How Retargeting Ads Work for Mobile Games: Android Audiences, Re-engagement Attribution Windows, Segment and Deep Link Design, and Proving Incremental Lift With Holdouts in 2026

Paid app retargeting for mobile games in 2026: GAID audiences, MMP sync, re-engagement windows, segment design, deep links, exclusions, and holdout tests.
Nov 18, 2026
How Retargeting Ads Work for Mobile Games: Android Audiences, Re-engagement Attribution Windows, Segment and Deep Link Design, and Proving Incremental Lift With Holdouts in 2026

Retargeting ads are paid-media campaigns aimed at people who already have your game installed, or had it installed at some point. They serve one of two goals. Re-engagement brings users who still have the app back into the game. Re-attribution, also called a retargeting reinstall, gets users who deleted the game to install it again. Depending on the ad network and the MMP, the same idea goes by remarketing, app retargeting, or re-engagement campaigns.

Its share of the market keeps growing. AppsFlyer's 2025 trends report, built on 32 billion paid installs, 45,000 apps, and 45 billion remarketing conversions, puts global app remarketing spend at $31.3 billion in 2025, up 37% year over year. Remarketing's share of total app marketing spend (UA plus remarketing) rose from 25% in 2024 to 29% in 2025. Most of that growth came from iOS ($17 billion, up 71%), while Android reached $14 billion, up 10%. Android remarketing spend in the gaming category grew 5%, which the report classifies as flat.

It isn't hard to see why games lag here. Games already have plenty of owned channels that reach existing players, such as push notifications, in-game events, and community spaces. Paid retargeting means spending money on the players those channels can't reach: people who turned off notifications or deleted the app. So for games, the question is less "should we retarget?" and more "whom, under what conditions, and without paying for players who would have come back anyway?"

This article covers how paid-media retargeting works as of November 2026. Diagnosing why players left and designing re-engagement messaging across push, email, and ads was covered in Winning Back Lapsed Mobile Game Players: A Re-engagement Framework That Works. This article focuses on the part run through ad networks: how it is set up and how it is measured.

How is retargeting different from UA?

Retargeting differs from UA because the audience is fixed in advance as "people who already know my game," and that one difference changes how audiences are built, how attribution works, and how performance should be judged.

UA uses the ad network's data to find people who don't yet know your game and are likely to install it. Retargeting works the other way around. The advertiser hands its own data to the network (who installed, when they last played, how much they spent), and the network shows ads to users on that list. Targeting control sits with the advertiser's first-party data.

UA campaign

Retargeting campaign

Audience

Users who have never installed the game

Users with an install history (app still installed or deleted)

Targeting data

Network data, lookalikes, interests

Advertiser's user data (install, session, and purchase history)

Conversion

Install

Re-engagement (app open) or re-attribution (reinstall)

Landing

Store page

A specific in-game screen via deep link (reinstalls go through the store)

KPIs

CPI, D7 retention, ROAS

Cost per re-engagement, post-return retention, repurchase rate, incremental ROAS

Core risk

Low-quality and fraudulent traffic

Cannibalization: paying for players who would have returned anyway

That last row runs through this entire article. In UA, paid channels claiming organic installs is a problem with some channels. In retargeting, it is structural. Everyone in a retargeting audience already knows the game, and many of them will come back eventually without any ad. Because attribution counts every user who returned after seeing an ad as a retargeting result, measured performance almost always overstates the real contribution.

How does app retargeting work on Android?

Android retargeting works like this: the advertiser builds a list of advertising IDs (GAIDs) for users who meet certain conditions, passes it to ad networks through an MMP or a direct upload, and the networks find those devices on their own inventory and show them ads.

The flow breaks down into four steps. First, define a segment from game data (BI or MMP events) using behavioral conditions, such as "made a purchase in the last 30 days and hasn't opened the game in 7+ days." Second, pass those users' identifiers to the networks. An MMP's audience feature lets you define a segment once and send it to several networks at the same time. According to AppsFlyer's Audiences documentation, audiences are refreshed automatically once a day and pushed to connected partners again, and you can control which identifiers are shared with each partner. Third, the network serves ads to those devices, and a click opens a specific screen in the game via deep link. Fourth, the MMP records that app open as a re-engagement and credits in-app events for a set period afterward to the retargeting network.

The details vary by network type.

Network type

How the audience is delivered

How attribution is reported

What the studio should check

Self-attributing networks (SRNs): Google, Meta, TikTok, etc.

