Soft Launch: What Creative Testing Should Actually Tell You

The short version

A soft launch exists to answer questions before you spend at scale. The creative question is usually the one that gets least attention and matters most, because a game that cannot be advertised profitably cannot be scaled regardless of how good the retention looks.

The three questions worth answering, in order:

Does the game retain? If not, nothing else matters.
Can it be advertised? Is there a creative angle that acquires at a viable cost.
Does the economy work? Does the value justify the acquisition cost.

Most soft launches answer the first, assume the second, and estimate the third.

The creative question specifically

The question is not "which creative performed best." It is "is there an angle that works at all, and can we make more of it?"

One winning creative from a batch of five is not a finding you can scale on. It might be a fluke, or a one-off idea with no variations. What you want to leave soft launch with is a direction that produced several viable creatives, because scaling consumes creative faster than anything else.

A soft launch that produces one winner and no repeatable direction has not answered the question.

How many creatives

Enough to cover distinct angles rather than distinct executions.

Five variations of the same idea test execution. Five different ideas test direction. In soft launch you want direction, because you do not yet know what the game's advertising story is.

A practical shape: several clearly different angles, each with two or three executions, so you can tell whether a losing angle failed as an idea or as a particular version of it.

Reading the results without fooling yourself

Sample size first. Soft launch budgets are small. Most creative differences at that spend level are noise, and ranking five creatives on twenty installs each is describing randomness.

IPM before CPI. IPM isolates creative pull from auction cost, which varies a lot across soft-launch geos.

Retention by creative. The single most valuable soft-launch output and the one most often skipped. A creative that installs well and retains badly is a warning about the promise it makes, and scaling it multiplies the problem.

Geo caveats. Soft-launch markets are chosen for cost, not similarity. Creative performance there is directional for your scale markets, not predictive.

What a finished soft launch looks like

You should be able to state:

The angle that works, and why you think it works.
Two or three executions of it that performed consistently.
Which angles failed, and whether as ideas or executions.
Retention by creative, not just in aggregate.
A rough cost expectation, held loosely because geos differ.

If you cannot state the first point, the soft launch is not finished, regardless of how good the retention curve is.

Practical rule

Test angles, not executions, and read retention by creative before you scale anything. A soft launch that proves the game retains but not that it can be advertised has answered the easier half of the question.