The Science of Dunning: How We Recovered $4.2M in Failed Subscription Payments
Machine learning-driven retry logic that knows when to push and when to pause
Key Takeaways
- ML-optimized retry timing recovers 73% of failed transactions
- Personalized retry windows based on historical payment patterns
- Pre-dunning card updates reduced failures by 23%
- $4.2M recovered for merchants in Q4 2025 alone
Subscription businesses lose an average of 9% of monthly recurring revenue to involuntary churn—payments that fail due to expired cards, insufficient funds, or temporary issuer holds. Smart dunning recovers the majority of these transactions.
The Prediction Model
Our dunning engine starts before the payment even fails. We analyze 127 signals to predict which upcoming renewals are at risk: card expiration dates, historical decline patterns, issuer decline rates, and time since last successful transaction.
We reach out to update payment methods 48 hours before predicted failures. This single intervention reduced our clients' involuntary churn by 23%.
Intelligent Retry Timing
When a payment fails, the retry strategy matters enormously. Retry too quickly and you'll hit the same temporary hold. Retry too slowly and the customer may churn. Our model learns optimal retry windows per card type, issuer, and failure reason.
- Insufficient funds: Retry post-payroll (optimized per customer's historical deposit patterns)
- Temporary holds: 4-hour minimum delay before first retry
- Expired cards: Immediate outreach, no retry until updated
- Issuer downtime: Aggressive retry with exponential backoff
Written by Nina Patel
Lead Data Scientist
