How does mobile app attribution work? A mobile measurement partner (MMP) receives click and impression data from ad networks and matches it to the app’s first open by device identifier, install referrer (on Android, via Google Play) or a statistical model. It then applies attribution windows and assigns the install to a source. On iOS, when users decline tracking, installs from ads that support SKAdNetwork or AdAttributionKit are reported as aggregated postbacks.

This guide is for UA leads, growth marketers and app owners who buy traffic from several networks and need numbers they can act on.

Key Takeaways

  • Use the MMP as the single record for budget and payouts. Network dashboards apply their own windows, so their install counts overlap.
  • Pick one click-through and one view-through window per campaign goal and apply them to every network you compare. AppsFlyer and Adjust both default to 7 days for clicks and 24 hours for views.
  • On iOS, report consented users and SKAdNetwork or AdAttributionKit data side by side, and map the event that drives budget into the first conversion window: later windows return only low, medium or high.
  • On Android, attribution stays mostly deterministic: the store referrer and the advertising ID still carry it, and Google is retiring the Privacy Sandbox APIs once planned to replace cross-app IDs.
  • Before disputing a gap with a network, reconcile in a fixed order: report setup first, then attribution rules, then data quality.

What Is Mobile App Attribution?

Mobile app attribution decides which ad gets credit, and payment, for each new user. Every budget shift, bid change and partner invoice depends on that decision. A UA team therefore needs to know where the record is deterministic, where it is modeled and where it is missing.

Mobile app attribution is the process of linking an app install, and the in-app actions that follow it, to the ad click or impression that led to it. The result is a source for each user: network, campaign, ad set and creative. Teams use it to compare channels, optimize campaigns and calculate partner payouts.

Mobile attribution answers three practical needs. The first is budget allocation: spend moves toward sources whose users register, deposit or subscribe. The second is campaign optimization, because ad networks bid better when they receive conversion signals for the users they brought. The third is payment. On cost-per-install (CPI) and cost-per-action (CPA) deals, the attributed record is the invoice, so a contested attribution becomes a payment dispute.

Attribution assigns credit; it does not prove that the ad caused the install. That distinction matters later, when incrementality tests enter the picture.

How Mobile App Attribution Works

Attribution is a matching job. Networks report touchpoints, the app reports first opens, and the MMP pairs them under a set of rules. Most disputes about mobile app attribution come from one of the six steps below, so it helps to know which party controls each one.

Figure 1. How a paid install is attributed: from ad click or impression to postback and BI

  1. Touchpoint. A user clicks or views an ad. The network records the click or impression with a timestamp, campaign and creative.
  2. Touchpoint reaches the MMP. Ad networks pass it through an attribution link (a tracking URL with campaign parameters). Self-attributing networks keep it on their side and answer the MMP’s query instead. Apple’s frameworks carry it inside a signed ad.
  3. Install and first open. The user installs from the App Store or Google Play and opens the app. The MMP SDK inside the app sends an install event with the identifiers available on that device.
  4. Matching. The MMP looks back across all touchpoints for that device within the attribution window and picks a winner, normally the last click.
  5. Attribution. The install is assigned to a source, or recorded as organic if no qualifying touchpoint exists. Later in-app events inherit that source.
  6. Postbacks and reporting. The MMP sends server-to-server (S2S) postbacks, which are automatic notifications about installs and events, back to the winning network so it can optimize. The same data flows into dashboards, raw data exports and the advertiser’s BI. Payouts on CPA deals run on this step; see how postbacks work between agency and advertiser.

Deterministic vs Probabilistic Matching

Deterministic matching pairs a touchpoint and an install through a shared identifier: the IDFA (Apple’s Identifier for Advertisers), the GAID (Google Advertising ID) or the Google Play install referrer string. When the identifier matches, the touchpoint and the install are known to come from the same device. That makes the match reliable, but it does not show that the ad caused the install.

Probabilistic matching (also called probabilistic modeling) estimates the match from non-unique signals such as timing, IP address and device type. AppsFlyer describes it as a statistical fallback with a click lookback of up to 24 hours, used only when deterministic methods are unavailable. Adjust ranks it the same way, as its secondary method.

On iOS the boundary is set by Apple policy. Apple’s User Privacy and Data Use page states that developers may not derive data from a device to uniquely identify it. Apps that reference SDKs doing so may be rejected from the App Store. Apple does not name individual MMP modeling methods in that page, so each MMP’s iOS approach should be checked against Apple’s current wording before a contract is signed.

