Designing payment recovery for subscriptions

Why this matters

For a one-off ecommerce purchase, a failed payment is a transaction. For a subscription, a failed payment is a moment of churn risk.

Most LATAM subscription businesses (streaming, telco, SaaS, fitness, education, premium content) lose 5-15% of MRR every month to payment failures that have nothing to do with customer intent. Cards expire. Issuers churn. Funds run short on the 28th. The customer is still a customer; the payment instrument has simply failed under them. The recovery question is not “how do we close this sale”, it is “how do we keep this relationship intact without forcing a re-decision.”

The honest version of subscription recovery is more delicate than vendor narratives suggest. Aggressive dunning recovers more MRR in the short term and triggers more churn in the medium term. The operator job is to find the line, hold it, and move it carefully when the data tells you to.

What goes wrong

The most common failure modes in subscription recovery are not technical. They are framing mistakes:

• Borrowing cart-recovery copy. “Complete your purchase” reads as cold and transactional to someone who has been a customer for two years. The voice has to acknowledge the relationship.
• Retrying too aggressively. Five retries in 24 hours generates issuer noise, triggers fraud flags, and sometimes blocks the card outright. Smart retries are spaced, classified by decline reason, and capped.
• Treating all declines as equal. Insufficient funds on the 28th has a 70%+ recovery curve if you wait until payday. A “do not honor” decline at any time is usually permanent. Same dunning ladder for both = wasted attempts and burned trust.
• No account-updater coverage. Networks offer account updater services that quietly refresh expired card details. Subscription operators not using them are leaving 2-5% of MRR on the floor every month.
• No alternative-method offer. If the card has churned and won’t recover, the only path forward is a different method: Bre-B, PSE, Pix, SPEI, a wallet. A dunning sequence that only ever retries the same card is a dunning sequence designed to lose customers.
• Confusing voluntary and involuntary churn. When customers genuinely want to leave, recovery is the wrong tool. When the payment failed and the customer didn’t notice, recovery is everything. The dunning ladder has to give the customer a clear exit, or you cross the line from recovery into hostage-taking.
• Compliance drift on regulated verticals. Telco and financial-services subscriptions have rules about how often you can attempt charges and what notifications are required. The recovery system has to honour them.

Operating model

Subscription recovery is best operated as a coordinated ladder with three concurrent disciplines.

The trust math is the discipline most often skipped. It is also the one that decides whether your recovery program compounds or quietly erodes the base.

Recommended approach

A working subscription recovery operating model looks roughly like this:

Use network tokens and account updater as table stakes. Network tokens reduce decline rates on lifecycle events. Account updaters reduce decline rates on expirations. Neither is optional at scale.

Time the first retry by decline reason, not by clock. Insufficient funds: retry on the 1st or 15th. Issuer timeout: retry in 1 hour. Expired card: do not retry; route to account-updater and outreach.

Make the dunning ladder a path, not a barrage. Day 0: in-app banner. Day 1: email. Day 3: WhatsApp. Day 7: pause notification. Day 14: cancellation. Each step explicitly offers an alternative method.

Offer alternative methods, especially in Bre-B / Pix / SPEI markets. A subscription customer in Colombia whose card has churned can often be saved by a one-tap Bre-B mandate. The same is true in Brazil with Pix recurring and in Mexico with SPEI direct debit.

Make cancellation honest. Hiding cancellation under five clicks recovers some MRR and destroys customer trust. The cancellation flow should be findable and respectful. Operators who do this report higher reactivation rates downstream.

Watch involuntary-churn-driven complaints. App store reviews, support tickets tagged “billing surprise”: these are leading indicators that your dunning is too aggressive long before the cohort data catches it.

Re-tune quarterly. The right retry schedule for January is not the right schedule for July. Holiday cycles, payroll cycles, and macro events all shift the failure distribution.

Example architecture

A reference architecture for subscription recovery at scale:

Recurring charge attempt (PSP)
    → approved → ledger entry → continue
    → declined → decline classifier
        ↓
    Reason class:
        - insufficient funds → schedule retry near payday
        - expired card → account-updater lookup
            → updated token found → retry with new token
            → no update → outreach loop
        - issuer timeout → retry in 1h
        - hard decline → outreach loop (no retry)
    ↓
Outreach loop (dunning ladder)
    → in-app banner (day 0)
    → email (day 1)
    → WhatsApp (day 3)
    → pause notification (day 7)
    → cancellation (day 14)
    ↓
Each step: offers alternative method (Bre-B / PSE / Pix / SPEI / wallet)
    ↓
Recovery outcome → recovered MRR + involuntary-churn cohort

Two pieces drive most of the value: the decline classifier with reason-aware retry scheduling (so retries are smart, not lazy), and the alternative-method offer at every dunning step (so a churned card doesn’t have to end the subscription).

Metrics to watch

Involuntary churn is the lagging metric and the one that matters most over a year. Recovery rate is the day-to-day operational metric. Account-updater hit rate and alternative-method save rate together tell you whether your stack is doing the unglamorous work that holds the base together.

If recovery rate is high but six-month retention of recovered customers is low, the dunning ladder is winning the battle and losing the relationship. That number is worth watching more carefully than any current-month rollup.

What to read next