right item, the person continues; if it opens the home screen, many stop. Links also carry the source, through campaign parameters, that analytics needs to split behaviour by channel.

Events matter because they are the raw material of every metric an app is judged by. Activation, retention, conversion and lifetime value (LTV) are all computed from events. In the AARRR model, every stage after acquisition is an event or a sequence of them, so a poor taxonomy makes the whole model unreadable. The mobile measurement hub shows how links and events fit beside attribution.

There are three kinds of link that open an app, and they differ in reliability and fallback behaviour.

Link typePlatformHow it worksFallback if app absentMain weakness
Custom URL scheme (myapp://product/123)iOS and AndroidThe app registers a scheme; the OS opens the app for that schemeError or nothingNo web fallback; any app can claim a scheme; blocked or unsupported in many contexts
Universal linkiOSAn ordinary https link on a domain the app has verified with an association fileOpens the web pageBreaks if the association file or domain setup is wrong; some in-app browsers do not honour it
Android App LinkAndroidAn ordinary https link on a domain verified with a digital asset links fileOpens the web pageVerification failure silently falls back to the browser or a chooser
Deferred deep linkBothA link service records the destination, sends the person to the store, and passes the destination to the app on first openStore, then the intended screenNeeds a link service or MMP; matching can be limited by platform privacy rules

Universal links and App Links are the default choice for anything marketing sends, because they behave like normal web links when the app is missing and open the app directly when it is present. The web page at the same address should be a real, useful page, not a dead end: a person without the app should still be able to see the product or content.

Routing is the other half. The app needs a route table that maps each link path to a screen and handles the awkward cases: the person is logged out, the item no longer exists, the link was built for a screen that has since been renamed. Marketing needs a readable copy of the route table, because every link a campaign sends depends on one of its rows. Test links in the places people will actually tap them, including email clients and the in-app browsers of social apps, which often behave differently from the system browser. The tutorial on testing app deep links before a campaign turns that advice into a test matrix of platforms, app states and tap contexts.

A deep link only works if the app is installed. For someone who taps an ad or a shared link without the app, the journey runs through the store, and the store does not pass the destination to the app. Deferred deep linking closes that gap: a link service records what the person tapped, sends them to the store, and when the app opens for the first time it asks the service whether this install followed a link and, if so, where it was meant to go.

The payoff is a first session that continues the story the link started. A person who tapped an ad for a specific recipe opens the app on that recipe, not on a generic sign-up screen. A person invited by a friend lands on a screen that knows who invited them. Onboarding can then be shortened, because the context is already known.

Matching a first open to an earlier tap is subject to the same platform limits as install attribution, so deferred links work best where a deterministic signal exists (the Android install referrer, or a clipboard or login-based method the platform permits) and less reliably otherwise. Most mobile measurement partners and dedicated link services provide deferred deep linking alongside attribution, which keeps the link that was tapped and the install it produced in one record. Design a sensible default first-run experience for the cases where the destination cannot be recovered.

Those journeys start from the campaign links on the store listing, which the app store listing checklist checks before each listing update.

An event taxonomy for apps

An event taxonomy is the agreed list of events an app sends, their names, their parameters and what each one means. It is the app's part of the tracking plan, and it should be written before instrumentation starts, from the questions in the measurement plan rather than from the list of screens.

LayerPurposeExamplesNotes
Automatically collectedSessions, first opens, screen views, app updatesfirst_open, session_start, screen_viewCollected by the SDK; do not rebuild them
ActivationThe moment a new user gets the app's core value for the first timetutorial_complete, first_lesson_complete, account_fundedOne per app, agreed across teams; sent to the MMP and networks
Core valueThe repeated actions that show the app is being used for its purposeworkout_logged, search, shareA handful at most
Commerce and subscriptionMoney movingadd_to_cart, purchase, subscription_startUse the platform's recommended names and parameters
Engagement with marketingMessages and links acted onnotification_open, deep_link_openCarry campaign parameters as event parameters

Use the recommended event names that Google Analytics and Firebase document wherever one exists, because reports and integrations understand them; add custom events only for behaviour the recommended set does not cover. Name events consistently (one case, one tense, object and action in a fixed order), keep parameters few and typed, and record every event in a shared document with its definition and owner. The analytics section's guide to events and parameters explains the GA4 event model in detail; the same model applies to app data streams.

Links and events meet in the campaign parameters. Tag every marketing link with UTM parameters or the equivalent, pass them through the deep link into the app, and record them on the session or the first event, so that in-app behaviour can be split by source. For paid installs, the source comes from the measurement partner's Attribution instead, and the two should use the same naming.

User properties and audiences

User properties describe the person rather than the action: subscription tier, preferred language, acquisition source, the date of activation. They let reports and audiences segment behaviour without repeating the same parameter on every event. Keep them few, stable and free of personal data that the privacy notice does not cover; Google Analytics, for example, does not permit personally identifiable information to be sent as event or property values.

