How Should Mobile Games Buy Rewarded Traffic in 2026?

Why did rewarded traffic get more attractive in 2026 without getting cheaper?

Because the alternatives got harder to read, not because the rate card moved. AppsFlyer's report, built on 9,600 gaming apps and 14.1 billion paid installs, found that games raised user acquisition budgets 3.8% in 2025 to $25 billion, and got 10% more paid installs for it while absorbing 20% more ad impressions.

Read those three numbers together. Impressions grew twice as fast as the installs they produced, on a budget that barely moved. That is a market paying more attention for the same result, which is what channel saturation looks like from the buyer's seat.

Games responded by widening the media plan. The same report found advertisers running more media sources in 2025: up 11% among mid-tier hypercasual advertisers and up 16% among large midcore advertisers on iOS, with iOS advertisers running roughly 15% more sources than their Android counterparts. Competition for the same inventory also intensified from outside: China-headquartered publishers took 35% of global gaming user acquisition spend outside China, a 22% jump in a year.

Casual games, which carry over half of all mobile game UA spend, cut Android budgets 7% and left iOS roughly flat. Budgets are not expanding. The list of places to spend them is.

Rewarded traffic is one of the places that list now reaches.

What can an iOS campaign still tell you about a player?

At full fidelity, only what happened in the first two days. Apple's AdAttributionKit documentation is explicit: the framework runs three conversion windows from first app launch, spanning days 0 to 2, days 3 to 7, and days 8 to 35. Your app can update a conversion value in all three. What comes back differs sharply.

The fine-grained conversion value appears only in the first postback. For the second and third windows, and for any download in a lower postback data tier, the system substitutes a coarse conversion value. Coarse means a bucket, not a number. Whatever you encoded about level progression, session depth, or first purchase gets flattened into it.

Timing compounds the loss. Postbacks are released after a random delay of 24 to 48 hours for the first and 24 to 144 hours for the second and third. A day-30 signal can reach your network six days after it fired, and the whole apparatus stops at day 35.

Then there is crowd anonymity. Apple assigns every app download a postback data tier based on the crowd size around the publishing app, the advertised app, the install country and the network's source identifier. At Tier 0 the system sends only the first postback. At Tier 1 the postback carries a two-digit source identifier and a coarse value in place of the conversion value. The identifier telling you which publisher app the player came from appears only at Tier 3. Small campaigns and narrow geographies sit at the bottom of that ladder, so the campaigns most in need of a clear read are the ones least likely to get one.

An offerwall buy is not routed through any of this. When a player completes your offer, your own server confirms the event to the network over a server-to-server postback, and the network bills against that confirmation. The event is whatever you wrote down. It carries the publisher app that sent the player, because the publisher supplied the click identifier. No crowd-size threshold applies, nothing gets bucketed, and there is no day-35 ceiling.

Why is the same argument weaker on Android?

Because Google stopped building the restrictions that make it compelling. On 17 October 2025 Anthony Chavez, VP of Privacy Sandbox, announced that Google was retiring ten Privacy Sandbox technologies across Chrome and Android, citing low levels of adoption.

The retired list includes the Attribution Reporting API, the SDK Runtime, Topics, Protected Audience and Protected App Signals. Google's own feature status page, last updated 14 August 2026, lists all six Android technologies as scheduled for phaseout.

The practical effect is that Android measurement in 2026 looks much as it did in 2021. A game running Android user acquisition can still observe a deterministic path from ad to install to any downstream event on any channel, at any depth, on its own timetable. The offerwall's measurement advantage over Meta or AppLovin is small there, because there is not much to be advantaged over.

So be honest about where the argument applies. On Android, judge rewarded traffic on cost per player who reaches your value event, the same test you would apply to any other source, and expect it to win or lose on that alone. On iOS, the measurement difference is real and it is worth paying something for.

GameAnalytics benchmark data covering 11,600 games and 1.48 billion average monthly active users puts top-quartile D1 retention at 31% to 33% on iOS against 25% to 27% on Android. The platform where players are most likely to come back is the platform where you can see the least about whether they did.

Which in-game milestone should you bill on?

Pick the deepest event that at least 5 in every 100 installed players reach. Below that frequency the network has too few conversions to optimize against and publishers cannot forecast what a placement will pay, so your offer loses position on the wall and the test never gathers the volume to mean anything.

Above that floor, every step deeper you push the billable event strips out another slice of players who came only for the reward. Finding where those two pressures meet in your own game is the job.

