Mobile funnel experimentation: what to test beyond the paywall
A catalog of in-app experiments for onboarding, permissions, upsells, and win-backs that compound revenue when you are not waiting on App Store releases.
Paywall A/B testing gets the headlines. Superwall, Adapty, and RevenueCat have trained the market to think "experimentation" means headline price tests on screen five.
That is necessary but narrow. The subscription apps pulling ahead in 2026 run experiments across the full funnel: first open, permission moments, activation checkpoints, upsells, cancellation, and win-back. They ship variants over the air because waiting on App Store review for every hypothesis is how funnels go stale.
This post is a catalog of what to test beyond the paywall, how to prioritize when you cannot run fifty tests at once, and what tooling you need so experiments compound instead of conflicting.
Map the funnel: install to retain
Before ideating tests, draw stages. Experiments attach to stages, not random screens.
Install → First open → Onboarding → Activation (aha) → Monetization → Habit → Expand → Retain / Win-back| Stage | Question the user is answering | Example experiments |
|---|---|---|
| First open | "Is this for me?" | Value prop order, social proof placement |
| Onboarding | "How does this work?" | Flow length, interactive vs passive |
| Activation | "Did I get value?" | Checklist vs free exploration |
| Monetization | "Is it worth paying?" | Paywall timing, trial length messaging |
| Habit | "Will I come back?" | Notification pre-prompt, streak UI |
| Expand | "Should I pay more?" | Upsell timing, tier presentation |
| Retain | "Should I stay?" | Cancel survey, save offers |
| Win-back | "Should I return?" | Reason-based offers, what's new |
Paywall tests live in monetization. Most teams under-test activation and retain because those screens are scattered across code owners.
Experiment ideas by stage
Onboarding and first session
| Test | Hypothesis | Primary metric |
|---|---|---|
| 3-screen vs 7-screen flow | Shorter completes more | Onboarding completion |
| Value-first vs account-first | Value before signup lifts activation | Signup rate, D1 retention |
| Progress bar vs none | Progress reduces anxiety | Step drop-off |
| Personalized headline by channel | Paid social needs different hook | Trial start by attribution |
| Interactive demo vs static slides | Doing beats reading | Activation event |
Permissions (ATT, push, location)
| Test | Hypothesis | Primary metric |
|---|---|---|
| Pre-prompt copy variants | Clear benefit lifts opt-in | ATT authorization rate |
| Permission after aha vs launch | Timing beats early ask | Opt-in × retention |
| Branch on deny | Alternate path keeps engagement | D7 retention (denied cohort) |
| Skip vs required push | Forced push hurts completion | Completion vs opt-in tradeoff |
Permission flows are underserved in competitor content and high leverage for paid acquisition quality.
Monetization (including paywall)
| Test | Hypothesis | Primary metric |
|---|---|---|
| Paywall on day 0 vs after activation | Later paywall lifts paid % | Trial-to-paid |
| Annual default vs monthly | Annual default lifts LTV | ARPU, annual mix |
| Soft paywall vs hard | Soft lifts completion with revenue tradeoff | Revenue per install |
| Social proof on paywall | Testimonials lift convert | Purchase rate |
| Onboarding + paywall as one variant | Siloed tests miss interaction effects | Flow-level conversion |
Post-convert: upsell and expansion
| Test | Hypothesis | Primary metric |
|---|---|---|
| Upsell at day 3 vs day 14 | Early upsell annoys or converts | Tier upgrade rate |
| Feature gate vs explicit upsell screen | Contextual gates win | Upgrade revenue |
| Trial reminder before charge | Reminder reduces involuntary churn | Renewal rate |
Lifecycle: cancel, feedback, win-back
| Test | Hypothesis | Primary metric |
|---|---|---|
| Cancel survey length | 5 reasons beat 12 | Survey completion |
| Save offer by reason | Personalized save beats generic | Save rate |
| Win-back on return vs push-only | In-app beats email alone | Resubscribe rate |
| NPS after aha moment | Timing lifts response | Response rate |
Stories-style and full-screen sequential flows
| Test | Hypothesis | Primary metric |
|---|---|---|
| Stories vs card onboarding | Stories lift engagement for visual apps | Completion |
| Tap-to-advance vs auto-advance | Control lifts comprehension | Activation |
Breadth is intentional. Your backlog should span stages, not ten paywall headline tests.
Velocity: why OTA changes prioritization
When each test costs a release cycle, teams pick "safe" incremental changes. When publish is instant, bold variants become affordable.
| Release-bound culture | OTA flow culture |
|---|---|
| Test button color | Test remove half the screens |
| One experiment per quarter | Continuous backlog |
| Debates in eng planning | Debates in growth weekly |
| Winner ships in v2.4 | Winner ships Tuesday |
OTA does not remove the need for statistics. It removes the need to wait six weeks to start collecting data.