Audiences are built from events and properties together: people who activated but have not returned in a week, people who added to cart without purchasing, subscribers on a monthly plan. They feed the messaging tools that send push and in-app messages, the ad networks for retargeting and exclusion, and the experiments that test changes on a defined group. The push notifications and in-app messaging guide covers how such audiences are used for messaging.

Two practical rules keep audiences useful. Define each one in plain language before building it, so the definition outlives whoever created it. And set membership durations deliberately, since an audience that keeps people forever slowly fills with users who no longer match the intent.

Retention and cohort reporting

Retention is the metric that decides whether an app's acquisition spend is worth it, and it can only be read properly by cohort. A cohort is a group of users who share a starting point, usually the week they installed or activated; the retention report then shows what share of each cohort returned on each later day or week. Blended figures (all active users this month, say) mix new and old users and hide whether recent cohorts are better or worse than earlier ones.

Read cohorts by source as well as by date. A paid network that brings many installs with weak retention can look fine in a blended active-user count and poor in its own cohort curve. Read them against changes, too: a new onboarding flow, a price change or a messaging programme should show up as a shift in the cohorts that started after it. The growth section's guide to cohort analysis explains the method and how it feeds lifetime value estimates.

Define retention precisely before reporting it. Choose whether "retained" means any session or a core-value event, whether the measure is the share active on a given day or within a range, and which starting event defines the cohort. Firebase and Google Analytics both offer cohort and retention views, and most teams also export events to a warehouse for analysis the built-in reports cannot do.

Illustrative example — not a real result

Juniper Recipes fixes its links and finds its best channel

Juniper Recipes, an invented cooking app, sends 30,000 people a month from Instagram posts to recipes. Its links open the app's home screen, and only 12% of those who tap view a recipe. It switches to universal links and App Links routed to each recipe, with deferred deep links for new installs, and tags every link with campaign parameters carried into the app. Recipe views after a tap rise to 58%. Cohort reports by source then show that Instagram installs retain at 31% by week four against 17% for a paid network the team had been scaling. Juniper moves £5,000 a month from the paid network to creator partnerships on Instagram.

Deferred links are the joint in every invite cycle, which is why the guide to growth loops built for app installs gives the link its own owner and arrow.

The events defined here feed a driver tree, and the KPI tree for app installs shows which leaves come from app analytics and which are modelled under platform privacy rules.

The in-app events defined here become optimisation goals in the TikTok Ads setup for app installs, which moves delivery from installs to an event that predicts retention.

App experiments depend on this reporting: the push permission prompt test, for example, reads opt-in and early retention by install cohort, with each user's arm logged as an event. The web-to-app switch test relies on the shared user identifier instead, joining web and app orders so that a prompt on the website is read on both surfaces at once.

Common mistakes

  • Custom schemes for marketing links. They fail with no fallback when the app is missing. Use universal links and App Links with a real web page behind them.
  • Untested link routes. Links break when screens are renamed or users are logged out. Keep a route table and test it in email clients and in-app browsers with every release.
  • No deferred deep links. New users from a specific link land on a generic start screen. Carry the destination through the install.
  • Events named by whoever built the screen. Inconsistent names make reports unusable. Write the taxonomy first and enforce it.
  • Tracking everything. Excess events add cost and confusion. Track what the measurement plan needs.
  • No agreed activation event. Without it, attribution postbacks, onboarding and retention all measure different things. Agree one and use it everywhere.
  • Campaign parameters dropped at the app boundary. Behaviour cannot be split by source. Pass parameters through the link into the first session.
  • Blended retention. An overall figure hides cohort trends and weak sources. Report retention by cohort and source.

Frequently asked questions

What is the difference between a deep link and a deferred deep link?

A deep link opens a specific screen in an app that is already installed. A deferred deep link remembers the intended destination when the app is not installed, sends the person to the store, and then opens that destination on first launch after install. Deferred linking needs a link service or measurement partner to carry the context across the install.

How many events should an app track?

As few as answer the questions in the measurement plan, defined consistently. Most apps need an activation event, a handful of core-value events, the commerce or subscription events that represent money, and the standard screen and session events the analytics SDK records automatically. An event that nobody will use in a report or audience is a cost, not an asset.

Can Google Analytics measure apps as well as websites?

Yes. Google Analytics collects app data through the Firebase SDK into app data streams, and the same property can hold web streams, so app and web behaviour can be reported together. The event model is shared, with automatically collected events, recommended events and custom events, as the Google Analytics and Firebase help pages document.

Terms used on this page

Frameworks

Where this page is used

Also related

Sources

  1. Firebase documentation — Google
  2. Google Analytics Help — Google Analytics Help

Related