Trial and free to paid conversion audit

Most trial conversion numbers answer the wrong question because they treat every non-conversion as a buying decision. A trial whose first charge failed is not a verdict on the product, it is a billing failure, and conflating the two is the central error here. This audit cohorts on trial start, requires a settled payment before counting a conversion, and splits non-conversion into buckets that lead to different fixes.

Steps

1. Resolve plan and trial identity before building any cohort. stripe.list_prices with type: "recurring" and active: true, paged fully, to build a map of price to unit_amount, recurring.interval, recurring.interval_count and recurring.trial_period_days. Pair with stripe.list_products for readable names, falling back to product or price metadata when nickname is null. On Polar, call polar.get_metrics once to confirm which trial slugs this workspace populates. On Dodo, call dodo_payments.list_products and read trial_days off the subscription rather than the plan. Never infer trial length from the plan alone: a subscription can override the price's trial period.

2. Build the cohort by trial start date. stripe.list_subscriptions with status: "all" and created_gte / created_lte bracketing the cohort month plus the full trial length, paged on starting_after until exhausted. Keep trial_start, trial_end, status, cancel_at_period_end, canceled_at, cancellation_details, latest_invoice and items. On Polar use polar.list_subscriptions; on Dodo use dodo_payments.list_subscriptions with created_at_gte / created_at_lte. Separately pull status: "trialing" for the live pipeline and value it as pipeline, not revenue.

3. Require a settled payment for the numerator. A status change is not a conversion. For each cohort member, confirm a real paid first invoice via stripe.list_invoices with subscription: <id> and status: "paid", or a succeeded intent via stripe.list_payment_intents. Run this only against the ambiguous subset, not the whole cohort, to keep page counts sane. On Polar, polar.list_orders with status: ["paid"] serves the same purpose, and polar.get_customer_state resolves single accounts under discussion.

4. Decompose non-conversion into four buckets. Explicit cancel during trial: cancel_at_period_end set while trialing, or cancelled before trial_end. No payment method: Stripe status paused, which it reaches only when a trial ends without a payment method. Attempted and failed first charge: incomplete, or past_due immediately after trial_end, with attempt_count at 1 or more. Hard expired: incomplete_expired, reached when the first invoice goes unpaid for 23 hours, which is terminal. Read decline reasons from dodo_payments.list_payments with status: "failed" and its error_code, from polar.list_payments, or from last_payment_error.decline_code on stripe.list_payment_intents. Buckets two and three are recoverable and are the entire reason this audit exists.

5. Measure time to convert, trial length and abuse. Distribution of days from trial_start to first settled payment, so trial length can be tuned to where the mass actually sits. Compare conversion across the trial lengths genuinely in use, detected from trial_start to trial_end deltas rather than assumed, and treat a mid-period trial length change as a series break. Screen for repeat trials on the same email domain or card fingerprint and for trials that reached trial_end with zero product usage: use stripe.search with resource: "customers" or resource: "subscriptions" for metadata and name matching, and posthog.query or postgres.query (after postgres.get_schema) for the activation side. Keep the abuse bucket out of the honest conversion rate rather than deleting it.

6. Report. One row per trial start cohort with columns: cohort month, trials started, converted with settled payment, conversion rate, cancelled in trial, ended with no payment method, failed first charge, hard expired, abuse-flagged, median days to convert, converted MRR, recoverable MRR in the failed and no-payment-method buckets. Then deliver one judgement: state the conversion rate against the model-matched Growth Unhinged benchmark (8% median overall, 30% when a card is required up front), and name how many points of the shortfall are a billing failure rather than a buying decision.

Gotchas

• A failed first charge is not a non-conversion. This is the trap the whole playbook exists for. Stripe's revenue recovery analytics explicitly exclude the first invoice payment following a trial, so these failures appear in neither the conversion funnel nor the dunning dashboard. Report them as their own line and price them, because they convert with an email, not with a product change.
• Trial subscriptions emit $0.00 invoices. Counting invoices to count conversions counts every trial twice, once at zero and once at the real price. Count distinct subscriptions with a settled non-zero payment instead.
• The 23 hour fuse on incomplete_expired. Stripe moves a subscription to incomplete_expired if the first invoice is not paid within 23 hours, and the status is terminal. A cohort pulled the next morning shows these as permanently lost even though an intervention at hour 20 would have worked, so always report the age of the window you pulled.
• paused means no payment method, not customer paused. Stripe enters paused only when a trial ends without a payment method. Reading it as a voluntary pause hides a pure collections problem. Separately, pause_collection does not change status at all, so those accounts still read active.
• Cohort on trial start, and watch the timezone cut. Conversions lag the cohort by the trial length, so a conversion-dated numerator over a start-dated denominator mechanically breaks the newest month. Stripe filters are UTC epoch seconds, Polar takes an IANA timezone, Dodo takes ISO timestamps, so a trial starting at 23:00 UTC on the last day of the month lands in a different cohort for a US team.
• Minor units, interval_count and multi-currency. Amounts are integer minor units, so divide by 100 once and not at all for zero-decimal currencies such as JPY. Monthly normalisation is unit_amount divided by 100 divided by the months in the interval, and forgetting interval_count triples a quarterly plan. Never sum trial MRR across currencies without stating the FX source.
• Test clocks and status spellings. Trial logic is the most test-clocked part of any billing integration, so filter on livemode and drop non-null test_clock. Dodo spells the cancelled state with two Ls while Stripe spells it canceled, and a status comparison written for one silently returns zero rows against the other. On Polar, prefer the pre-computed trial_monthly_recurring_revenue and checkouts_conversion slugs over recomputing from raw orders; on a Stripe-only or Dodo-only account state plainly that there is no checkout-step rate available and the measurement starts at the trial subscription.