Casual game marketing strategy: buying installs against ad revenue

India downloads casual games in enormous volume and pays for them barely at all. Sensor Tower's India Q2 2026 report, reported via TechCrunch, puts the market at roughly 6.3 billion quarterly downloads, flat since 2023, against $345M (₹3,036 crore) in consumer spend, up 35% year on year. The growth belongs to spend, not to downloads — India's story is monetisation, not volume. Games were only $106M (₹933 crore) of that spend.

That is the structural fact. Sensor Tower's 2026 ad monetisation work reports that in India, Indonesia and Brazil, advertising delivers 55–70% of total game revenue, while in the US in-app purchase carries the business. An India-primary casual game is an advertising business with a game attached. Most teams pick the India-shaped audience and plan the US-shaped economics, and the plan dies at month two.

The eCPM rows are the weakest data here: RevenueLab's 2026 AdMob compilation is a vendor blog and its India column is derived from a Tier-3 discount band rather than measured. Take the 10–20x India-to-US eCPM gap as load-bearing and the dollar values as order of magnitude. Tenjin's 2026 report charts eCPM but publishes revenue share only, so nobody has a clean measured number to give you.

I have cut three figures that appeared in the first draft of this post — a North America gaming CPI of $1.68, a LATAM figure of $0.14, and a paid-to-organic install ratio of 3.33 global and 3.81 India. I could not trace any of them to a named Adjust report with a stated period and sample, and a number I cannot source does not belong in a planning table, however convenient its shape.

Ludo King has passed 1.25 billion cumulative downloads per Sensor Tower, and India's Q2 2026 charts carry a cluster of minimalist puzzle titles — Arrow Puzzle, Arrows Puzzle Escape, Arrows GO! — while hypercasual revenue in India grew 180% quarter on quarter.

Has hybrid-casual displaced pure hyper-casual by 2026? Commercially yes, by volume no. Pure hypercasual no longer produces standalone chart-toppers and every large publisher has hybridised — Voodoo internally declared its old core business dead, SayGames reached four billion downloads by hybridising its hits, Habby defined the category. But Sensor Tower still puts hypercasual at roughly 40% of all game ad revenue, casual at 40% and hybrid at only 16%, with puzzle alone at 53%. And games outside the top 1,000 take 29% of ad revenue against 9% of in-app purchase revenue. The long tail pays in advertising, which is why an ad-monetised design is the structurally correct choice for a small publisher, not the poor cousin.

Decision rule: short sessions and an India-only audience means an ads-dominant economy, and — as the payback arithmetic two sections below shows — a hybrid revenue layer is closer to a requirement than an upside.

You are acquiring impression volume, not players. The activation event that predicts revenue in an ad-monetised game is a cumulative ad-impression threshold — instrument ads_watched_3 and ads_watched_10 as discrete events and optimise toward them. Tutorial completion is a quality gate, not a bidding signal.

Tune the level rung to your own data: usually 5–10 for pure hypercasual, 10–20, or the first meta stage, for hybrid.

Retention context, from Mistplay genre medians surfaced by Segwise. The sample size is undisclosed, so read these as medians among games with real install volume, not the median game. These are the numbers the payback table below runs on, so their weakness is the payback table's weakness too.

Hybrid-casual is absent from that table. Game Growth Advisor, a vendor blog, puts hybrid at roughly 20% D7 and 10% D30 — three to four times hypercasual D7, and the whole economic case for hybridising. The same source cites a market median D7 of 3.4–3.9% across 11,600 games; both are true, because genre medians are computed on games with volume and the median game has none.

Multiply ARPDAU by the cumulative DAU-days each install produces. That product is your cumulative revenue per install at any horizon, and it is the maximum CPI you can pay at 100% payback. Everything else in this post is downstream of it. Run it honestly and the India hypercasual case does not clear at the CPIs this category quotes.

First, bridge eCPM and ARPDAU

ARPDAU is not an independent benchmark you look up. It is an identity:

ARPDAU = impressions per DAU × eCPM ÷ 1,000

That identity is missing from most casual UA writing, and it is where the published bands fall apart. At India's rewarded eCPM band of $0.70–3.96 and a banner band of $0.03–0.09, a realistic blended eCPM across a mixed rewarded, interstitial and banner stack is around $1.50. Impressions per DAU is a design decision, not a benchmark; the figures below are model assumptions, stated as such.