MMP audience sync or direct upload to the ad account

The network claims credit via API, the MMP decides

Differences between the network's own lookback and inactivity rules and your MMP settings

Retargeting-specialist DSPs

MMP audience sync

Retargeting attribution links (click and impression)

Whether impression data is shared, bidding unit, frequency controls

Ad networks

MMP audience sync

Retargeting attribution links

Whether retargeting is supported, sub-publisher transparency

Owned media (cross-promotion, etc.)

Internal lists

MMP link management

Separating results from natural returns

Google's retargeting product, App campaigns for engagement (ACe), has its own entry requirements. Per Google Ads Help, you need at least 50,000 installs, deep links, audience lists, and conversion tracking through Firebase or an app attribution partner. AppsFlyer's documentation adds two points. Because of a domain mismatch, ACe can't use third-party deep linking solutions (including AppsFlyer's OneLink), so you have to use your own App Links or custom URI schemes. And to record re-engagement conversions, you need to import the session_start event into Google Ads. Skip that, and Google Ads and MMP numbers diverge sharply, and Google's targeting optimization stops working. Requirements for each App campaign type are summarized in Should Google App Campaigns Be Your Game's Main UA Channel? 2026 Campaign Types, Bidding Thresholds, Budget Floors, What You Can't Control, and a Pilot Plan.

Relying on GAIDs also comes with limits. According to Play Console Help, once a user deletes their advertising ID in Android settings, any later request for the identifier returns a string of zeros, and apps targeting Android 13 (API 33) or higher must declare the AD_ID permission in their manifest. Users who delete their ad ID drop out of device-ID-based audiences, so the reachable audience is smaller than the number of users who match the segment definition. The Android Privacy Sandbox, which was supposed to replace the GAID, went away when Google announced in October 2025 that it was retiring its core technologies, so Android retargeting is likely to stay GAID-based for now. What that change means is covered in Privacy Sandbox Is Gone. What That Actually Means for Android Game Marketing.

How is re-engagement attribution different from install attribution?

Re-engagement attribution asks "who got the user to open the app again?" rather than "who drove the install?", so it adds several windows that install attribution doesn't have, and it introduces dual attribution, where the same in-app event is credited to both the UA source and the retargeting source.

Here are the main settings, based on AppsFlyer's retargeting attribution guide. Names and defaults differ between MMPs, so check the same items in your own MMP's documentation.

Setting

Default

Range

What it means

Click lookback

7 days

1 hour to 30 days

Maximum time from a retargeting ad click to the app open

View-through lookback

1 day

1 hour to 7 days

Maximum time from an ad impression to the app open

Re-engagement window

30 days

1 hour to 90 days, or lifetime

How long in-app events after a re-engagement are credited to the retargeting source

Minimum inactivity period

None

0 to 30 days

Users active within this period are not recorded as re-engagements or re-attributions

Minimum time between re-engagements

None

1 hour to 30 days

Prevents another retargeting conversion from overwriting an existing re-engagement window

Re-attribution window

90 days

1 to 24 months

Reinstalls within this period after the first install are not new installs; only reinstalls via retargeting are credited to the retargeting source

Three differences from install attribution matter most.

First, dual attribution. Once a re-engagement is recorded, in-app events during the re-engagement window are credited to both the retargeting source (primary) and the original UA source (non-primary). Simply adding revenue from the UA dashboard and the retargeting dashboard counts the same purchase twice. When you report combined channel ROAS, state which view it is based on.

Second, windows get overwritten. By default, if a user responds to another retargeting campaign during an active re-engagement window, the existing window closes immediately and a new re-engagement is recorded. Running several retargeting networks at once turns the same user into a credit hand-off between networks, and whichever network showed the latest ad takes the result. Setting a minimum time between re-engagements prevents this.

Third, the active-user filter is off by default. AppsFlyer's documentation notes that the minimum inactivity period is not enabled for media sources by default, and says directly that many retargeting campaigns underperform because they don't filter out active users. With a 3-day inactivity period, users who opened the app in the last 3 days are not attributed to the retargeting source; only users inactive for 4 days or more are. Without this setting, a player who played yesterday, sees an ad once today, and logs in as usual gets counted as a re-engagement.

SRNs apply their own inactivity and lookback rules, so if they differ from your MMP settings, network dashboards and MMP numbers will disagree. Setting the same values in the ad account and in the MMP is the simplest way to reduce discrepancies.

Which player segments should you retarget?

Retargeting segments should be chosen on two axes at once, how valuable the player is and how likely they are to come back without an ad, because without the second axis the budget drifts toward the players who are easiest to bring back.

If you build segments on value alone, lapsed payers always come first. They have a spending history, so expected revenue is high. But payers are also more attached to the game, and a larger share of them return on their own after an event announcement or an update. This is the segment where measured ROAS looks best and incremental lift may be smaller than expected. D7 non-returners are the opposite: individually less valuable, but likely to have a relatively low natural return rate, so a larger share of the returns you drive may be incremental.