Mobile Measurement Partners (MMP): What They Do

A mobile measurement partner is an independent attribution service that sits between the app and every ad network. It applies one set of rules to all sources, removes duplicate claims and gives the advertiser a single record of which network brought which user. That single record is why the MMP, and not any network dashboard, is the source of truth for budget.

Each network’s dashboard reports conversions by its own rules: its own windows, its own view-through policy and its own idea of which touchpoint counts. Two networks can claim the same install, and each is correct by its own logic. The MMP resolves that overlap once, for all sources, and only one network wins each install.

Self-attributing networks (SANs, which AppsFlyer calls self-reporting networks or SRNs) are the main exception to the attribution-link model. Meta, Google Ads, TikTok and Apple Ads do not pass clicks through a tracking URL. AppsFlyer’s SRN documentation explains that the MMP notifies the SAN of an install and the SAN replies whether it had a qualifying touchpoint. The MMP still makes the final decision, but the network’s own report will keep showing its own windows.

How to Choose an MMP

Comparing mobile app attribution tools comes down to fit with the channel mix. Check every mobile app attribution platform on the shortlist against these criteria before price:

  • Network integrations: native integrations with every network and SAN in the current and planned media plan, including cost data import.
  • iOS privacy support: SKAdNetwork and AdAttributionKit postback decoding, conversion value schema management and reporting next to user-level data.
  • Fraud protection: real-time blocking and post-attribution rejection, with reasons visible per install.
  • Raw data access: user-level exports and APIs that reach the BI without manual downloads.
  • Configurable rules: attribution windows, re-attribution and re-engagement settings per network.
  • Pricing model: per attributed install, per event or flat fee, and how it scales with organic traffic.

Attribution Models in Mobile Marketing

A mobile app attribution model is the rule that decides how credit is split between touchpoints. Mobile defaults to last click because installs pass through an app store, which breaks the user path. The MMP sees the touchpoints and the first open, but nothing in between that could support weighting several ads at once.

Table 1. Attribution models and how they apply in mobile

ModelHow it creditsWhere it fits in mobileLimitations
Last click100% to the last click inside the click-through windowDefault in MMPs; basis for CPI and CPA payoutsIgnores earlier ads; rewards networks that are present at the end of the path
Last touch with view-throughLast click wins; if there is no click, the last impression inside a shorter view-through windowVideo, playable and programmatic campaigns where users rarely clickEasy to over-credit high-frequency display; windows differ by network
Multi-touchSplits credit across several touchpoints by position or weightsAnalysis of assisting networks, not payoutsNeeds user-level paths, which ATT and SKAN remove on iOS
Incrementality testingCompares an exposed group with a holdout that saw no adsValidating a channel, a retargeting program or a large SANNeeds volume and a clean test design; answers one question per test
MMMRegression of outcomes on spend by channel over timeBudget split across channels, including those without click dataNeeds long, varied spend history; weekly or monthly rather than daily

Last click survives in mobile because it is the only model that every party can verify against the same record. View-through attribution matters most for programmatic advertising for mobile apps, where the impression often does the work. Multi-touch has lost most of its ground on iOS since user-level paths disappeared. Incrementality and marketing mix modeling (MMM) answer a different question: they estimate what a channel added, which attribution alone cannot show.

Attribution Windows: How to Set Them

An attribution window, or lookback window, is the time between a touchpoint and an install during which that touchpoint can win credit. Click windows usually run longer than view-through ones. The defaults below are a starting point; the right window depends on how long users in a given vertical take to install after seeing an ad.

Table 2. Default attribution windows in AppsFlyer, Adjust and AdAttributionKit as of November 2026

SettingAppsFlyer defaultAdjust defaultApple AdAttributionKit default
Click-through window7 days (range 1–30 days)7 days with device matching (range 1–30 days)30 days (configurable 1–30 days)
View-through window24 hours (range 1–24 hours)24 hours with device matching1 day (configurable 1–7 days)
Probabilistic click windowUp to 24 hours6 hours (range 1–24 hours)Not applicable
Re-attribution or inactivity90-day re-attribution window7-day inactivity period before re-attributionNot applicable

Two more windows govern returning users. The re-attribution window is the period after the first install during which a reinstall is not counted as a new install; AppsFlyer sets it to 90 days by default. The re-engagement window is the period after a retargeting conversion during which in-app events are credited to the retargeting campaign; AppsFlyer’s retargeting guide gives a 30-day default. Both decide who gets paid for users who come back, which is why they belong in any app retargeting setup from day one.

