A mobile game soft launch is a process, and the process is a sequence of decisions taken in a fixed order. Most studios instrument the metrics thoroughly and never agree on what any of them would have to read for the answer to be no. Having managed 50+ launches at Gameloft, SFR and Blacknut across mobile and cloud gaming, I have watched well-instrumented teams spend six figures of UA on a soft launch and still ship a doomed title, because the go, extend and kill thresholds were written after the data arrived rather than before.
This guide is the playbook that closes that gap: the store gate you have to clear before you can run a soft launch at all, market selection against CPI bands with a named source, how to size the budget from cohort readability instead of a headline number, and the explicit criteria for extending, scaling, pivoting or killing.
Every retention figure on this page derives from our mobile game KPIs and benchmarks reference, which carries the population, the percentile and the vintage behind each one.
The Soft Launch Decision Framework
A soft launch exists to answer four binary questions, in this order:
- Does the core loop retain? Does D1 clear the line below which paid acquisition is wasted, without paid traffic itself flattering the early cohort?
- Does retention compound? Does the D7 and D30 curve hold, or does the game leak faster than the peer group?
- Does it monetise against your acquisition cost? Once retention is proven, does projected LTV clear the CPI you will actually pay at global scale?
- Does it scale? When you triple daily UA spend, do the unit economics survive?
Treating each phase as a separate decision rather than a rolling “are we ready yet” is the single biggest discipline gap I see in the field. Each question gets a phase, a budget, and an explicit go / extend / kill threshold agreed before data is collected.
What a Soft Launch Is For, and What It Is Not
A soft launch is a limited, controlled release in selected markets whose only job is to convert those four questions into a defensible go-global decision. It is not a marketing campaign, it is not a slow ramp to global, and it is not the place to validate creative strategy at scale. That last one is a separate UA discovery exercise with its own budget.
Titles rarely fail at global launch because the build was weak. They fail because nobody wrote down the kill criteria, so every weak signal got rationalised away in the Monday review and the runway ran out before the answer did.
The Gate Before the Gate: Google Play Closed Testing
This one catches small studios by surprise and can add a month to the plan before a single install is bought.
Google Play requires personal developer accounts created after 13 November 2023 to run a closed test before they can apply for production access. The Console help page is explicit: “At least 12 testers must be opted-in to your closed test when you apply for production access”, and “they must have been opted-in for the last 14 days continuously.” Accounts registered as an organisation sit outside the requirement.
Fourteen continuous days is the floor, not the expected duration. Recruiting twelve testers who actually opt in and stay opted in, then passing the production access review, realistically adds three to five weeks. Plan it as a phase of the soft launch rather than as a formality, or register as an organisation before you need the release.
Choosing Your Soft Launch Markets
Market selection decides what your data means. You want install costs you can afford, a language you already support, and player behaviour close enough to your eventual audience that the numbers transfer.
These are the ranges for casual and puzzle on Android, which is the segment most small studios are planning for. They come from our UA cost benchmarks reference, which explains why published CPI figures disagree so violently.
| Market | Android CPI, casual and puzzle | Confidence | Use it for |
|---|---|---|---|
| United States | $1.50 - $3.50 | Moderate to high | Not a first soft launch market |
| UK, Canada, Australia | $1.00 - $2.50 | Moderate | Clean early-funnel read in English |
| Western Europe (FR, DE) | $0.60 - $1.50 | Moderate | Tier 1 behaviour at lower cost |
| LATAM (Brazil, Mexico) | $0.15 - $0.60 | Moderate | Volume for retention reads |
| Southeast Asia | $0.20 - $0.60 | Low to moderate | Ad tolerance, cheap iteration |
| India | $0.08 - $0.30 | Low to moderate | Volume only, weakest LTV transfer |
Three things this table will not tell you. It is Android: for iOS in the same market and genre, apply a 3x to 4x multiplier, and treat any published iOS-to-Android ratio outside that band as a methodology artefact rather than a market fact. The Philippines sits inside the Southeast Asia band and does not have a separately published figure. And New Zealand, a perennial soft launch favourite, has no published band at all, so price it with your own test spend before you plan around it.
