SuperScale

Insights

Predictive ROAS for mobile games: forecast payback before you scale

What ROAS prediction is, how to read ROAS by day, how to set a payback window, what App Tracking Transparency changed, and how SuperScale predicts it at cohort and creative level.

What is ROAS prediction?

Return on ad spend (ROAS) is the revenue a cohort of installs has returned, divided by what those installs cost, measured at a given day after install. Unless a campaign is buying brand awareness, ROAS is the number that says whether it made money.

Predicted ROAS is an estimate of where that cohort's ROAS will land at a later day (day 30, day 90, day 360) from what it has done so far. No model is exact, so a useful prediction comes as a curve with a range on either side rather than a single figure, and the range narrows as the cohort ages. The point of predicting rather than waiting is that the budget decision has to be made now: a campaign that will pay back at day 180 needs its spend defended or cut in week one.

How to read ROAS by day

ROAS is only meaningful with a day attached. Each checkpoint answers a different question, and the mistake most teams make is asking a late-day question of an early-day number.

DayWhat it tells youWhat to do with it
D0 to D1Install quality and first-session monetisation. Too early for ROAS.Read retention and CPI instead; kill obvious misfires.
D3Whether the cohort monetises at all. First creative-level signal.Pause creatives with no purchases; keep budgets small.
D7The standard early checkpoint. Network optimisers and SKAN postback windows line up around it.First usable prediction; set per-network and per-creative targets.
D14Confirms or reverses the D7 read; the anchor for ratio-based forecasting.Scale what held, cut what reversed.
D30The common payback checkpoint for casual and hybrid-casual titles.Judge against the payback target, net of fees.
D90 to D360Where payback is decided for mid-core and strategy titles with long tails.The prediction carries the decision; waiting is not an option.

Retention shapes the whole curve. Adjust's all-games baseline for 2025 is 27% on day 1, 13% on day 7 and 5% on day 30 (Mobile app trends 2026); a cohort sitting under the baseline for its genre will not grow into an optimistic prediction, however good day 1 looked. For a market anchor on the checkpoint itself, Liftoff's 2025 Casual Gaming Apps Report (with Singular; data Feb 2024 to Feb 2025) measured casual-game D30 ROAS at 47% on iOS and 15% on Android. The Android figure looks worse than it is. The same report has casual installs at $1.41 on iOS against $0.14 on Android, so the comparison is per dollar, not per install, and a cheap install returning 15% by day 30 can still be the better buy.

Setting a payback window

Pick the day

The payback window is the day by which a cohort must have returned its cost, and it is set by two things: the shape of your LTV curve and how long your cash can wait. A casual or hybrid-casual title earns most of its lifetime revenue early, so day 30 to day 90 is a typical window. Mid-core and strategy titles earn over a long tail and commonly work to day 180 or day 360. Our guide to player LTV covers how to read the curve from your own cohorts. Whatever day you pick, the target has to be net: the store's cut, the cost of running the campaigns and, on managed spend, the management fee all come off before a cohort has paid back.

Set the target

Breakeven is 100% ROAS at the window, net of those fees; the target sits above it by whatever margin the business needs. Then hold every network, geo and creative to the same window, so the numbers are comparable. It is always better to run a conservative prediction that may understate ROAS than an optimistic one that motivates the team to spend past the point where it pays. On one title we took over, ROAS went from 50% to 153% and CPI from $2.80 to $0.60, reaching breakeven in four months.

What changed for predictive analytics in mobile game and blockchain apps?

Since App Tracking Transparency arrived with iOS 14.5 in April 2021, iOS attribution for most installs comes through SKAdNetwork and its successor, AdAttributionKit. The postbacks are aggregated, delayed and privacy-thresholded, and they carry a coarse conversion value rather than a user-level record. For prediction that means three things: the early revenue signal on iOS is coarse, the later windows arrive days to weeks after install, and there is no user-level join between an install and its later lifetime value. Predictions on iOS therefore lean on aggregated cohorts, modelled attribution and the deterministic signal that still exists on Android, where Google retired its Privacy Sandbox initiative in October 2025 without deprecating the GAID. The mechanics are in our guide to SKAN and IDFA; the practical consequence for a model is that it has to work from cohort curves, not from stitched user journeys.

With blockchain and Web3 games the landscape is different. Mass adoption is still arguable, and there are few platforms that can offer reliable predictive analytics for them; even sourcing a dashboard that pulls on-chain and off-chain data into one place is difficult.

How to use a prediction in the budget decision

A prediction is only worth having if it changes what you spend. The decision is the same at every checkpoint: compare each cohort's predicted ROAS at your window with the target, and act on the gap.

  • Above target: scale in steps, and watch CPI as you do. A network's cost per install usually rises as budget rises, so the prediction for the next cohort is not the prediction for the last one.
  • Near target: hold, and let the cohort mature to the next checkpoint before deciding. Most bad calls are made on cohorts that were three days too young to read.
  • Below target: cut, and move the budget to the network or creative that clears. Keep a testing reserve so the plan can keep learning.

Apply the rule per network and per creative rather than per campaign, because a campaign blends cohorts that pay back with cohorts that never will. Judge networks on incremental lift rather than on the last click, since networks that share an audience will each claim the same install. This is what the prediction is for: forecast payback before you scale, so budget flows to the cohorts that compound, not the ones that only look good on day one.

How does SuperScale predict ROAS for mobile and blockchain games?

There is a lot to consider when predicting ROAS. The first is how far in the future the ROAS can be predicted with confidence (3 months, 1 to 2 years). The longer the horizon, the harder it is to predict with certainty.