Now compare that to the ARPDAU band this category publishes: Game Growth Advisor gives hypercasual $0.03–0.08 and hybrid-casual $0.15–0.50, and flags its own range as too wide for reliable planning. These do not reconcile for India. To reach $0.03 ARPDAU at a $1.50 blended India eCPM you need 20 impressions per DAU — a cadence that runs straight into Play's Ads policy on consecutive and repeated interstitials. To reach $0.03 at 6 impressions per DAU you need a $5.00 blended eCPM, above the top of India's entire rewarded band.

The resolution is that the $0.03–0.08 band is a global, Tier-1-weighted figure and does not describe India supply. Model India ARPDAU bottom-up from your own impressions per DAU and your own measured eCPM. Do not import the band. For the tables below I carry $0.02 as a generous India planning figure — generous, because the bottom-up identity says $0.009–0.018 is likelier at a policy-safe ad cadence.

Second, derive cumulative DAU-days from the retention curve

An install is worth the sum of the days it is active. Day 0 counts as 1.0 by definition — the install day. After that, retention at day d is the fraction of installs still active. Sum those fractions and you have DAU-days per install.

The published curves give you D1, D7 and D30 only, so days 2–6 and 8–29 have to be interpolated. Two reasonable interpolations bracket the answer, and I show both rather than pick one silently. Using hyper-casual, D1 29.31% and D7 5.90%:

Call it 1.95 DAU-days per install to day 7 — the midpoint. Extending the same method from D7 to D30 (5.90% to 1.38%) adds roughly 0.58 more, so about 2.5 DAU-days to day 30.

The same exercise on the other profiles:

*Hybrid D1 is not published anywhere; 32% is a model assumption taken from the match and puzzle rows. D7 and D30 are Game Growth Advisor's vendor-blog figures.

Third, multiply

Maximum payable CPI at 100% payback = ARPDAU × cumulative DAU-days at that horizon.

What the arithmetic says, including the part that is unflattering

At the published global band, the top of the recommended CPI band exceeds the top of the revenue band. $0.03–0.08 ARPDAU over 1.95 DAU-days is $0.06–0.16 of cumulative D7 revenue, against a recommended CPI band of $0.05–0.20. Only the bottom half of that CPI band pays back at day 7, and only if your ARPDAU sits at the top of its band.

At the $0.02 India ARPDAU this post says is likely, D7 revenue is $0.039 — below the entire CPI band. Not at the margin: below the cheapest install in the band. At D30 it reaches $0.051, which clears the very bottom of the band and nothing else, and does so on a 30-day payback that contradicts the seven-day discipline hypercasual is supposed to run on.

Run it bottom-up from the eCPM identity and it gets worse. At 12 impressions per DAU and a $1.50 blended India eCPM, ARPDAU is $0.018, D7 revenue per install is $0.035 and D30 revenue is $0.045. Both sit under ₹4.

The honest conclusion: pure hypercasual, India-only, with ad monetisation alone, does not clear at the CPIs this category quotes. You cannot fix that with better creative or a cheaper network, because the gap is structural — India eCPM is a tenth to a twentieth of Tier 1, and hypercasual retention gives you under two DAU-days to monetise. The three things that do change the answer:

1. Push impressions per DAU up, within Play's Ads policy. Twenty policy-safe impressions per DAU at $1.50 blended puts you at $0.030 ARPDAU and $0.059 D7 revenue, which clears the bottom of the CPI band. This is a session-design problem, not a media problem.
2. Add a meta layer and IAP. The hybrid row is the only one in the table with real headroom: $0.42 payable at D7 against a ₹13–35 CPI band. That headroom is why every large publisher hybridised, and the payback table is the argument in one line.
3. Buy at the very bottom of the band or not at all. If your validated India CPI comes in at ₹4–6 and your measured ARPDAU at $0.02+, you have a business. Above ₹10 on pure hypercasual you are buying revenue at a loss and the loss is visible by day seven.

What payback horizon does that imply, and what should kill criteria be?

The seven-day payback horizon that hypercasual writing repeats is a description of when you find out, not of when you get paid. In-app advertising realises 89% of its D60 revenue by day 7, so an IAA cohort is nearly fully read inside a week — but only against the revenue that cohort will ever produce, which the table above shows is under ₹5 per install in India.

For casual, the published anchor this post previously dropped is the one that matters: D7 ROAS 7.6% (7.8% iOS), peaking 9.34% in July, and D30 ROAS approximately 47% — Udonis citing Liftoff for the D7 figure and Segwise citing Adjust for D30. A curve that reaches 47% at day 30 and keeps flattening implies a four-to-five month payback, not a seven-day one. That is the real horizon for casual, and it is a different business from hypercasual's.