Segment

Example definition

Example inactivity threshold

Deep link destination

Primary KPI

Watch out for

Lapsed payers

Purchase in last 60 days, inactive 7+ days

7 days

Comeback reward inbox, new season or shop

Repurchase rate, post-return ARPPU

Highest natural return rate, highest cannibalization risk. Holdout is mandatory

D7 non-returners

Played D1 to D3 after install, then inactive 4+ days

4 days

The next goal where they stopped (next stage, unclaimed reward)

D14 and D30 return rate, next-step conversion

Low individual value, so bid caps need managing

High-intent non-payers

Top playtime, visited the shop, no purchase, inactive 3+ days

3 days

First-purchase offer or the relevant item screen

First-purchase conversion

Including active users overlaps with in-game offers. Keep the inactivity condition

Uninstallers

Uninstall detected, reached a certain level before uninstalling

Within re-attribution window

Through the store, then a deferred deep link to where they left off

Reinstall rate, post-reinstall D7

Whether progress can be restored (account linking) drives results

Long-term dormant

Inactive 30+ days, previously reached a certain level

30 days

Major update or new content

Post-return D7 retention

The message must match the update. Audience size shrinks

Exclusion lists matter as much as segments. Recent players, recent payers, users already inside another retargeting campaign's re-engagement window, suspected fraudulent devices, and users who withdrew ad consent should be excluded from every segment by default. You also need a priority order so segments don't overlap and each user sits in only one campaign. That keeps per-campaign results readable and reduces cases where two networks bid against each other for the same user.

A retargeting ad's promise has to be kept on the in-game screen that opens after one tap, and if that screen doesn't match the message, or the ads keep running after the player has already returned, cost goes up and the experience gets worse.

Choose deep link destinations to fit the segment and the ad message. An ad that promises "claim your comeback reward" should open on the reward screen, not the home screen. An ad announcing "a new season starts" should open on the season pass or event page. Uninstallers don't have the app, so they go through the store, and sending them to the original destination on first launch after reinstalling requires a deferred deep link. How it is implemented, where it breaks, and how to QA it is covered in Deep Links and Deferred Deep Links for Mobile Games in 2026: How They Work on Android and iOS, What Replaced Firebase Dynamic Links, Where Games Use Them, and a QA Checklist. Before launch, it is worth checking how each segment's destination opens across login states, when an update is required, and after an event has ended. If an ad keeps sending players to an event screen after the event is over, their first moment back in the game is an empty screen.

Frequency and exclusion settings are easiest to manage as three layers.

The first layer is the audience refresh. A player who returns no longer meets the segment condition (inactivity) and drops out at the next refresh. But if the MMP audience refreshes once a day, ads can keep running to that player for up to a day after they return. The second layer is the MMP's minimum inactivity period. It stops active users who haven't yet dropped out of the audience from being recorded as re-engagements if they see an ad and log in. It doesn't stop the spend itself, but it reduces inflated results and wrongful billing under click- or conversion-based pricing. The third layer is the network's frequency cap. Retargeting audiences are much smaller than UA targets, so the same user easily sees the same creative repeatedly in a short time. Where the network supports it, set per-user daily or weekly impression caps, and run separate creatives for each segment.

Setting

Where

What it prevents

Practical example

Segment condition (inactive N days)

BI or MMP audiences

Targeting active users

3 to 30 days by segment

Audience refresh frequency

MMP-network sync

Continued delivery to returned players

The shortest available cycle; if once a day, back it up with an inactivity period

Minimum inactivity period

MMP partner integration settings

Re-engagements recorded for active users

At least the same value as the segment condition

Minimum time between re-engagements

MMP app settings

Networks overwriting each other's credit

Similar length to the re-engagement window

Re-engagement window

MMP link or network settings

Overly long credit that absorbs results from other activity

7 to 30 days, matched to when the segment's goal event happens

Frequency cap

Network

Creative fatigue, wasted impressions

Per-user daily or weekly cap within the network's recommended range

Shared exclusion list

MMP and networks

Recent payers, suspected fraud, consent withdrawn, overlap with other campaigns

Applied identically to every campaign

Choosing the re-engagement window length is also a measurement decision. A longer window credits more revenue to the retargeting source, but some of that revenue comes from purchases driven by post-return LiveOps events or push notifications. If the segment goal is "repurchase within 7 days," keeping the window close to that keeps the read clean.

What changes for retargeting on iOS?

On iOS, using the advertising identifier (IDFA) requires ATT consent in both the app and the ad network's app, so far fewer users can be reached through device-ID-based audiences than on Android.