Run two markets in parallel. One that behaves like your eventual paying audience, one that gives you install volume for the price. A single cheap market will validate retention and mislead you about monetisation. Keep the United States, Japan, South Korea and China out of the first phase: competition is high, install costs are unforgiving, and burning budget there teaches you nothing you could not have learned for a fifth of the price.
For the trade-offs between Nordic, Canadian and Southeast Asian test markets in more depth, our soft launch market selection guide works through the full decision framework.
Sizing the Test: Cohorts, Ad Sets and Budget
Most soft launch budgets get set by picking a round number. Build it from the two constraints that actually bind.
Constraint one: cohort readability. Reading a D1 near 30% to within three points at 95% confidence requires 896 installs. Budget roughly 1,000 purchased installs per cell to absorb the drop between install and first session. Comparing two variants needs that on each arm. One readable cohort of 900 beats nine unreadable cells of 100, and the studios that split their traffic across six cells to “learn faster” learn nothing at all.
Constraint two: the algorithm’s learning floor. Below roughly $300 to $500 per day per ad set, the buying platform cannot separate signal from noise, and your CPI reading is an artefact of undersampling. That is the binding constraint at small scale, not the cohort maths: $300 to $500 a day is $9,000 to $15,000 per market per month, which at UK, Canada and Australia install costs buys somewhere between 3,600 and 15,000 installs. Comfortably more than one readable cohort, and enough for two arms.
Everything else is a line item on top:
| Line item | Lean run: 4 weeks, one market, one ad set | Full programme: 12 weeks, two markets |
|---|---|---|
| Paid media | $9,000 - $15,000 | $54,000 - $90,000 |
| Creative waves, 6-10 variations each | $2,400 - $6,000 (two waves) | $4,800 - $12,000 (four waves) |
| Attribution (MMP) | $0 on a free tier | $600 - $2,100 |
| Freelance UA management | $0 - $2,000 | $2,400 - $6,000 |
| Total | $11,000 - $23,000 | $62,000 - $110,000 |
Notes on the line items. Creative production runs roughly $150 to $350 per variation, consistent with two to four days of freelance work at typical European day rates, and variations on one proven concept cost far less than six original concepts. Attribution is free at the volumes a soft launch generates: Tenjin’s free tier covers 2,000 conversions, and paid plans start around $200 a month, which is where you land if you run two markets for a full quarter. Freelance UA management at one and a half to four days a month is a freelance rate, and it is what is available to you at this scale. Specialist UA agencies decline accounts below roughly $10,000 to $30,000 of monthly media spend, so a single-market soft launch running $9,000 to $15,000 sits at or under the floor where outsourced execution exists at all.
The variable that moves the total is how many cells you run in parallel, not how long the soft launch lasts. Adding a second ad set costs the same as adding four weeks.
Diagnostic Thresholds: the Stop-and-Fix Signals
The full target benchmarks live in the mobile game KPI guide. What matters during soft launch is the diagnostic line below which you stop spending and fix something.
| Phase | Diagnostic signal | What to do about it |
|---|---|---|
| Phase 1, core loop | D1 below 25% on Android | Stop buying. Onboarding or the first-session loop is broken. More traffic will not diagnose it. |
| Phase 2, retention curve | D7 below 8%, or D30 below 3% | Stop scaling. Deepen the meta layer and progression. Reassess after two to four weeks of iteration. |
| Phase 3, monetisation | Projected LTV below the CPI you will pay in Tier 1 | The economics do not close. Fix the retention curve or the ad load before touching pricing. |
| Phase 4, scale test | CPI rises more than 50% when daily spend triples | Channel saturated. Validate the next channel before committing to global. |
A word on each of these, because the temptation is always to negotiate with them.
The 25% D1 line is not a target, it is a floor. Median D1 across 11,600 live games measured by GameAnalytics is around 22%, and the top quartile is 25-27% on Android. So 25% puts you at roughly the top quartile of the whole market, and it is still the point below which paid acquisition destroys money. The casual and puzzle peer group on Android sat at 28-32% in AppsFlyer’s Q3 2022 genre data, the most recent broad genre breakdown anyone has published. That peer band is the number to beat, and it is four years old.