So kill criteria have to be a fraction of a stated horizon, not free-floating percentages:

• Reference trajectory: D7 ROAS 7.6%, D30 ROAS 47%, full payback at month four to five.
• Kill at D7 ROAS under 4% — roughly half the casual median. Half the median trajectory implies a payback horizon of eight to ten months, which no ad-monetised casual game survives, because your D60 revenue is already 89% realised by then and there is nothing left arriving.
• Kill at D7 ROAS under 6% at scale, because at scale you are buying broader supply and a cohort that starts below 6% will not average up.

State the horizon next to the kill line every time. A 4% D7 ROAS is only meaningful as "about half of the 7.6% casual median, implying roughly double the four-to-five month payback."

Three things: a mediation stack reporting impression-level ad revenue, an event taxonomy ending in a cumulative impression event, and a build clearing Google's technical quality thresholds. Miss any one and your first ₹5 lakh ($5,700) buys data you cannot act on.

Mediation and bidding, because UA cannot be planned without it. Your CPI ceiling is a function of ARPDAU, and ARPDAU is a function of your mediation setup. Pick one platform with full in-app bidding — AppLovin MAX or Unity LevelPlay — and run bidding rather than a hand-tuned waterfall. The industry moved decisively to bidding; hybrid setups keep a shrinking manual waterfall only for partners that lack it. At launch volume you will not have the impressions to make manual price floors statistically meaningful, so hand-tuning is a way to lose money slowly while feeling busy.

No 2026 dataset quantifies bidding's revenue lift over waterfall, so anyone quoting a percentage is quoting a sales deck.

The demand-partner count and the Families policy are in direct conflict, and you must resolve it before you build. The general advice is six to ten demand partners, with Mintegral and Pangle mattering more for India fill than their global share suggests — Tenjin's Q2 2026 split puts Pangle at 9% and Mintegral at 12% of Android game ad revenue, disproportionately Tier-3 supply. Sensor Tower puts AppLovin at 36% and AdMob at 29% of all game ad revenue globally.

But casual and arcade skew young in India, and Google Play's Families policy requires that apps targeting children or users of unknown age "Only use Google Play Families Self-Certified Ads SDKs." Google publishes that certified list and maintains it; Mintegral and Pangle are not generally on it. AdMob and Google Ad Manager are, and Unity Ads and Unity LevelPlay have historically appeared on it. Check the published list against your intended stack on the day you build it — the list changes, and a stack you assembled from a blog post is not a compliance position.

The consequence, stated as a cost rather than a caveat: if you declare a child or mixed audience, you lose roughly a fifth of Android game ad demand by Tenjin's Q2 2026 revenue split, and disproportionately more of your India fill, because the excluded networks are exactly the Tier-3-heavy ones. Model your ARPDAU on the certified subset from the first spreadsheet — at the bottom of the eCPM bands above, not the middle — and treat the full ten-partner stack as available only to a title that genuinely declares an adult audience and can defend that declaration against Google's audit of actual content.

Decide the age declaration before the ad stack, not after. Reversing it later means rebuilding mediation and re-forecasting revenue at the same time.

Impression-level ad revenue is table stakes. Send it from mediation to your MMP and forward it to Google and Meta, because tROAS on an IAA game has nothing to bid against without it. This is the most common instrumentation gap in a first casual launch.

Firebase or an MMP. Firebase is free and needs no new event code once the SDK is in, but it does not feed Meta and has no fraud detection. The moment a second paid network exists an MMP earns its fee in schema consistency alone, because Firebase for Google plus the Meta SDK for Meta gives you two incompatible definitions of the same event. Of the seven certified App Attribution Partners — Adjust, Airbridge, AppsFlyer, Branch, Kochava, Singular, Tenjin — pick one handling impression-level ad revenue natively.

The stability gate. Google publishes user-perceived quality thresholds — 1.09% crash rate, 0.47% ANR — and states apps below the bar are excluded from prominent discovery surfaces. Check Android vitals against both before switching on any campaign, because paid traffic amplifies whatever your build already does.

The production gate. Play requires 12 testers running a closed test for 14 continuous days before a personal developer account created after 13 November 2023 can apply for production access; organisation accounts are exempt. Testers who opt in for fewer than 14 days then opt out do not count, and if a tester opts out and returns the clock restarts. Review is typically under seven days on top. Older guides say 20 testers — that is outdated. Start the closed test at least four weeks before your intended launch date, because it is the one gate on this list that cannot be bought back with budget.