Windows change the comparison between networks. A network judged on a 7-day click window and a 24-hour view-through window will look weaker than in its own dashboard if it reports on longer defaults. AppsFlyer lists a 28-day default click window for Snapchat and 30 days for Apple Search Ads (now Apple Ads), against 7 days for Meta ads. Set one window per campaign goal in the MMP, apply it to every source, and change it only when time-to-install data shows a reason.

Attribution on iOS: ATT, SKAdNetwork and AdAttributionKit

On iOS 14.5 and later, apps need the user’s permission through App Tracking Transparency to read the IDFA. Users who allow tracking can be attributed at user level. For everyone else, installs from ads that support SKAdNetwork or AdAttributionKit reach the ad network as aggregated, delayed postbacks with a campaign identifier and a conversion value, without any user or device data.

App Tracking Transparency (ATT)

The ATT framework appeared in iOS 14, and Apple’s privacy page makes the permission mandatory from iOS 14.5 for tracking or for access to the advertising identifier. Without permission, the IDFA is returned as all zeros. Apple’s definition of tracking explicitly includes advertising measurement, which is why consented and non-consented users end up in two different reporting streams.

Figure 2. iOS attribution with and without ATT consent

SKAdNetwork 4

SKAdNetwork (SKAN) is Apple’s privacy-preserving attribution API. According to Apple’s documentation, apps can call it regardless of their tracking authorization status. Since iOS 16.1, version 4 sends up to three postbacks per install, one for each conversion window: days 0–2, days 3–7 and days 8–35 after the first launch (Apple).

Each postback carries a conversion value that the advertiser defines. The first postback can hold a fine-grained value from 0 to 63 (Apple); the second and third carry a coarse value of low, medium or high. Postbacks arrive after a random delay of 24–48 hours for the first window and 24–144 hours for the later ones.

Crowd anonymity limits the detail further. Apple assigns each install a postback data tier based on how many installs share the same parameters. At Tier 0 the network receives only a two-digit source identifier and no later postbacks. Higher tiers add a source identifier of up to four digits, the fine value and, at Tier 3, the publisher app and country code. Small campaigns therefore get thinner data than large ones.

AdAttributionKit

AdAttributionKit (AAK) is Apple’s newer attribution framework, available from iOS 17.4. It keeps the SKAN model of signed ads, three conversion windows, crowd anonymity and conversion values. On top of that, it covers ads in alternative app marketplaces and re-engagement conversions for users who already have the app. In iOS 18.4, Apple added configurable attribution windows and cooldowns, conversion tags for overlapping re-engagements and country codes in postbacks (Apple, WWDC25 session).

The two frameworks work together. Per Apple’s interoperability guide, the system ranks SKAN and AAK impressions in one pool, clicks beat views, the most recent wins, and only one impression can win a conversion. Apps integrated with both frameworks call both update APIs.

Table 3. What iOS attribution returns with and without ATT consent

Tracking allowed (IDFA available)Tracking declined or not asked
IdentifierIDFA shared with the app, MMP and networksIDFA is all zeros
MatchingDeterministic, user-levelSKAN or AAK postback to the network; MMP decodes the network’s copy
GranularityCampaign, ad set, creative, userCampaign-level source identifier; detail set by crowd anonymity tier
Post-install dataAll in-app events and revenue, by cohortOne conversion value per window (fine in window 1, coarse later)
TimingNear real time24–48 hours after window 1; up to 144 hours after windows 2 and 3
Apple AdsAttribution via the AdServices API (iOS 14.3+)AdServices attribution token plus SKAN or AAK

Attribution on Android

Android attribution is mostly deterministic. The Google Play Install Referrer API passes the referrer and timestamps of the click that led to an install, and the GAID is available unless the user deletes it. With Google’s Privacy Sandbox advertising APIs on Android being retired, the referrer and GAID remain the working signals.

The Play Install Referrer API returns the referrer URL of the install, the click timestamp, the install begin timestamp and the app version at first install. Because the referrer comes from Google Play itself, it is the strongest attribution signal on Android: the MMP reads campaign parameters straight from the store record. The two timestamps also expose fraud, as the fraud section below shows.

