Why Every Dashboard Shows a Different Install Count: A Game Marketer's Guide to Mobile Attribution in 2026

When the ad platform reports 120 installs, the MMP reports 80, and the backend counts 65 for the same campaign, nothing is misconfigured. That is how attribution works by design. Global gaming UA spend reached $25 billion in 2025 against 14.1 billion paid installs, and the MMP is the official ledger that decides which channel earned that money. Fewer marketers than you would expect know how to read the ledger correctly.

This post is the companion to Media Mix Modeling for Mobile Games: Measuring What Attribution Can No Longer See. That piece covered what attribution cannot see; this one covers what it still sees and how to avoid misreading it. The shift in the Android measurement environment itself is covered in Privacy Sandbox Is Gone. What That Actually Means for Android Game Marketing.

What Does an MMP Actually Do?

An MMP ties an ad touchpoint and an install (or in-app event) to the same device and assigns credit to a single source. Ad networks send clicks and impressions, the SDK inside the app sends the first open, and the MMP matches the two records.

There are two matching methods. Deterministic matching links records through an identifier that appears on both sides, such as the advertising ID (GAID on Android) or the Play Store install referrer. Probabilistic matching infers that two records are "probably the same device" from indirect signals: IP address, device model, OS version, and the time between click and install. Because accuracy differs, MMPs prioritize deterministic matches and allow probabilistic matches only a much shorter window.

The attribution window is the rule for how many days after a touchpoint an install still counts as that touchpoint's credit. As of September 2026, the defaults across major systems look like this, and the gaps between them are the most common source of the discrepancies discussed below.

The assignment model remains last-touch by default: the last eligible touchpoint inside the window takes all the credit. Multi-touch splits credit across touchpoints, but on mobile, where an install happens once and cross-platform identification is limited, it mostly serves as a reference report rather than the system of record.

How Does Android Attribution Work in 2026?

Android did not become the identifier-free world that was announced two years ago. Google deprecated Privacy Sandbox on Android as of October 17, 2025, retiring ten APIs including the Attribution Reporting API. What disappeared was the replacement technology, not the existing method.

The 2026 Android attribution stack therefore has three layers. First, GAID-based deterministic matching works as before for users who have not disabled their advertising ID. Second, the Google Play Install Referrer passes the referrer string, click timestamp, and install-begin timestamp into the app, linking click to install independently of the advertising ID. This is where click-to-install time (CTIT) comes from, and abnormally short or long intervals are the signal used to filter click injection and click spam. Third, Google's own ads apply modeled conversions across all App campaigns for users who opted out of the advertising ID. By Google's definition, modeling uses data that does not identify individuals to estimate conversions it cannot observe directly, which makes those numbers a different kind of count from the MMP's deterministic tally.

The practical implication is that on Android, deterministic MMP attribution can still serve as the basis for campaign-level decisions, which is why correcting the misreads below pays off far more here than on iOS.

Why Only a Brief Note on iOS?

iOS is structurally different, so Android methodology does not transfer. As of Q1 2026, global ATT opt-in stands at 38%, and gaming leads all categories at 39%, but six users in ten remain outside device-level attribution and are measured only through Apple's aggregated postbacks.

One fact worth settling: SKAN 5 never shipped as a separate version. The re-engagement support it promised was moved into AdAttributionKit, a new framework built on top of SKAdNetwork. UA approaches for the post-IDFA environment are covered in Mobile Game User Acquisition in the Post-IDFA Era: What Has Changed and What Works Now. The rest of this post assumes Android.

Why Do Three Dashboards Show Three Install Counts? Four Common Misreads

Three dashboards producing three numbers is normal; the problem is which number you use for budget decisions.

First, double counting by self-attributing networks. Meta, Google Ads, Apple Search Ads, TikTok, Snapchat, and X perform attribution inside their own platforms and pass the result to the MMP as a claim. If a user saw ads on two platforms, both report the install as theirs, but the MMP credits only one. The sum of platform dashboards exceeding the MMP's total is expected, and MMP figures running 10-20% below platform figures is the normal range. A wider gap points to a configuration issue.

Second, organic cannibalization. When an ad reaches a user who would have installed anyway, last-touch records the install as paid. Attribution answers what the last touchpoint was; it cannot answer whether the install would have happened without it. Only incrementality answers that question.

Third, window mismatch. Leave the defaults in the table above untouched and Google Ads sees the same user on a 30-day click window while the MMP sees 7 days. A user who installs on day eight exists in the platform and not in the MMP. Align the windows first, then interpret only the difference that remains.

Fourth, view-through inflation. Crediting an impression alone is weaker evidence than a click, which is why MMPs keep the default window at 24 hours, but platform reports include their own view-through counts. The higher a channel's impression volume, the larger this gap.