For an ad-monetised game, the iOS ATT opt-in rate is the primary determinant of iOS eCPM and therefore of whether iOS user acquisition pays back at all. An opted-out user cannot be served personalised advertising against an IDFA, so their impressions clear at contextual rates. Since your entire revenue is impressions, ATT is not a measurement footnote here — it is the revenue line.

The mechanism, as an identity you can plan on:

blended iOS eCPM = opt-in rate × personalised eCPM + (1 − opt-in rate) × non-personalised eCPM

No ATT opt-in benchmark with a named report, stated period and disclosed sample appears in this series' research base, so I will not print one as a benchmark. What follows is a model assumption you replace with your own measured rate in week two:

Read that as: at a plausible 30% opt-in, your iOS ARPDAU is roughly a third lower than your consented cohort's ARPDAU, and you must plan iOS CPI against the blended figure, not the consented one. Run the payback table above with iOS ARPDAU discounted by that share before you set an iOS bid. On India iOS, where the audience is a small slice to begin with, that discount is frequently what turns a marginal iOS plan into a losing one.

Prompt design, which is the only lever you control:

• Never show the system prompt on cold launch. A prompt fired before the user has seen the game is a denial you cannot re-request — iOS allows the system prompt once per install.
• Run a pre-prompt first, an in-game screen in your own art that explains what allowing does and gives the user a reason. Show the system prompt only to users who say yes to the pre-prompt; users who decline the pre-prompt can be asked again later, because you have not spent the system prompt on them.
• Fire it at a moment of demonstrated value — after the first rewarded ad completes, or at level three, not at level zero. The user has to have something to lose before "keep the game free" means anything.
• Do not offer a reward for allowing tracking. Apple's ATT guidance prohibits incentivising the request, and a rewarded-currency bribe attached to the prompt is a review finding.
• Localise the pre-prompt copy. The system prompt string itself is your NSUserTrackingUsageDescription and is also localisable; write it as a plain sentence about ads staying relevant, not as a legal notice.

The consequence for planning: validate the ladder on Android, where GAID persists and attribution is deterministic, measure your Android ARPDAU and your iOS opt-in rate separately, then price iOS from the blended identity above. Treating iOS as Android with a 1.5x bid multiplier is the specific mistake this section exists to prevent.

Two things are category-specific: your title carries the mechanic rather than a coined brand word, and your ad cadence is a store policy surface, not only a UX choice. Play's Ads policy is the rule most hypercasual launches trip, and Apple's 4.3 escalates to Developer Program removal rather than app rejection.

iOS gives 30 characters of title, 30 of subtitle and a 100-character keyword field users never see — put competitive and long-tail terms there and never repeat a title word. Play gives a 30-character title, an indexed 80-character short description carrying heavy weight, and a 4,000-character indexed long description. That short description is the highest-leverage field most teams waste. Run Custom Store Listings by channel and Store Listing Experiments from day one.

Keyword clusters observed in live listings, not from a volume tool — commission a keyword pull before committing title copy. India: ludo, ludo offline, carrom, offline games, free games, puzzle game, cricket game, plus the arrow-puzzle terms the Q2 2026 charts surfaced. Hindi and transliterated Hindi are under-competed on Play India and the localised listing is indexed separately.

The category-specific risks:

