ASOhack
Back to Blog
Methodology

A/B Test or Just Ship? A Framework for App Experiments

Not every change needs an A/B test. A framework for deciding when to test rigorously, when to ship and watch, and when to trust your gut.

ASOhack TeamMay 19, 20266 min read

A/B testing is a powerful tool that's also routinely misapplied. Indie devs either over-test (test every button color, never reach significance, ship slowly) or under-test (ship blindly and miss obvious improvements).

This is the framework for deciding when to A/B test, when to ship-and-watch, and when to trust your gut.

The three decision modes

Mode 1: A/B test rigorously

  • Change has potentially big impact (10%+ on a critical metric).
  • You have enough volume for significance (1k+ samples per variant).
  • You can isolate the change (one variable).
  • The cost of being wrong is high.

Mode 2: Ship and watch

  • Change is plausible improvement based on data / first principles.
  • Risk is low (easy to revert).
  • Volume is too small for proper A/B testing.
  • You can monitor post-ship metrics.

Mode 3: Just ship (no test)

  • Change is obvious improvement (a typo fix, a clear UI bug).
  • Cost of testing exceeds value.
  • No interesting metric to test against.

Most indie devs default to Mode 2 / Mode 3, which is usually correct.

When to A/B test (Mode 1)

Conditions

  1. Volume: ≥1,000 conversions per variant for meaningful significance.
  2. Time: ≥14 days to capture weekly patterns.
  3. Effect size: hypothesized lift of 10%+ on the metric.
  4. Isolated variable: only one thing changes between variants.
  5. Measurable: clear before/after metric.

High-value A/B tests for mobile apps

  • Paywall variant (hard vs soft).
  • Trial length (3 vs 7 vs 14 days).
  • Pricing tier ($4.99 vs $9.99).
  • First screenshot (lifestyle vs UI vs outcome).
  • Onboarding length (3 steps vs 7 steps).
  • App icon variant (background color, character).
  • Subtitle phrasing (outcome vs feature).

These are tests worth running rigorously.

When to ship and watch (Mode 2)

Conditions

  1. Plausible improvement based on user feedback or data.
  2. Low risk — easy to revert if it breaks.
  3. Volume too small for proper A/B testing.
  4. Time-sensitive — can't wait 14 days.

Examples

  • Adding a new feature based on user requests.
  • Improving error messaging.
  • Refactoring onboarding flow based on drop-off data.
  • Updating copy based on feedback patterns.

Methodology

  • Ship the change.
  • Monitor key metrics for 7-14 days.
  • Compare to pre-ship baseline (same period last week/month).
  • If metric improves: keep.
  • If metric drops: revert or iterate.

This is most indie product work. A/B testing every change would slow you to a crawl.

When to just ship (Mode 3)

Conditions

  • Pure quality improvement (bug fix, typo).
  • Following platform best practices (compliance, accessibility).
  • Internal-facing change (analytics events, code refactor).
  • Obvious win with no plausible downside.

Examples

  • Fix a crash.
  • Add accessibility labels.
  • Improve performance.
  • Fix typos.
  • Tag analytics events.

No need to test. Just ship.

The trap of over-testing

Over-testing burns time and slows progress:

  • 2 weeks per test × 10 tests/year = 20 weeks/year on testing.
  • Most tests show "no significant difference" — feels like wasted time.
  • Even meaningful tests need samples; indie apps often lack samples.

The right framing: A/B testing is for high-stakes decisions where you can't afford to be wrong.

The trap of under-testing

Under-testing produces compounding mistakes:

  • Ship a paywall variant that loses 30% conversion.
  • Doesn't notice for 2 months.
  • Cost: $30k+ in lost revenue.

For high-stakes decisions (pricing, paywall placement), test rigorously.

Sample size math

Quick estimator for A/B tests:

Required sample size per variant ≈ 16 × (1 - p) / p × (1/MDE)²
where:
  p = baseline conversion rate
  MDE = minimum detectable effect (e.g., 0.10 for 10%)

For 5% baseline conversion and 10% relative MDE:

≈ 16 × 0.95 / 0.05 × 100 = 30,400 per variant

For most indie apps, this means tests on conversion need months of runtime, which is impractical. Test bigger effects (20%+ MDE), accept lower confidence, or use Mode 2 instead.

Frequentist vs Bayesian

Some A/B testing tools use Bayesian methods (faster to conclude) vs frequentist (more conservative). Both are valid; both can be misinterpreted.

For indie purposes:

  • Use whichever your A/B testing tool offers by default.
  • Be skeptical of "winning" variants with <100 conversions per group.
  • Trust 14+ day tests more than 3-day tests.

What to do at indie scale

Most indie apps have insufficient volume for proper A/B testing. Strategies:

Strategy 1: Test the biggest decisions only

Test paywall + pricing + first screenshot. Skip everything else.

Strategy 2: Pre/post comparison

Ship a change, compare 4 weeks before vs 4 weeks after. Not rigorous but better than nothing.

Strategy 3: Direction testing

Don't try to detect 5% lifts. Test changes you expect to move metrics 30%+. Smaller changes either ship-and-watch or skip testing.

Strategy 4: Combine signals

A/B test + qualitative feedback + analytics drill-down combined gives more signal than any single source.

A/B testing tools

Built into stores (free)

  • App Store Connect: Product Page Optimization.
  • Google Play Console: Store Listing Experiments.

Use these for icon / screenshot / video tests.

In-app testing

  • Firebase A/B Testing (free, integrated with Remote Config).
  • Statsig (free tier).
  • PostHog (open source, included).
  • RevenueCat / Adapty / Apphud paywall A/B testing (built-in).

Don't pay for testing tools at indie scale

Built-in tools are sufficient until you have substantial volume.

Common mistakes

  • Testing too many variables at once. Can't isolate the cause.
  • Stopping tests early. Below required sample = unreliable conclusion.
  • Calling weak winners. A 3% lift with 200 samples isn't a real win.
  • Testing trivial changes. Burns time; learns nothing.
  • Never testing big decisions. Misses easy wins.
  • Over-trusting frequentist p-values. Multiple-test problem.
  • Ignoring qualitative data. Numbers + user feedback together.

A pragmatic indie cadence

For most indie devs, a working rhythm:

  • Test rigorously: 4-6 times per year (the big decisions).
  • Ship and watch: 20-40 times per year (the normal product work).
  • Just ship: continuously (the obvious wins).

Run an audit

A/B testing compounds with good baseline data. Run free ASO audit regularly to see the absolute landscape — testing only matters if your baseline is healthy.

Try the tools

Ready to Optimize Your App Store Listing?

Try our free ASO tools — no signup required.