All four misreads flow directly into the denominator of CPI. How the same campaign produces different CPIs depending on which dashboard's install count you use is covered in How to Calculate eCPI: Common Pitfalls and Optimization in UA.

Attribution, Incrementality, and MMM Are a Stack, Not Substitutes

The three tools answer different questions, so none can replace another.

The working sequence is to run day-to-day optimization on attribution, correct the channels attribution overvalues with quarterly incrementality experiments, and feed those results into MMM as priors. Incrementality design is covered in Rewarded Ads Are Working — But Can You Prove It? The Case for Incrementality Measurement, and the owned-data foundation under all three in A Working Guide to First-Party Data: Turning Player Signals Into a Marketing Asset.

Making "Did This User Actually Play" the Conversion: The Playio Perspective

The question an install pixel cannot answer is whether the user actually played the game. AppsFlyer's analysis of 106.4 billion installs from Q1 2025 to Q1 2026 found that 52% of fraudulent installs came through organic traffic, which happens because an install proves existence, not intent.

Playio defines the conversion not as an install but as reaching a playtime threshold or completing a specific in-game action, verifies that behavior against the platform's own first-party behavioral data, and matches it in the advertiser's MMP. On top of the question of who the last touchpoint was, this adds a second check: did the user that touchpoint delivered actually reach the core loop. The ambiguity that install-time fraud and organic cannibalization create shrinks by that margin. Campaigns run on CPI or CPE pricing.

You can find more details here. (https://playioadsen.oopy.io/bizdeck)

Key Takeaways

As of September 2026, mobile attribution on Android still works deterministically. With Privacy Sandbox shut down in October 2025, GAID and the install referrer remain the backbone, and only Google's own ads rely on modeled conversions. Three dashboards showing three numbers is normal, and nearly all of it is explained by four misreads: self-attributing network double counting, organic cannibalization, window mismatch, and view-through inflation. In an environment where half of fraudulent installs arrive as organic, moving the conversion definition from the install to post-install behavior is the most practical way to make attribution more trustworthy, and attribution only becomes a budget-grade number once it is corrected by incrementality and MMM.

For inquiries about Playio's advertising solutions, reach out at: [email protected]

Sources

• AppsFlyer, State of Gaming for Marketers 2026 ($25B gaming UA spend, 14.1B paid installs): https://www.appsflyer.com/company/newsroom/pr/gaming-marketing/

• Google for Developers, Privacy Sandbox on Android ("As of October 17, 2025, Privacy Sandbox on Android is deprecated"): https://developers.google.com/admob/android/privacy/sandbox

• AdExchanger, Google Pulls The Plug On Topics, PAAPI And Other Major Privacy Sandbox APIs (ten APIs retired including Attribution Reporting): https://www.adexchanger.com/privacy/google-pulls-the-plug-on-topics-paapi-and-other-major-privacy-sandbox-apis-as-the-cma-says-cheerio/

• Google Ads Help, About modeled conversions (definition, expansion to all App campaigns): https://support.google.com/google-ads/answer/10081327

• Google Ads Help, About conversion windows (30-day click, 1-day view, 3-day engaged view defaults): https://support.google.com/google-ads/answer/3123169

• Adjust Help Center, Attribution windows (7-day click and 24-hour impression defaults, probabilistic 1-24 hours): https://help.adjust.com/en/article/attribution-windows

• Singular, What is a Play Install Referrer (referrer string, click and install-begin timestamps, fraud detection): https://www.singular.net/glossary/install-referrer/

• Tenjin, Self-Attributing Networks (SAN list, MMP deduplication role): https://tenjin.com/glossary/self-attributing-networks-sans/

• Linkrunner, What is a Self-Attributing Network (10-20% platform-vs-MMP discrepancy as normal range): https://linkrunner.io/glossary/what-is-a-self-attributing-network

• Linkrunner, Mobile Attribution After ATT and GDPR 2026 (Meta default 7-day click / 1-day view): https://linkrunner.io/blog/understanding-mobile-attribution-in-today-s-privacy-landscape

• Adjust, Mobile App Trends 2026 via PocketGamer.biz (ATT opt-in 38% global, 39% gaming, Q1 2026): https://www.pocketgamer.biz/report-global-app-installs-climbed-10-in-2025-as-sessions-rose-7-year-over-year/

• Singular, AdAttributionKit: the new SKAdNetwork? (SKAN 5 never shipped, AAK re-engagement): https://www.singular.net/blog/adattributionkit-the-new-skadnetwork/

• AppsFlyer, State of Fraud for Marketers 2026 (52% of fraudulent installs organic, 106.4B installs analyzed): https://www.appsflyer.com/company/newsroom/pr/organic-traffic-ad-fraud/