• Apple 4.3(a) prohibits multiple Bundle IDs of the same app, which kills the hypercasual habit of shipping ten skins of one mechanic on iOS. Apple's remedy is one app with variations via in-app purchase.
• Apple 4.3(b) is account-level: "Don't submit apps that are indistinguishable from what's already widely available... Repeated submissions of this kind may lead to removal from the Apple Developer Program." Mitigate with distinct art direction and a real meta layer. Separate developer accounts work only where ownership is genuinely separate; shell accounts read as evasion.
• Apple 3.1.1 on loot boxes, verbatim: "Apps offering 'loot boxes' or other mechanisms that provide randomized virtual items for purchase must disclose the odds of receiving each type of item to customers prior to purchase." Google Play runs an equivalent odds-disclosure requirement for randomised virtual items. This post recommends a meta layer and a dual-currency economy, and those two design choices are exactly what produce a gacha or chest mechanic. Prior to purchase is the operative phrase: the odds have to be visible on the purchase surface, not buried in a settings screen or a support page. Build the odds panel into the chest UI in the first version, because retrofitting it means a store submission at the moment your economy starts earning.
• Apple 2.3.8 requires icon, screenshots and previews to hold to a 4+ rating even if the app is rated higher. Your ad creative can be edgier than your store assets, so plan two creative tracks.
• Apple's privacy manifest and SDK signature requirements apply to every third-party SDK you ship, which in a six-to-ten-partner mediation stack means every one of them. Each SDK on Apple's commonly-used list must supply a privacy manifest declaring collected data types and any required-reason API usage, and must ship a valid signature that Xcode validates at build. Your own app's privacy report is assembled from those manifests, so a partner who has not updated theirs blocks your submission, not just theirs. Audit manifest and signature status for every network before you add it, and re-audit before every release — a mediation partner added in a hurry for India fill is the most likely thing to fail this check.
• Play Ads policy prohibits interstitials at the start of a level or content segment and before splash screens, requires full-screen interstitials to be closeable after 15 seconds except opted-in rewarded ads, and disallows consecutive ads after one user action. Interstitial-after-every-level is enforcement-exposed. Note the interaction with the payback table: the impressions-per-DAU number that makes India economics work is capped by this policy, not by your ambition.
• Play Families policy triggers if your audience skews young, which casual and arcade do in India. It forces certified ads SDKs for child or unknown-age users, neutral age screening for mixed audiences, and no persistent identifiers from child-targeted apps. See the mediation section above for the demand-share cost.
• A certified consent management platform for EEA and UK ad serving. If you serve ads to users in the EEA or UK — and an India-only media plan does not stop organic installs arriving from those markets — Google requires a CMP certified under its own programme and integrated with the IAB Transparency and Consent Framework for traffic monetised through AdMob and Ad Manager. Non-consented traffic must be served non-personalised ads or none. Wire the CMP in before launch; retrofitting consent into a live mediation stack means a period where you either break policy or serve nothing.
• India's Promotion and Regulation of Online Gaming Act 2025, Rules in force 1 May 2026. Your game is most likely an "online social game" and may need registration — budget legal time. Any entry fee, stake or cash-out makes it an online money game, which is banned outright, with advertising separately criminalised at up to two years or ₹50 lakh. Be careful with Ludo, Rummy, Carrom and Teen Patti-adjacent mechanics, which is exactly the tabletop genre topping India's download charts, and refuse cross-promo inventory from real-money gaming apps.

Run Apple's documented four-campaign structure, expect Games to be the platform's most expensive vertical, and size the budget against a hard fact about the India audience: India iOS is roughly 4–6% of devices. Apple Ads India is a quality channel, not a volume channel, and any allocation above about 15% of an India-only budget is buying a slice of the market that is smaller than the slice of budget you assigned it.

AppTweak's 2025 dataset — roughly 3,500 apps, 50,000 campaigns, $1B in spend across 38 countries — puts US search-results cost per tap for Games at $4.21 and cost per install at $12.28, against a global median CPI of $1.80 and India at ₹78 ($0.89), with India tap-through at 5.2% and install-per-tap at 48%.

Where AppTweak is your planning anchor, carry the cross-panel range rather than a single number. Adapty's panel, 1M+ ad groups across 90 countries, puts US cost per tap at $1.58 against AppTweak's $1.91 — the same metric, two panels, a 21% spread, and no India cut from Adapty at all. Plan India against ₹78 as a midpoint with an explicit ±20% band, not as a point estimate.

Add every keyword from the first three campaigns to Discovery as exact-match negatives so Discovery only spends on genuinely new terms. Splits like 30/35/30/5 are convention, not Apple guidance.

Search results now supports Maximize Conversions, launched 25 February 2026. Apple's documented guidance: budget for at least five conversions a day, run at least two weeks before judging, and treat target CPA as a weekly average goal rather than a per-query ceiling. Search Match is mandatory on the automatic ad group, and it is unavailable for pre-order campaigns.

That mandatory Search Match conflicts with the exact-match discipline of the four-campaign structure and Apple has published nothing reconciling them. Run the manual structure, or run a Maximize Conversions campaign alongside it with separated budgets. Do not merge them.

Lifetime budgets were paused in June 2026, so daily budget is the only model and monthly maximum is daily times 30.4. Apple documents no minimum daily budget for Advanced, but five conversions a day is the real floor — about ₹390 ($4.40) a day in India at the ₹78 CPI, against roughly ₹5,400 ($61) a day on US search results at a $12.28 Games CPI. The India floor is trivially clearable, which is precisely why the channel tempts over-allocation. Size it against the 4–6% device share instead.

Custom Product Pages are worth the setup — up to 70 per app, search-results ad groups only, no new app version needed. Build one per core mechanic so Category and Competitor land on matching screenshots.

Android is where an India-only casual game lives, and App Campaigns have published budget floors most launches violate. Installs campaigns with target CPI need a daily budget of at least 50 times bid; with target CPA on an in-app action the floor is 10 times bid; Engagement campaigns need 50,000 installs before you can run them at all, plus 15 times bid.

