Why Revenue Per Exposed User Beats Conversion Rate for Subscription Apps

Conversion rate counts the percentage of users who start a trial or make a purchase, treating all conversions as equally valuable. In subscription apps, this assumption breaks down immediately. A user who converts to a trial then cancels before paid billing contributes nothing to revenue but inflates conversion rate the same as a user who remains subscribed for years.

The metric becomes even more misleading when experiments affect user quality. Aggressive tactics like heavily discounted trials or exaggerated benefit claims can dramatically increase conversion rates while attracting users with low intent to pay. These users convert at high rates but churn quickly, often during trial periods or immediately after the first charge. The conversion rate metric registers success while revenue actually declines.

Revenue per exposed user divides total revenue generated by all users assigned to a variant, regardless of whether they converted. This approach measures what actually matters: the average financial value of exposing a user to your onboarding experience. A variant that converts fewer users but attracts higher-quality subscribers will correctly show superior performance.

The metric naturally incorporates timing effects that conversion rate ignores. Users who convert but refund within days contribute negatively. Users who subscribe at lower rates but remain subscribed for months generate compounding value. Trial-to-paid conversion, first billing success, and early retention all flow into a single number that reflects true business impact rather than a proxy metric.

Subscription revenue accrues over time, creating a measurement challenge for experiments. A 7-day attribution window might show Variant A winning because it drives more immediate trial starts, while a 60-day window reveals Variant B generates more revenue through better trial-to-paid conversion and first-month retention. Short windows bias toward quick conversions rather than quality.

The right attribution window depends on your subscription model and billing cycle. Weekly subscriptions show meaningful retention signals within 30 days, while annual subscriptions need longer windows to capture trial conversion and initial renewal behavior. Most subscription apps benefit from 30 to 90-day windows that capture trial conversion, first billing, and early churn patterns without waiting so long that learning cycles become impractical.

Conversion rate remains useful as a diagnostic metric when revenue per exposed user changes. If revenue increases while conversion rate drops, you've successfully filtered out low-quality users. If both metrics move together, the experiment affected user volume without changing quality. Breaking down the funnel helps identify which specific step drove the revenue change.

Track conversion rate at multiple stages: trial start rate, trial-to-paid rate, and first billing success rate. These component metrics help diagnose why revenue per exposed user changed and guide iteration. A variant might reduce trial starts but dramatically improve trial-to-paid conversion, resulting in net positive revenue impact that conversion rate alone would miss entirely.

Accurate revenue measurement requires matching purchase events to experiment assignments at the user level. Revenue systems like RevenueCat track purchases by App User ID, which must be connected to the user identifier in your experiment platform. Without proper attribution, revenue from variant B might be credited to variant A if user identifiers don't align correctly between systems.

Tools like Spinosis handle this attribution by matching RevenueCat purchases to experiment user IDs, ensuring revenue is credited to the correct variant. The system tracks which variant each user saw and attributes their complete purchase history accordingly, including trials, conversions, renewals, and refunds. This infrastructure enables revenue per exposed user to serve as your primary decision metric rather than relying on conversion rate proxies.