Industries, Dating apps
Decisions for the highest-stakes paywall in mobile.
Subscription tiers, profile-boost pricing, and matching-policy personalisation, with the causal proof your investors will ask for.
Dating apps run the highest-LTV paywalls in consumer mobile, with the highest variance in per-user value and the most volatile weekly cohorts. Payment-friction sensitivity is extreme; a small mis-step on pricing shifts the entire funnel. Metapolicy gives growth teams in this category the substrate to run aggressive personalisation with defensible holdout-validated lift.
What agents handle
Workflow
Tier-upgrade paywall variant
Before
Same monthly / annual / lifetime grid for everyone
After
LinUCB picks per user from frozen-context features (engagement, region, device)
Result
+17% conversion to annual on the same impression budget
Workflow
Boost / super-like pricing personalisation
Before
Fixed pricing tile across all users
After
Bandit-priced per propensity-to-purchase cohort
Result
+11% revenue per boost-eligible session
Workflow
Push cadence per engagement decile
Before
Same daily-message cadence for active and dormant alike
After
Per-decile bandit picks frequency and channel
Result
+8% D14 retention, −22% unsubscribe rate
Workflow
Day-1 onboarding step skip
Before
Fixed 8-step profile-completion flow
After
Bandit picks which steps to skip per user platform and entry source
Result
+14% first-match-within-24-hours
The challenges we solve
- Extreme variance in per-user value, the average tells you almost nothing
- Payment-friction sensitivity: every additional tap drops conversion measurably
- Weekly-cohort instability around marketing pushes and competitor releases
- Investor diligence asks for holdout-validated lift, not dashboard A/B claims
- Trust signals matter, users notice when pricing feels arbitrary
Contextual bandits handle the cohort drift
When the weekly user mix shifts (marketing push, competitor launch), the bandit rebalances. The propensity log captures that drift so OPE remains valid.
Bootstrap CIs reflect the actual variance
Studentised bootstrap with B=2000, the confidence interval you see is the variance your data actually has, not a t-test assuming normality.
Outcome pricing aligns our incentives with your retention
We only earn on lift that survives doubly-robust evaluation with an ESS guardrail. The math itself caps over-claiming. Your CFO sleeps; your investor diligence finishes.
Worked audit example
What a doubly-robust re-analysis surfaced for Velvet.
Series-B dating app · ~$26M ARR · 318,940 boost-eligible users · case ID VELVET-2026Q2-AUDIT-005
What their team reported
Boost price test: $7.99 vs $4.99. Conversion +1pt to 7.8%; revenue per eligible user $0.54 → $0.39. Team shipped $4.99 globally; the revenue drop was attributed to "low elasticity."
What our re-analysis found
Doubly-robust per engagement decile: light users gained on both conv. and rev. at $4.99, but heavy-engagement deciles cost $0.42/user, they would have paid $7.99. Across ~87,000 users in those deciles, ~$330K of foregone revenue in one quarter.
Recommendation · projected annualised impact
Bandit-personalise boost pricing on (engagement decile × days-since-last-match × current online-cohort density). +14% boost revenue without changing the offer.
+$1.1M / yr boost revenueSame shape we'll send back on your last A/B test, in three business days.
Read the full audit PDF →Audit your last paywall test — free.
One CSV, one experiment config. Same-shape readout back in three business days.