Google's learning threshold is a count, not a window, and its wording is exact: "Making changes to your in-flight campaign before the first 100 conversions have registered may disrupt learning." Fund the campaign to clear 100 conversions and keep your hands off it.

Also documented: one action only for target CPA; bid iOS at roughly 1.5 times Android; bid roughly 20% above baseline target CPI when targeting in-app-action users; keep one approved asset of each type per ad group and Ad Strength at "Good".

Asset specs, since games fail on these often. Headlines one to five at 30 characters, descriptions one to five at 90, images up to 20 per ratio at 5MB. Video runs 10 to 60 seconds in 16:9, 9:16 and 1:1 and must be hosted on YouTube first — supply none and Google generates video from your store listing, which for a game is always worse than what you would ship. HTML5, 20 ZIPs per ad group at 5MB, is your playable slot.

Two silent deep-link failures: robots.txt must allow AdsBot-Google and AdsBot-Google-Mobile to reach apple-app-site-association and assetlinks.json, the most common misconfiguration in the channel; and Android installs need deferred deep links in your measurement SDK or post-install routing fails quietly.

For launch-day velocity, Play pre-registration runs 90 days maximum per country, two apps at a time, with a day-1 push to everyone registered and automatic install on launch day — but only up to 200MB on Wi-Fi and cellular, Wi-Fi only from 200MB to 2GB, and not at all above 2GB. Nobody mentions those size cliffs and they matter for anything with a large asset bundle. Exactly one reward exists for the campaign's lifetime and cannot be edited once created. Google publishes no statement that pre-registration confers a ranking benefit; the install-concentration mechanism is real, the ranking claim is inference.

Before you buy a burst. Apple's February 2026 Guidelines revision added clause 5.6.3, "Discovery Fraud", under the Developer Code of Conduct, where the remedy is termination of the Developer Program account rather than app rejection. Apple's sentence in full: "Manipulating any element of the App Store customer experience, such as charts, search, reviews or referrals to your app, erodes customer trust and is not permitted." Section 3's preamble catches agencies: "or engage with third-party services to do so on your behalf." Google filters incentivised installs out of ranking systems, escalating to top-chart then store removal, so a burst can be silently zeroed while you still pay for it — and the account remedy cascades: "any related Google Play developer accounts will also be permanently suspended." Games carry a third mechanism: Google's published thresholds of 1.09% crash rate and 0.47% ANR. A burst of low-intent installs onto an unstable build pushes you past that line, and exclusion from prominent discovery surfaces is the only ranking penalty either store has ever quantified.

The legitimate version produces the same curve: pre-registration cohort, waitlist, PR embargo, creator posts and paid campaigns all landing on one day. The independent evidence on what a burst actually buys is thin — an ACM IMC 2020 study of real incentivised campaigns put top-chart appearance at 3.1% baseline, 7.5% on vetted platforms and 2.5% on unvetted ones, so the cheap platforms produced no detectable ranking benefit at all.

Meta is your video and UGC channel and explicitly not your playable channel. Liftoff's creative report states directly that playables do not operate at meaningful scale on Meta, TikTok or Snap. So you run two creative systems — playables on the ad networks, video on Meta — and budget for both rather than expecting one to serve both.

Optimisation goals in the order your volume allows: App Installs is Meta's default and the right cold start; App Event Optimization, charged on impressions, is where you move once ads_watched_3 has volume; Value Optimization needs purchase data and Meta describes it as available on a limited basis to partners and advertisers, so do not plan on it. Meta discourages Link Click Optimization if the SDK is installed.

The learning-phase figure everyone quotes — roughly 50 optimisation events per ad set per rolling seven days — is trade press only, not primary-verifiable, so attribute it that way. Budget, creative, audience and event changes all reset it, and raising budget purely to escape the "learning limited" label typically yields 5–10%. Consolidating into fewer, larger ad sets is the structural fix.

On iOS the eight-event Aggregated Event Measurement model still governs app campaigns even though the standalone AEM tab was removed for many web accounts in 2026. Rank ads_watched_3 high, because reordering pauses affected ad sets for a 72-hour cooldown. On the Conversions API for App Events, the mandatory parameters are action_source = "app", advertiser_tracking_enabled, extinfo and event_id, with deduplication on event_id plus event_name. Install events get automatic 90-day deduplication; post-install events do not, so impression events need their own dedup discipline. Note that advertiser_tracking_enabled carries the ATT state discussed above — the same signal that sets your iOS eCPM also sets what Meta can optimise against.