The level of granularity needed for a useful prediction also matters. In some cases, daily cohort analysis is enough, but in most cases you need at least campaign or user-level data, which is hard to get for mobile games after IDFA. Wherever historical data exists, the models are more accurate and therefore more useful.

Standard modeling approaches to predict ROAS

Option 1: take the last available ROAS number (say day 14) and assume a stable ratio between 6-month ROAS and 14-day ROAS. From historical data you can estimate that ratio and use it to predict 6-month ROAS from the 14-day figure.

Option 2: estimate retention to calculate the lifetime of a cohort and multiply it by average revenue per daily active user (ARPDAU).

Both are useful, and both only work at an aggregated level. They need large user samples, so they are not much help on a single smaller campaign, which is exactly where the budget decisions get made.

What's different when SuperScale predicts ROAS?

We take into account more than the last available ROAS number. Daily growth over the previous days is a signal that a model reading only the latest revenue figure ignores entirely.

We identify cohorts that stopped growing sooner than the average cohort because of player churn, which keeps the model from overestimating. A conservative prediction that may underestimate ROAS beats an optimistic one that motivates UA managers to spend more than is profitable.

Recognising the different monetisation behaviours of different user segments is also key. Grouping players into daily cohorts mixes every segment together and loses information; separate predictions per segment give better overall accuracy.

The prediction pipeline is built on experience from 200+ games across genres. The outputs are cohort- and creative-level ROAS predictions, computed in SuperPlatform, deployed in your own BigQuery, and surfaced in SuperVYZR, so the same numbers that direct the budget are the numbers your own team can query. You own the data.

On-chain and off-chain data on one dashboard

SuperVYZR dashboards give you data visibility across both mobile games and blockchain apps: on-chain data (Ethereum, Solana, Binance Smart Chain, Tron and others) and off-chain data (Meta, Google UAC, TikTok and the rest) pulled into one place.

The predictive dashboard runs the ROAS model described above, so marketers can allocate budget from one set of numbers and, for mobile games specifically, from post-IDFA predictions they can rely on. Our team supports the setup from the first connector to the finished dashboards and beyond.

Results from games we run

On one title we took over, ROAS went from 50% to 153% (3x uplift) and CPI from $2.80 to $0.60 (-78%), reaching breakeven in 4 months. On another, UA ROI moved from -39% to +11%, worth $1.37M in extra profit after all fees. The write-ups are in our case studies; the managed service that runs on these predictions is SuperMedia, 35+ ad networks under one IO.

FAQ

Frequently asked questions

What is D7 ROAS, and what is a good D7 ROAS for a mobile game?
D7 ROAS is the revenue a cohort of installs has returned by day 7, divided by what those installs cost. It is the first checkpoint most UA teams optimise on because the networks' own optimisers and the SKAN postback windows line up around it. There is no universal good number: a casual title that has to pay back by day 30 needs a far higher D7 than a strategy title paying back at day 180 or later. Read D7 against your own LTV curve and payback window, not against a benchmark.
How do you track ROAS for mobile game campaigns?
Attribute installs and revenue to campaigns through a mobile measurement partner; on iOS that means SKAdNetwork or AdAttributionKit postbacks with conversion values rather than user-level matches. Pull the cost from every network and the attributed revenue into one dataset with one set of definitions, then report ROAS by cohort day (D1, D3, D7, D14, D30, D90, D360) per network and per creative. Two consoles that both say "ROAS" rarely mean the same thing by it.
How long until you can see ROAS results from a UA campaign?
The first monetisation read arrives around day 3. A usable prediction needs about a week and enough installs to make a cohort readable. Payback itself is only confirmed at the window you set, anywhere from day 30 for casual to day 360 for mid-core and strategy titles. On iOS, SKAN's later postbacks add days to weeks of delay on top of that, which is why predictions carry the decision on iOS rather than waiting for the actuals.
How do you predict LTV for mobile game users?
The two standard approaches are a retention curve multiplied by average revenue per daily active user, and a fixed ratio between an early ROAS (say day 14) and the later one you care about, both estimated from historical cohorts. They work at an aggregated level and need large samples. Cohort-curve modelling adds the daily growth signal, adjusts for cohorts whose growth stopped early because of churn, and predicts per player segment rather than per calendar cohort, which is what SuperScale runs. Our LTV guide covers the cohort maths.
What is a good ROAS for a mobile game?
Breakeven is 100% at your payback window, net of platform fees and the cost of running the campaigns; anything above it is profit. For a market anchor, Liftoff's 2025 Casual Gaming Apps Report (with Singular; data Feb 2024 to Feb 2025) measured casual-game D30 ROAS at 47% on iOS and 15% on Android, on installs that cost $1.41 and $0.14 respectively. Genre moves the number a long way: a strategy title's D30 ROAS is lower by design because its revenue arrives over a much longer tail.
How does predictive targeting work for mobile game advertising?
On the network side, value-based optimisation steers delivery toward users the network's own models predict will install, purchase or spend, based on the conversion events you send back. On your side, the job is to send the right optimisation event, give the model enough conversions to learn from, and judge each network by the predicted ROAS of the cohorts it delivers rather than by its own reporting, since every network grades its own homework.
How do you improve ROAS for a mobile game?
In order: fix retention before spending more, because a cohort under the market baseline will not grow into its prediction; cut the networks, geos and creatives whose predicted ROAS sits under target and move that budget to the ones above it; test creative continuously, since video and playables now take most game ad impressions; and expand beyond the core networks you have maxed out, where the cheaper inventory sits. Managed UA across 35+ networks under one IO is how we do that for publishers.