App install campaigns on Meta and Google: CPI and what comes after

Google app campaigns, named App campaigns in Google Ads, take a few lines of text, a bid, a budget, the languages and locations, and your assets, then build ads from them and from your store listing and run them across Google Search, Google Play, YouTube, Discover and the Display Network, testing combinations and serving the ones that perform. You can add up to 20 images, 20 videos and 20 HTML5 assets. There are three subtypes: App campaigns for installs, App campaigns for engagement, which need at least 50,000 installs and deep links and re-engage people who already have the app, and App campaigns for pre-registration, Android only, for titles that will launch on Google Play within 90 days.

Bidding is the whole setup. Target cost per install is what you are willing to pay for a new user; target cost per action pays for a user likely to complete the in-app event you choose; target return on ad spend asks for a dollar value back per dollar spent; Maximize conversions runs without a target. Google's own examples give the scale: a $100 daily budget at a $2 target CPI is built for about 50 installs a day, and $300 at a $10 target CPA for about 30 actions. Its best practices make that a rule: a daily budget of at least 50 times the bid for target CPI campaigns, at least 10 times for target CPA, and no changes before the first 100 conversions have registered, with one-time changes rather than cycling bids up and down.

On Meta, app installs run under the App promotion objective, and app promotion campaigns are Advantage+ app campaigns. Meta describes them as built to deliver high-performing creative variations to more relevant audiences on more effective placements, and to sustain performance as the budget increases. The trade is targeting: the only options are operating system, countries and language. No exclusions, no interests or behaviors, no device, no gender, no lookalike or custom audiences at creation, and the minimum age defaults to 18, or 21 in some countries. A 90-day exclusion window is applied automatically, so anyone who installed or opened the app in the last 90 days is not shown its ads.

The performance goal does the work targeting used to do. Meta's own chart maps app installs optimization to cost per install, app events optimization to cost per event, and value optimization to return on ad spend, and notes that the only bid type is lowest cost, with a bid cap possible but likely to limit results. You can run as many Advantage+ app campaigns as you need for different events, regions or creative tests, and Ads Manager stops you from duplicating a campaign with the same setup. In Adrails, a Meta campaign with the App promotion objective is built in the editor from a brief or by the Campaign Manager, checked with Meta before it is sent, and created paused, so nothing spends until you switch it on.

The lead table gives the published figures. In Liftoff's 2025 Casual Gaming Apps Report, covering February 2024 to February 2025, casual games cost $1.41 per install on iOS and $0.14 on Android, casino games $21.03 on iOS, and simulation, action and racing games about ten times more on iOS than on Android. Liftoff's 2024 Mobile Ad Creative Index, on 2023 data, has native gaming ads at $3.09 on iOS and $1.22 on Android, finance apps under $1 on Android across formats and over $5 on iOS for three of four formats, and entertainment video and interstitial ads at $5.31 on iOS against about $2 on Android. RevenueCat's 2025 report notes that in North America iOS CPIs reach nearly three times the Play Store's in some categories.

Three caveats. First, every one of these numbers is one network's inventory in one period: Liftoff is programmatic in-app, and Meta and Google auctions clear at their own prices. Second, the platform gap is not a bargain: RevenueCat's 2025 report finds that over 67% of apps in every region earn at least 80% of their revenue from iOS, and Liftoff's casual games returned a day-30 ROAS of 47% on iOS against 15% on Android despite the tenfold difference in CPI. Third, spend keeps rising: AppsFlyer projected global app install ad spend at $94.9 billion in 2025, 20% above 2023, and in April 2025 reported iOS ad spend up 26% from 2023 to 2024. A benchmark from last year is a floor for this year's auctions, not a target.

• One network, one period: compare with your own last quarter first
• Cheap installs on Android against iOS revenue: read ROAS, not CPI
• Format matters as much as platform: native and banner cheapest, video dearest
• Rising spend: last year's benchmark is this year's floor

An install is the first step of a funnel the campaign has to pay for in full. RevenueCat's 2025 report, on 75,000 subscription apps and 2024 data, gives the shape: a median trial start rate of 6.2% of installs, 20.3% for the top decile and 9.8% for high-priced apps; 82% of trials start the day of the install; trials of 17 to 32 days convert at a 45.7% median; and nearly 30% of annual subscriptions are canceled in the first month. Run the formulas above with those figures and a $3.00 install becomes a $48 trial and a $121 subscriber. The campaign that reports the lowest CPI is often the one buying installs that never open the app twice.