Pick the deepest event that still clears its platform threshold at your intended budget, working backwards from the published gate rather than forwards from ambition.

The rate that drives all of this is your install-to-ads_watched_3 rate, and no published benchmark for it exists in any source I could find. The figure below is a model assumption you replace with measured data in week two:

At a 25% ads_watched_3 rate and Meta's roughly 50 events per ad set per week, you need about 200 installs per ad set per week. At an India Android CPI of ₹13–35 ($0.15–0.40) that is ₹2,600–7,000 ($30–80) per ad set per week — trivially affordable, which is the whole reason to validate the ladder in India.

Do the budgets in this post clear the gates? Yes, and here is the demonstration.

• Validation, ₹4 lakh a month, 85% to Android India. That is ₹3,40,000 a month, or ₹11,184 a day. Google's tCPI floor is 50 times bid; at a ₹35 India casual bid the floor is ₹1,750 a day. The Android line clears it with roughly 6.4x headroom, and buys about 9,714 installs a month, so the 100-conversion learning threshold on installs clears inside the first day.
• Scale, Meta at 20% of ₹15 lakh, so ₹3 lakh a month. That is ₹69,079 a week. At a ₹35 CPI that is 1,974 installs a week; at the assumed 25% ads_watched_3 rate, about 493 events a week. Against Meta's roughly 50 events per ad set per week, that supports about nine ad sets — enough for a real creative test matrix without any ad set sitting in permanent learning.

Move to tCPA on ads_watched_3 only once you have measured the real rate. If it comes back at 10% rather than 25%, the same ₹3 lakh Meta budget supports three or four ad sets, not nine, and the correct response is fewer, larger ad sets — not more budget to escape the learning label.

Four compounding mechanisms make premature depth fail. Statistical starvation: the ad set never exits learning, delivery throttles, CPA inflates. Sparse-signal overfitting: the model fits noise, which is why Google tells you not to select more than one action for tCPA. Reset cascades: every fix restarts learning. And iOS signal decay, the one most playbooks miss — deep events land in AdAttributionKit windows two and three, which return coarse values and never fine, and at Tier 0 crowd anonymity those windows produce no postback at all.

The consequence is concrete: encode your most important signal — cumulative impressions, or bucketed day-0 ad revenue — into the first postback, days zero to two, the only window carrying a fine conversion value and only at Tier 2 or 3. Validate the ladder on Android, where GAID persists and attribution stays deterministic, then port it. Android Privacy Sandbox was cancelled in October 2025, so any playbook preparing Android for an ATT-style event is wrong.

Playables and gameplay-first video, on different networks. Liftoff's creative report — 4.7 trillion impressions and 1.1 billion installs between January 2023 and May 2025 — found playables deliver eight times the impression-to-install rate of non-playable formats for top-spending game advertisers and sixteen times for everyone else. The advantage is larger for smaller advertisers.

Three more findings from the same report: top games put 35% more spend into playables; introducing UGC lifted impression-to-install by 152%; creative-audience alignment lifted installs per mille by 93%.

The hook pattern has not changed. Real gameplay in the first one to two seconds, no logo, no studio bumper, vertical 9:16, legible with sound off. For hybrid, add the meta reveal at five to eight seconds — the base, the collection, the upgrade screen — because that pre-qualifies for retention rather than raw installs.

The failure mode worth naming: scaling UA on a hook that works before the meta is tuned. A great two-second hook buys installs your retention curve cannot hold, and the payback table earlier in this post is what that costs — under two DAU-days per install and under ₹5 of revenue to recover the CPI from.

Plan against the maximum payable CPI derived earlier, not against a subscription LTV you do not have. Cumulative ARPU by your stated payback horizon must exceed CPI, and the honest horizon for casual is four to five months on a 7.6% D7 / 47% D30 ROAS trajectory — not the seven days hypercasual folklore quotes.

AppsFlyer's State of App Monetization 2026, covering January 2025 to March 2026 across $900M of in-app purchase, $800M of subscription and $7.2B of ad revenue. Note the two ARPU rows are different metrics with different denominators, which the first draft of this post collapsed into one:

The blank cell is the number this post most needs and nobody publishes it. There is no D90 in-app-advertising ARPU for casual anywhere in AppsFlyer's report or in any other source I could find, which means the single figure that would settle whether an ad-monetised casual game pays back in India does not exist. That absence is why this post derives the answer from ARPDAU and retention instead, and why the derivation carries an explicit assumption at every step.