Remote native flows (Rheo, not JS bundle OTA) let product managers publish screen structure and branches without merging React Native navigation changes. Engineering keeps SDK integration stable; growth owns the flow graph.
Prioritize with ICE
You cannot run everything. Score ideas:
- Impact: expected lift on revenue or retention (1-10)
- Confidence: how sure you are from data or analogs (1-10)
- Ease: ship time with your tooling (1-10)
ICE = (Impact + Confidence + Ease) / 3. Sort descending. Re-score after each result.
Heuristics that beat ICE alone:
- Fix largest step drop-off before cosmetic tests
- Run flow-level tests when onboarding and paywall tell conflicting stories
- Sequence permission tests after value is clear
- Hold 10-20% holdout on big swings to measure incrementality
Anti-patterns
Paywall-only myopia. Lift on paywall convert means nothing if onboarding shed 80% before users arrived.
Overlapping experiments. Two tests on the same channel without mutual exclusion contaminate results. Use experiment layers or sequential runs.
Underpowered calls. 200 users per variant is not an A/B test. It is a coin flip. Use longer runs or bolder variants when traffic is low.
Winner forever. Seasonality and acquisition mix shift. Re-test winners quarterly.
WebView funnels on native apps. Fast to ship, costly in conversion on critical paths. Native rendering for monetization and cancel flows is worth the integration.
Ignoring involuntary churn. Win-back creative does not fix broken billing retry.
Tooling requirements
To run funnel-wide experimentation you need:
| Requirement | Why |
|---|---|
| Visual flow editor | PMs ship without eng queue |
| OTA publish | Same-day variant start |
| Step-level analytics | Find screen 4, not just "funnel" |
| Experiment assignment | Clean A/B/C with holdouts |
| Trait targeting | Channel, locale, plan, cancel reason |
| Native rendering | Conversion-critical trust |
| Billing integration | RevenueCat etc. for purchase truth |
| MMP hooks | Attribute installs to funnel variants |
Point solutions cover slices (paywall-only, surveys-only). A flow platform covers the graph: onboarding through win-back on one canvas with one analytics model.
Rheo platform map (how pieces connect)
| Rheo capability | Funnel stage |
|---|---|
| Channels | Entry points (first launch, settings, return) |
| Canvas | Screen design and branches |
| Targeting | Traits and audiences |
| Experiments | Variant assignment and holdouts |
| Analytics | Per-step completion and conversion |
| Publish | OTA without App Store review |
| Native SDK | RN (iOS/Android), more platforms on roadmap |
RevenueCat remains source of truth for entitlements. MMPs remain source of truth for acquisition. Rheo is the experimentation layer for everything between install and subscription state change.
Case pattern: permission branch after aha
A meditation app ran ATT pre-prompt tests only on paywall results for months. Funnel data showed 40% drop on screen 2 (permission) before anyone saw pricing.
They shipped three remote variants over two weeks without releases:
- A: ATT on screen 2 (status quo)
- B: ATT after first completed session (aha)
- C: ATT after aha with reason-specific branch on deny
Variant C lifted ATT opt-in by nine points and D7 retention by four points among denied users because deny-branch skipped broken attribution-dependent tips and showed offline content instead.
Paywall convert barely moved. Revenue per install rose because more users reached the paywall. That is the case for funnel-wide experimentation in one sentence.
Collaborating with engineering
Remote flows do not remove engineering from the loop. They change the contract:
| Engineering owns | Growth owns |
|---|---|
| SDK version, init, auth | Screen copy and order |
| Purchase and entitlement callbacks | Branches and experiments |
| Channel trigger points | Publish timing |
| Crash and performance SLAs | Hypothesis backlog |
Weekly thirty-minute sync beats async ticket queues. Review live experiments, upcoming native releases that need new SDK features, and trait definitions both sides agree on.
Localization and global funnels
Remote flows support localized copy per locale without duplicating app binaries. Subscription apps with EU, LATAM, and US growth often test:
- Currency and price presentation copy (offerings still from RevenueCat)
- GDPR consent screen placement
- Shorter flows in markets with higher install-to-trial drop on long onboarding
Publish locale variants from the same canvas using targeting rules rather than maintaining four git branches of onboarding screens.
Stories and sequential UI: full-screen tap-through flows behave differently by category. Education apps may need longer stories; utility apps need faster time-to-aha. Tag experiments by app_category in your learnings doc so patterns transfer across portfolio companies if you operate more than one app.
Build your backlog this week
- Export last 30 days step funnel from analytics (or instrument if missing)
- Mark the single worst drop-off step
- Pick three tests from the tables above tied to that stage
- ICE score them
- Ship the top one as a remote flow variant
- Set 7-14 day run with predefined success metric
The apps winning on conversion in 2026 are not smarter. They run more learning cycles per month because friction to ship is low.
Start for free. Map your flows on the canvas and test beyond the paywall.