The D7 line is the one that ends most soft launches. Median D7 across the market is 3.4% to 3.9%. Top quartile is 7% to 8%. Meanwhile paid user acquisition at Tier 1 install costs on hybrid casual only starts to work around 18% D7, which is more than double the top quartile of the entire market. That single comparison explains why so many technically competent games never find a channel that pays back. If your D7 sits at 6% and your plan assumes scaled paid UA, the plan is the problem.
Phase 4 has no measured benchmark behind it. The 50% figure is an operating convention I use, not a published elasticity. What is defensible is the shape: if tripling spend moves CPI by half, the channel was carrying you on a narrow audience and the next dollar costs materially more than the last one.
For the levers once you are under a line, see our retention strategies guide, and for meta-layer design that moves D7 and D30 specifically, the hybrid casual playbook. Monetisation model trade-offs sit in the F2P monetization models comparison.
Iteration Cycles: How a Soft Launch Actually Runs
A soft launch is a sequence of iteration cycles, each tied to one of the four decision questions. Each cycle has a build, a test, a review, and a decision, and the decision is one of three words: continue, extend, kill.
Cycles 1-2: Technical Validation
Decision to make: do we put money behind this build?
- Deploy to the test markets, monitor crash rate, cold start time, device coverage
- Fix critical bugs before spending UA dollars on a broken build
- Gate: stability is non-negotiable. No paid traffic until crash rate clears your own ceiling, which should sit well below the point where Google Play flags an app for bad behaviour
Cycles 3-4: Core Loop Validation
Decision to make: is D1 above the line?
- Read D1 by cohort, by channel and by creative, on cells large enough to be readable
- Diagnose onboarding friction across sessions one, two and three
- Start A/B testing onboarding variants, on two arms rather than six
- Gate: D1 below 25% on Android triggers a stop-and-fix cycle. Not a “let us see whether it improves with more data”
Cycles 5-8: Retention and Monetisation Layering
Decision to make: do the unit economics close?
- Iterate on the D7 and D30 curve. Add or rebalance meta layers where the curve leaks
- Layer monetisation events only once the curve is stable
- Model cohort LTV against blended CPI, and check the ratio rather than a conversion rate
- Gate: projected LTV supports a CPI at 30% to 70% of it in the markets you intend to scale in
Cycles 9-12: Scale Test and Pre-Global
Decision to make: do we green-light global?
- Triple daily spend in one market and watch what CPI does
- Lock the global launch strategy and the first 90 days of LiveOps calendar
- Final polish, localisation, store listing optimisation
- Gate: all four questions answered yes, global commit signed off
The Go / Extend / Kill Decision Matrix
Not every game should go global. The hardest discipline in soft launch is making the kill call early enough to preserve runway for the next bet. Use this in the weekly decision review.
| Signal | Go | Extend, one more cycle | Kill or pivot |
|---|---|---|---|
| D1 retention | At or above the casual puzzle peer band by cycle 4 | Between 25% and the peer band, rising build over build | Below 25% after three build iterations |
| D7 retention | Above the top quartile of the market and holding | Below it but improving with each build | Flat or declining despite two or more build iterations |
| Monetisation | Projected LTV supports CPI at 30-50% of it | LTV supports CPI at 50-70% of it, payback is slow but real | CPI at or above projected LTV with no identified lever |
| CPI scale test | Rises under 30% when spend triples | Rises 30-50%, suspect channel mix | Rises above 50%, channel saturated with no next channel |
| Technical | Crash rate inside your own ceiling on all target devices | Isolated device class issues | Architectural blockers requiring a rebuild |
The CPI scale bands are continuous on purpose. An earlier version of this page put the extend band at 30-40% and the kill band above 50%, which left every result between 40% and 50% in no box at all. That is exactly how a soft launch gets extended by default.
Pivot signals, meaning kill the current configuration but keep the IP:
- Specific features perform while the rest does not. Strip back and ship leaner
- Player behaviour diverges from the design intent in an interesting direction. Re-pitch the core loop
- Market feedback points at a different genre framing. Restart at cycle 3 with a new creative angle
What separates studios that ship hits from studios that ship runners-up is not analytic depth. It is the willingness to walk into the cycle 6 review with the kill option genuinely on the table.