The D7 realisation row is the most useful line in the table. In-app advertising realises 89% of its D60 revenue by day 7, against 52% for subscriptions. An IAA game finds out whether it works inside a week — brutal, and a gift, because you can kill a bad cohort cheaply.

On cost, the most-cited gaming CPI number is a global blended $0.56 (₹49) attributed to Adjust and reaching this post through Segwise's 2026 CPI, IPM and ROAS page. I cite the intermediary because I could not locate the underlying Adjust report with a period and sample, and because the same Segwise page also gives iOS at $4.22 and Android at $2.97, which cannot all describe one population. Udonis, citing Liftoff's casual gaming report, gives casual CPI around $1.00 overall, $0.63 Android and $2.23 iOS.

My planning position, stated as a position and not a benchmark: India Android hypercasual ₹4–18 ($0.05–0.20), India Android casual ₹13–35 ($0.15–0.40). These are inference from the Tier-3 eCPM ratio and are unverified — validate with a ₹2–4 lakh ($2,300–4,500) test. US figures, for contrast only: casual Android ₹132–264 ($1.50–3.00), casual iOS ₹220–440 ($2.50–5.00).

On the organic halo, use the causal number. The only causal estimate available is MIT's 2025 work by Ju, Zhao and Aral — three advertising-shutoff experiments across six apps over 500 days — which found every $100 of ad spend produced about 37 paid installs and about 3 organic, an organic halo of roughly 8%. Plan blended CPI as paid CPI divided by 1.08, and nothing more generous. A correlational paid-to-organic install ratio is not a causal organic bonus, and deflating your planning CPI by a quarter on the strength of one is how a plan that looked fine at 26% headroom turns out to have 8%.

One benchmark that does not exist. The Tenjin and GameAnalytics hyper-casual CPI series, the one every 2019-era deck cited, was discontinued around Q4 2022. Tenjin's current report covers eCPM only, with no CPI and no India cut. Anyone selling you "2026 hyper-casual CPI benchmarks" from that series is citing something that has not been published in four years. Any CPI number without a named report, a stated period and a disclosed sample is marketing.

The creative production line is not optional and is missing from most published budget scenarios. Fifty to eighty variants a month is the working cadence in this category and it has a cost; a media budget that does not fund it is a media budget that will spend month three re-running the same three ads at rising frequency.

The KPI ladder, in weekly checking order: tutorial completion, D1 retention, impressions per DAU, blended eCPM, ARPDAU, D0 and D3 ROAS, D7 ROAS, then CPI. CPI is last deliberately — a cheap install into a game showing two ads a day is worse than an expensive install into one showing eight, and the payback table is the arithmetic that proves it.

Seven failures account for most stalled casual launches, five of them self-inflicted before an ad runs.

Never doing the payback multiplication. Keeping eCPM, ARPDAU, retention and CPI in four separate tables and never multiplying them is how a plan gets approved that its own numbers say loses money. Cost: the full media budget, discovered at day 30 rather than on the spreadsheet.

Optimising on install and calling a cheap CPI a win. India CPI is cheap because India eCPM is cheap, so buying on CPI alone means systematically buying the lowest-ARPDAU supply on the market. Cost: the media budget, discovered at day 30.

Shipping hypercasual portfolio strategy to iOS. 4.3(a) prohibits multiple Bundle IDs of the same app and 4.3(b) escalates repeated variants to Developer Program removal. Cost: your account, not your app.

Ad cadence that violates Play's Ads policy. Cost: removal, at the point your mediation is finally earning — and the cadence you needed to make India ARPDAU work is exactly the cadence the policy caps.

A Families declaration made late. Declaring a child or mixed audience after building the ad stack forces certified SDKs only, cutting roughly a fifth of Android demand and disproportionately more India fill. Decide first.

Buying from a CPI network without the IO clauses. AppsFlyer's 2026 ad fraud work across 106.4 billion installs puts iOS fraud at 11.7%, Android at roughly 14–15%, and affiliates near 40% against self-reporting networks near 1% — and affiliate is exactly how CPI networks are structured. An MMP's fraud rate is its own detection rate, so it is a floor. And the tell: CPI-network rates running 30–60% below the equivalent App Campaigns or Meta rate for the same geo are not efficiency, because the discount lines up almost exactly with the affiliate fraud rate. Demand in writing that your MMP is source of truth, blocked installs are non-billable, clawback runs at least 30 days, sub-publisher IDs are transparent, and no incentivised inventory is used. These protect your money, not your account.

Real-money adjacency in India. Any entry fee, stake or cash-out makes your game an online money game under the 2025 Act. Cost: criminal exposure, not a policy strike.