So the target is a cost per paying user with a renewal rate behind it, and the allowed CPI is derived from it: target per subscriber times trial rate times trial-to-paid rate. Both platforms can optimize to the later event once it is sent back, Google through target CPA on the trial or purchase event and target ROAS on value, Meta through app events and value optimization, and both need enough events to learn, which is why Google asks that a campaign not be changed before its first 100 conversions. With RevenueCat connected, Adrails reads your app's revenue, MRR, customers and subscriptions, subscriptions every 30 minutes and revenue metrics every 6 hours, and sets them next to ad spend so app campaigns are judged on the subscriptions they brought; a campaign's Tracking tab lists the customers each ad brought and their monthly revenue.

Since iOS 14.5, Apple requires an app to ask permission through App Tracking Transparency before it tracks the user or reads the device's advertising identifier, and without permission the identifier is all zeros. Apple defines tracking as linking data from your app with data from other companies' apps or websites for targeted advertising or measurement, which is exactly what an install campaign's attribution used to do. AppsFlyer reported in April 2025 that about 50% of users shown the prompt now consent, a 10% increase since the first rollout, with France just under 51% and Germany at 47%, so roughly half of iOS installs are still measured without a user-level identifier.

For those, Apple's SKAdNetwork validates ad-driven installs with a postback that carries no user or device data, and lets the app update a conversion value as the person engages, with up to three postbacks over three conversion windows since iOS 16.1. Apple now points advertisers to AdAttributionKit, which builds on SKAdNetwork, attributes clicks within 30 days and views within 24 hours, and delivers a postback 24 to 48 hours after the app is launched. The constraints are the point: 64 possible conversion values, aggregated reporting, and values withheld when a campaign is too small to meet Apple's privacy threshold. Google's SKAdNetwork report shows installs and conversion values at the campaign level and uses a model trained on recent postbacks to redistribute the withheld values; its modeled conversions carry delays of up to 5 days, and integrated conversion measurement through an attribution partner is the more real-time option. None of this reaches a subscription renewal, which is why the revenue source has to be your own.

Work backwards. Decide the most a paying subscriber may cost from the plan price, gross margin and the renewal curve, then multiply by your measured trial rate and trial-to-paid rate for the CPI the campaign may pay. With a $90 target, a 6.2% trial rate and 40% trial-to-paid, that is $2.23. Compare it with the lead table and with your own last quarter: if the allowed CPI sits far below what the auctions cost, the work is in onboarding and pricing before it is in bidding.

Then size the budget to the target, not the other way round. Google's rule is 50 times the target CPI per day for an installs campaign and 10 times the target CPA for an in-app action campaign, so a $2.23 target CPI wants at least $112 a day, and a $48 trial target at least $480. Meta's advice is the lowest cost bid type without a bid cap, since a cap can limit results. A campaign starved below Google's figures reports a CPI that means nothing, because it never left the learning phase.

Read each campaign once a week on five lines: spend, installs and CPI; trials and cost per trial; paying users and cost per paying user from RevenueCat; day-7 or day-30 revenue per install for the cohorts old enough to have one; and the trial-to-paid rate of the campaign against the account's. The last two decide. A campaign with a low CPI and a trial-to-paid rate half the account's is buying the wrong installers, and the right move is a creative or a placement change, not a bid change. A campaign with a high CPI and the best revenue per install is the one to feed. Keep iOS and Android apart, since the postbacks arrive days later on iOS and the revenue per install differs by platform.

Change one thing per campaign per week and let it reach the conversion count before judging it. In Adrails, a budget step the Media Buyer prepares is capped at 30% per change and is a proposal you apply in one click, an automation can send the weekly table to Slack or email and pause what crosses a cost line, and the Analysis Expert explains a move with your RevenueCat figures next to the ad metrics, on Meta Ads and, once your workspace can connect it, Google Ads.

• Spend, installs, CPI: the platform's view
• Trials, paying users, their cost: RevenueCat's view
• Revenue per install at day 7 or day 30, by platform
• Trial-to-paid by campaign against the account: the quality check
• One change per campaign per week, read after it has reached its events