The GAID is a user-resettable and user-deletable advertising ID provided by Google Play services. According to Google’s Play Console help, apps targeting Android 13 or later must declare the AD_ID permission, and after a user deletes the ID, apps receive a string of zeros. Play policy also requires apps to use the advertising ID instead of other persistent identifiers for ad purposes.

Privacy Sandbox on Android was Google’s planned replacement for cross-app identifiers. On October 17, 2025, Google announced the retirement of the Attribution Reporting API, Topics and Protected Audience on Chrome and Android, along with SDK Runtime, citing low adoption. As of November 2026, Google’s feature status page marks all four for phase-out and names no successor API.

Cross-Channel Attribution for Mobile Apps

Cross-channel attribution means judging several networks against one rulebook. The MMP does the first part by deduplicating claims, but self-attributing networks, view-through policies and SKAN data still overlap. When the overlap is large, an incrementality test shows what each channel adds.

In practice, a UA team running Meta, Google, TikTok, Apple Ads and programmatic will see the sum of network-reported installs exceed the MMP total. Each SAN claims installs it touched within its own windows, including users who also clicked another ad later, so the sum of network dashboards overstates the real total.

Two habits keep the picture honest. Report view-through installs on their own line, so display and video networks are not ranked by numbers that click-based networks cannot earn. When one network grows its share of installs without a matching change in total installs, run a holdout or geo test before moving budget toward it.

For deposit-driven funnels, see cross-channel attribution models in iGaming.

Why Attribution Data Doesn’t Match (and How to Reconcile It)

Network dashboards, the MMP and internal analytics count different events at different moments under separate rules. A gap between them is expected. The job is to remove the explainable part of the gap step by step and raise only the remainder with the network, backed by click and install records.

Table 4. Common attribution discrepancies

DiscrepancyTypical causeWhat to check
Network shows more installs than the MMPLonger network windows or view-through claims the MMP does not acceptWindow settings on both sides; split click and view-through
Numbers shift between daysNetwork counts on click date, MMP on install or first-open dateDate basis of each report
Daily totals differ, weekly totals matchDifferent time zonesReport time zone in network, MMP and BI
Network installs exceed store installsNetwork counts installs without a first openInstall definition (download vs first open)
MMP installs are lower after fraud filtersReal-time blocked or post-attribution rejected installsFraud report and rejection reasons by sub-publisher
Reinstalls counted as new usersRe-attribution window shorter than the user’s gapRe-attribution and inactivity settings
iOS totals far apartSKAN or AAK postbacks reported separately, with delay and privacy thresholdsSKAN report vs network SKAN report for the same dates
Events and revenue differ from BIDifferent event names, currencies, refunds or test usersEvent mapping and revenue validation rules

Figure 3. Reconciliation order: network report vs MMP

Work through the comparison in a fixed order, because each step changes the numbers the next one uses. Start with report setup: the same date range, time zone, app, platform and event. Then align attribution rules: identical click and view windows, with view-through and SAN claims shown separately. Then remove data-quality effects: fraud rejections, reinstalls and SKAN totals, which belong in their own comparison.

Example: a payments app sees 1,240 installs in a network dashboard and 1,010 in the MMP for one week. Moving the network report from click date to install date and excluding view-through claims leaves 1,060. Rejected installs account for 30 more. The remaining 20 installs are the real dispute, and the network can check them against click IDs.

Any gap that survives these steps goes to the network with install-level evidence. If the advertiser and the network agree on a tolerance for that gap in advance, it belongs in the insertion order. Payouts then follow the MMP record without monthly renegotiation.

Attribution and Ad Fraud

Ad fraud on mobile mostly targets attribution itself: fake clicks and fake installs are designed to win last-click credit and the payout that comes with it. Attribution data is also where fraud shows first, in click-to-install timing and in installs that never produce real activity.

The common schemes each attack a different step of the match:

  • Click flooding (click spamming) sends large volumes of fake clicks so that one of them sits inside the window when an organic install happens (AppsFlyer).
  • Click injection is an Android scheme: a malicious app listens for install broadcasts and fires a click just before the install completes (AppsFlyer).
  • SDK spoofing fabricates install and event traffic that imitates a real SDK, without a real device or user (AppsFlyer).
  • Install farms use real devices and paid operators to install apps and complete cheap events, so the traffic passes identifier checks.