That range is narrower than most UA teams assume, and retention data is what pins it down. Across 11,600 games measured through 2024, GameAnalytics puts median D1 retention at roughly 17%, median D7 retention at 3.42% to 3.94%, and median D28 retention at 0.9% to 1.1%, with 75% of games below 3% at D28.

Retention figures are GameAnalytics medians and top-quartile values for 11,600 games across iOS and Android, measured 1 January to 31 December 2024. They describe the industry distribution, not your game. Run the same four rows against your own cohort data before using them to set a bid.

Read the table as a constraint rather than a menu. For a median game, a day-7 billable event fires for fewer than 4 players in 100, which is under the floor. For a top-quartile game it fires for 7 or 8, which works. Two studios can therefore make opposite and equally correct decisions about the same event, and neither can copy the other's setup.

Progression milestones are usually the better answer than return days, because you control where they sit. Tutorial completion, reaching a named level, or a first session past a set duration can be tuned until the completion rate lands in the workable band, and they are instrumented already in most games. Map your candidate milestones onto your own D1 and D7 curve, find the ones that clear 5%, and take the deepest.

This is where outcome-based buying stops being a reporting preference. On an offerwall the advertiser is charged for a completed action, so the milestone you name is the thing you are actually purchasing, and moving it rewrites the price as well as the report. RevU has operated that model across 20 years of adtech history, and for a buyer the practical consequence is that milestone changes have to be negotiated rather than toggled.

What do the quality numbers actually say about rewarded players?

They say rewarded players outperform other rewarded players, which is a narrower claim than the industry usually reports. A 2024 report found that users acquired through offerwalls showed 45.8% higher D1 retention, 86.1% higher D7 retention and 71.7% higher D14 retention.

The comparison population is users acquired through rewarded video and interstitials, not organic installs or search traffic.

That distinction carries the number. Read carelessly it says offerwall players are better than everyone. Read correctly it says that among players who arrived through an incentive, the ones who chose your offer from a list and worked through a defined task retain considerably better than the ones who watched a video. Selection and effort are doing the work.

It is 2024 data so treat the magnitudes as directional. The direction is well supported by the mechanism: a player who spent ten minutes completing your tutorial to earn currency has already done the thing that a video viewer never did.

The same report found the highest in-game engagement rates attach to gacha rewards at 31.1%, additional moves at 30.5% and daily rewards at 30.3%. Those are publisher-side reward types, but they point at something a buyer can use: the offers that work are the ones whose reward slots into a loop the player was already in.

When should a game not buy rewarded traffic?

When you have nothing to bill on. If your game has no instrumented event between install and day 7 that fires for at least 5 in 100 players, you cannot construct a workable offer, and buying on install alone will hand you a cohort that collects and leaves. Build the milestone first. That is a week of instrumentation, not a campaign.

When your D1 retention sits below the industry median of roughly 17%, the problem is not your traffic mix. Paid players will not retain better than your organic ones, and rewarded players entering a game with a weak first session churn at the point the reward stops. Buying traffic into that is buying a more expensive version of a result you already have.

When you need a specific audience rather than a specific action. Rewarded inventory selects for willingness to complete a task in exchange for currency. If your game depends on reaching people with a demonstrated interest in your genre, a store search or a lookalike audience is a better instrument, and no billable event will substitute for intent you did not buy.

And when you cannot run the campaign long enough to read it. With 75% of games below 3% D28 retention, a two-week test that stops before your cohort matures tells you what you paid, not what you bought.

How to run a first buy, by stage

Before you spend anything. Write the billable event down as a sentence, including what does not count, and get the network to confirm it in writing. Check your own cohort data for what share of installs reach it. If that share is under 5%, pick a shallower event.

First $5,000. One geography, one platform, one offer. Bill on the milestone you chose, not on install. Do not judge anything yet. You are confirming the postback fires, the reward credits, and players are not stalling in the offer.

$5,000 to $25,000. Get publisher-level reporting and use it, because this is the one channel where you can have it unconditionally. Cut the source apps whose players stall after the reward and raise your bid on the ones whose players keep going. Model this cohort's revenue curve on its own; folded into a blended LTV assumption it will drag your average and tell you nothing about either group.

Past $25,000. Run a geo holdout. You now have a strange asymmetry to manage: you can see this cohort in full and your iOS campaigns elsewhere only in coarse buckets after day 2, so a straight side-by-side flatters whichever source you happen to measure better. A holdout sidesteps the comparison entirely by asking what your revenue does with the channel switched off.

Ongoing. Move the milestone deeper one step and watch what happens to volume and to day-30 value. The billable event is the only lever on this channel that changes both price and player quality at once, which is why it deserves more of your attention than the bid does.