Four Ways Soft Launches Go Wrong
1. Launching in the wrong market
Starting in the United States costs roughly twice what the UK, Canada and Australia band costs on Android, and several times what Southeast Asia costs, for a read you could have bought far more cheaply. Starting only in Southeast Asia costs almost nothing and tells you almost nothing about whether Tier 1 players will pay.
2. Negotiating with early retention data
If D1 comes in at 22%, that is the market median and it is below the line where paid acquisition works. It will not drift upward on its own. Stop, diagnose the first session, ship a fix.
3. Rushing to monetisation
A title at the top of the retention distribution with weak monetisation is a fixable problem. A title at the median with excellent monetisation is not, because there is nobody left to monetise by D7. Retention sets the area under the LTV curve, and monetisation only sets its height.
4. Splitting traffic until nothing is readable
Six cells of 200 installs feel like more learning than two cells of 600. They are not. They are six numbers with confidence intervals wide enough to contain any conclusion you want.
The Soft Launch Readiness Checklist
Pre-launch
- Analytics and attribution SDK integrated, events validated end to end
- Google Play closed testing requirement cleared, or an organisation account in place
- Two test markets selected, one behavioural match and one volume market
- Media budget allocated at or above $9,000 per market per month, or the plan reduced to one market
- Diagnostic lines written down and agreed with stakeholders before any traffic runs
- Crash reporting and cold start monitoring active
Weekly during soft launch
- Crash rate and cold start inside your own ceiling
- D1 above 25% on Android and trending toward the peer band
- Cohorts at or above 900 installs per cell, or the cells consolidated
- At least two acquisition channels running
- A build shipping every cycle with a stated hypothesis behind it
- Cohort-level analysis, never aggregate
Go / no-go for global
- D1 at or above the casual and puzzle peer band, which was 28-32% on Android in AppsFlyer’s Q3 2022 data
- D7 above the market’s top quartile of 7-8%, with a documented path to whatever level your acquisition plan actually requires
- Projected LTV supports a CPI at 30% to 70% of it in your intended scale markets
- CPI stable within 30% when daily spend triples
- Technical stability confirmed across target devices and OS versions
- Global launch plan and budget approved
Notice what is not on this list: a D7 target above 15%. An earlier version of this page asked for one, which would have put the go/no-go gate at roughly twice the top quartile of the entire measured market. A gate nobody can clear is not a gate, it is a formality that gets waived.
What I Could Not Verify
Figures that appear in most soft launch guides, including earlier versions of this one, that I could not trace to a primary dataset. They are not published here.
- A CPI table by individual soft launch country. Only Android hypercasual country splits exist, from Tenjin in 2024. The Canada, Australia, Philippines and New Zealand figures this page used to print were attributed to a source that publishes nothing below the regional level
- A free-to-paying conversion rate for casual puzzle. Widely quoted as a 1.5% soft launch floor, never sourced. This page used to carry it as a phase gate
- The number of new game titles released annually across both stores. Frequently quoted at half a million, which is not the order of magnitude anyone’s release data supports, and I could not reach a primary release-count dataset to replace it. The claim is simply gone
- CPI elasticity when spend triples. The 30% and 50% bands in the matrix are my operating convention from running these tests. No published dataset measures the relationship
- Genre-level retention beyond casual and puzzle on Android. The last broad breakdown is AppsFlyer’s Q3 2022 data, and every 2026 genre grid I checked is reprinting it without saying so
Admitting a gap costs less than filling it. When a soft launch guide gives you a country-level CPI to the cent, check whether the page names a dataset and a date.
Conclusion
A soft launch that works is a soft launch where the decisions were written before the data arrived. Pick two markets for complementary reasons, size the budget from the learning floor rather than a round number, hold the diagnostic lines, and keep the kill option live through cycle 6.
Soft launch is one phase of a larger operating system. For the full picture from pre-production through scale, see the complete mobile game growth playbook. If your soft launch data is in and you want a senior read on whether to go, extend or kill, and on what to fix first, a focused engagement with a mobile game growth consultant turns the data into a defensible decision in weeks.
Need help with your soft launch strategy? With 20+ years of mobile gaming experience, I help studios size the test, read the cohorts and make the call. Explore our mobile game consulting services or book a free consultation to discuss your game.