Click-to-install time (CTIT) is the first diagnostic. AppsFlyer describes two warning shapes. A cluster of installs in the first few seconds after the click suggests hijacked attribution. An almost flat distribution between hours 2 and 24 points to flooding. On Android, comparing the click time with the referrer’s install begin timestamp catches clicks that arrived after the download started.

Attribution helps detect fraud, but it is not a fraud system on its own. MMP fraud filters, post-install quality checks and source-level reviews do that work. For the operating side, see clean tracking and data hygiene and fraud detection in iGaming.

How UA Teams Use Attribution to Make Budget Decisions

Attribution becomes useful when installs are joined to value. UA teams compare sources by cohort ROAS on day 7 and day 28, by CPA for the event that predicts revenue and by payback period. Each of these depends on the attributed source being correct, so attribution accuracy caps the quality of every decision built on it.

Cohort ROAS is revenue from users acquired in a given period divided by what was spent to acquire them, measured at a fixed age such as D7 or D28. It is the main budget metric because it compares sources on the same clock. D7 gives a fast read; D28 shows whether early buyers keep paying. Which window to optimize toward depends on the product’s revenue curve, covered in D7 vs D28 ROAS.

CPA by event works where revenue arrives late. A fintech app may judge sources on cost per verified account; a betting app on cost per first deposit. The event must be one that the MMP records reliably and the network receives as a postback; otherwise optimization runs on a signal nobody trusts.

Payback period, the time a cohort needs to return its acquisition cost, turns those numbers into a scaling decision. Definitions for the full set of KPIs are in our mobile app metrics framework.

On iOS, the same decisions use two inputs: user-level cohorts from consented users and SKAN or AAK conversion values for the rest. The conversion value schema should encode the event that drives the budget decision in the first window, because later windows carry only coarse values. In programmatic UA, view-through windows and supply-path quality have a large effect on attributed numbers, so both belong in the pre-launch setup.

How ROCKAPP approaches measurement
Before launch, the tracker, attribution settings, view-through rules and iOS privacy configuration are agreed with the client and set up in the client’s MMP, typically AppsFlyer or Adjust. Network reports are reconciled against MMP data on a fixed schedule using the order shown in Figure 3, and optimization decisions use MMP cohorts as the single reference.

Conclusion: Make the MMP Record the One Everyone Works From

Attribution is only as useful as the agreement around it: one set of windows, one reconciliation routine and one dataset for budget and payouts. Teams that settle those rules before launch can spend the first weeks of a campaign on optimization.

To plan mobile user acquisition with transparent measurement, get started with ROCKAPP.

FAQ

What is mobile app attribution?

It is the process of linking an app install and the user's later actions to the ad click or impression that led to them. The output is a source (network, campaign, creative) for each user, which teams use to allocate budget, optimize campaigns and pay partners.

What is an MMP and do I need one?

A mobile measurement partner is an independent service that attributes installs across all ad networks using one set of rules. If traffic comes from more than one network, an MMP is needed to deduplicate claims and keep one record for budget and payouts.

What is a good attribution window?

There is no single window that fits every app. Many MMPs start at 7 days for clicks and 24 hours for views (AppsFlyer and Adjust defaults as of November 2026). Tune the window to your time-to-install data and keep it identical across networks you compare.

How does attribution work after iOS 14.5?

Since iOS 14.5, reading the IDFA requires the user's ATT permission. Consented users can still be matched to their exact click or impression. For users who decline, installs from ads that support SKAdNetwork or AdAttributionKit come back as delayed, aggregated postbacks, and the MMP shows them in a separate report.

What is the difference between SKAdNetwork and AdAttributionKit?

AdAttributionKit is the successor framework Apple introduced in iOS 17.4. It reuses SKAdNetwork concepts and extends them to alternative marketplaces and re-engagement; iOS 18.4 added configurable windows and country codes. Impressions from both frameworks compete in one ranking, and only one can win a conversion.

Why don't my ad network and MMP numbers match?

The two sides usually differ in attribution windows, date basis (click vs install), time zone, view-through rules and fraud filtering. Align those settings first, then compare. The section "Why Attribution Data Doesn't Match" above lists each cause and what to check.

Can attribution detect fraud?

Attribution data exposes anomalies such as abnormal click-to-install time, click injection on Android and installs with no later activity. It supports fraud detection but does not replace dedicated fraud filters and source-level quality reviews.