Users who don't consent can't be passed to networks as part of a device-ID list, which leaves matching on other identifiers such as email where a network supports it, or relying on app-activity signals inside the network. On the measurement side, Apple's AdAttributionKit supports re-engagement conversions alongside installs on iOS 17.4 and later, but its postbacks are aggregated rather than user-level, which makes fine-grained segment analysis difficult. The 71% growth in iOS remarketing spend in AppsFlyer's 2025 data was led by non-gaming categories such as transportation, travel, finance, and shopping, and shouldn't be read as a sign that game retargeting got easier on iOS. For games with a large Android share, the practical path is to build the retargeting operation and its experiments on Android first, then carry the same segment logic over to iOS within what each network allows.

Why can't you trust retargeting results without a holdout?

Some of the players in any retargeting audience will come back on their own without an ad, so attributed re-engagements and revenue always include returns that would have happened anyway, and the only way to know how many is to compare against a randomized control group, a holdout.

AppsFlyer's documentation on incrementality for remarketing starts from the same point. Even if a campaign's conversion rate is 5%, you only know its real value after subtracting the users who would have converted without it. The method is simple. Randomly split the retargeting segment, show ads to one side (the test group) and not the other (the control group), then compare return and purchase rates between the two. The control group must not be sent to any network, so build it with the MMP audience split feature, or split the audience in BI and upload only the test group.

Here is a worked example. The numbers below are assumptions for illustration.

Test group (90%)

Control group (10%)

Users in segment

180,000

20,000

Return rate within 14 days

6.0%

4.5%

Users who returned within 14 days

10,800

900

Estimated natural returns at the control rate

8,100

-

Incremental returning users

2,700

-

If the MMP attributed 6,000 re-engagements to this campaign over the same period and spend was $30,000, the dashboard shows $5 per re-engagement. But the ads actually caused 2,700 returns, so the cost per incremental return is about $11.10, more than double. It also means 75% of the test group's returns (8,100 players) would have come back without the ads. The real test is whether that incremental cost is lower than the post-return value you expect from the segment.

Incrementality methods also differ. According to AppsFlyer's documentation, SRNs are measured with the intent-to-treat (ITT) method, while attribution-link networks that report impression data are measured with a reach-based method that compares only users who actually saw an ad. ITT assumes everyone in the test group was reached, so lift tends to come out low. Reach-based results look amplified, whether positive or negative. That is why incrementality rates measured with different methods shouldn't be compared directly across networks.

It is normal for holdout results to differ by segment. As noted above, lapsed payers can show high measured ROAS and low incremental lift, and D7 non-returners can show the reverse. So treat holdouts not as a one-time check but as a tool for tuning segments and inactivity thresholds. Use the same method to see whether incremental lift rises when you extend the inactivity threshold from 3 days to 7, or whether incremental returns hold when you lower bids on the payer segment. Test duration, sample size, and how to read results are covered in detail in How to Design an Incrementality Test for Mobile Game UA: Holdouts, Geo-Lift, Sample Size, and Reading the Result, and the same principles apply to retargeting.

Seen this way, the criteria for choosing retargeting channels change too. Whether you can run a holdout, and whether you can define success by what players do after they return, matter more than reach into your existing user base or the number of measured re-engagements. Playio is an Android-based community of 5 million gamers who love games, and beyond acquiring new players it also runs in-game action-based rewards aimed at existing players who already have a game installed. For example, success can be defined as actual play after the return, such as reaching a specific stage or hitting a playtime goal, rather than a simple app open. CPI is the default pricing model, but CPE is also supported, so you can define re-engagement success as "played again" rather than "opened the app" and compare it with other retargeting channels under the same holdout standard.

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

Key Takeaways

Retargeting ads show paid-media ads to users who have installed your game before. On Android, the GAID list for an advertiser-defined segment reaches SRNs, DSPs, and ad networks through MMP audience sync or a direct upload. As of November 2026, AppsFlyer data shows global remarketing spend of $31.3 billion in 2025 (up 37%) and a 29% share of total app marketing spend, while Android remarketing spend in gaming grew only 5%, which the report calls flat. Unlike install attribution, re-engagement attribution credits in-app events to both the UA source and the retargeting source during the re-engagement window (30 days by default in AppsFlyer), lets another retargeting conversion overwrite an existing window, and leaves the minimum inactivity period that filters out active users switched off by default. Choose segments on both value and natural return likelihood, match deep link destinations to the ad message, and layer audience refreshes, inactivity periods, frequency caps, and a shared exclusion list so you don't pay for players who have already returned. Above all, measured retargeting results include players who would have come back anyway, so don't scale budget before a segment-level holdout tells you the cost per incremental return.

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

Share article