Smart Retries

Smart retries is an ML-driven payment recovery system that automatically selects the optimal time and routing path to retry a failed payment, adapting to individual payment history, user profiles, transaction context, and global network patterns — rather than following a fixed schedule. It is primarily applied to subscription/recurring billing but also to one-time card-not-present ecommerce failures.

---

How it works

Smart retries is an automated, silent subset of a broader Dunning strategy: it handles the automatic retry aspect without customer involvement, while dunning management encompasses both automated retries and customer-facing communication (email, SMS) to encourage payment method updates. (GR4VY, 2026-06-02)

Between 70% and 90% of all failed card-not-present payments are soft declines — where the card itself is fine but something temporary went wrong (network timeout, insufficient funds at that moment, issuer velocity limit, conservative fraud rule) — meaning the transaction will likely succeed on a second attempt made at the right time and through the right path. (as-of 2026-06-02) (GR4VY, 2026-06-02)

Signal categories

The ML models underlying smart retry systems typically evaluate four to five factor categories:

• Customer behaviour / profile — subscription tenure, payment history, location, device fingerprint, local payday cycles
• Timing signals — local time zones, weekday vs weekend, salary days, local holidays
• Payment attributes — payment method type, card type, Decline Codes|decline reason, amount, currency
• Business / seasonality attributes — product category, geography, industry, billing mix
• Proven patterns — what has historically worked for the merchant, its geography, similar companies, and acquiring banks globally

(Solidgate, 2026-06-18; Stripe Engineering Blog, 2024-01-23)

Relationship with Adaptive Acceptance

Adaptive Acceptance and Smart Retries address different failure points:

• Adaptive Acceptance — retries false declines in real-time during the original transaction attempt, before the customer leaves checkout
• Smart Retries — schedules retries of genuinely failed recurring/subscription payments over subsequent days or weeks, silently in the background

Both are separate products at Stripe and are often deployed together. (Stripe Engineering Blog, 2024-01-23; Stripe Blog, 2025-02-12)

---

Card network constraints

Visa rules cap merchant-initiated retries at 15 attempts within a 30-day rolling period for the same transaction, and prohibit retries entirely for certain Decline Codes (e.g. code 04 "pick up card", 41 "lost card"). Excessive retries can trigger Visa's Acquirer Monitoring Programme with fines. (as-of 2026-06-02) (GR4VY, 2026-06-02)

In practice, 3–5 retry attempts spread over 10–14 days is the optimal range for most merchants; the network maximum is rarely the operationally optimal number. (GR4VY, 2026-06-02)

90% of recovered transactions occur within the first 10 days of a failed payment, meaning the quality of decisions made early in the retry window determines the majority of recovery outcomes. (as-of 2026-04-30) (Recurly, 2026-04-30)

---

Benchmarks

Best-in-class blended recovery rates (across all decline types) hover between 45–60%; AI-powered solutions can push recovery above 70% specifically for soft declines. (as-of 2025-12-03)

Companies switching from batch-based to intelligent individualised retry strategies typically see a 20–50% increase in recovered revenue. (as-of 2025-07-20)

---

PSP implementations

Stripe Smart Retries

Stripe's Smart Retries model analyses over 500 attributes per failed payment using an Auto-ML ensemble architecture (upgraded from XGBoost in 2023) that merges predictions from multiple base models; multimodal text and numerical data are converted into embeddings using Sentence Transformers. (as-of 2024-01-23)

Stripe reports its Smart Retries recovers an average of 57% of originally failed recurring payments, and $9 in revenue for every $1 customers spend on Billing. (as-of 2024-01-23)

Stripe's documented default Smart Retry setting is 8 attempts within 2 weeks for subscription billing; the window stops after 7 gateway declines, 20 total attempts, or 60 days from invoice creation. (as-of 2026 — live docs) (Stripe Docs)

Stripe Smart Retries operates exclusively within the Stripe ecosystem; it cannot route retries to alternate processors, which caps recovery potential compared to multi-gateway solutions. (as-of 2025-07-20)

Merchants can configure per-product retry strategies (e.g. separate B2B vs. B2C policies), the drop-dead retry day, and maximum retry attempts. (Stripe Engineering Blog, 2024-01-23)

Case studies (Stripe):

• Deliveroo recovered more than £100 million in one year using Smart Retries in coordination with Card Account Updater and Adaptive Acceptance (as-of 2024-01-23)
• Retool recovered more than $600,000 with Smart Retries in one year (as-of 2024-01-23)

(Stripe Engineering Blog, 2024-01-23)

Subscriptions recovered by Stripe tools continue on average for seven more months — equivalent to acquiring a new subscriber. (as-of 2024-01-23) (Stripe Engineering Blog, 2024-01-23)

Adyen Auto Rescue

Adyen's Auto Rescue product uses a contextual multi-armed bandit approach (epsilon-greedy with Thompson Sampling-inspired tree bootstrapping, implemented via a random forest classifier) to schedule retry timing across the retry window for failed recurring payments. It incorporates payday information based on customer location for insufficient-funds declines. The product has been expanded from card payments to SEPA payments. (as-of 2020-12-03 for ML methodology)

Initial experiments showed Auto Rescue converted approximately 6% more orders than the baseline strategy of retrying every four days. Pinterest saw a 4% recovery rate improvement, with expectation of reaching 10% over time. (as-of 2020)

Recurly Intelligent Retries

Recurly's Intelligent Retries uses ML based on hundreds of millions of data points. Optimised retry strategies improved failed payment recovery rates from approximately 53% to 71% compared to single-merchant retry logic. The retry window caps at 7 transaction declines from the gateway, 20 total transaction attempts, or 60 days since invoice creation. (as-of 2026-04-30) (Recurly, 2026-04-30)

Case studies (Recurly):

• Enterprise big-box retailer membership subscription: estimated $1.9M in incremental recovered revenue over two months (~$11.6M annualised) (as-of 2026-04-30)
• Global on-demand food delivery subscription: $3.6M projected over two months (~$21.6M annualised) (as-of 2026-04-30)

(Recurly, 2026-04-30)

Sean Douglas of PayPal stated in Recurly's 2026 State of Subscriptions report: "We tend to see plus 20% retention simply by doing the right Payment Orchestration|payment orchestration and enabling the consumer to have a frictionless experience without them even knowing it." (Recurly, State of Subscriptions 2026)

Chargebee Smart Retry / Revive

Chargebee's Smart Retry classifies gateway error codes into hard and soft declines and retries up to 12 times based on error type and transaction patterns, without requiring merchant configuration. Chargebee's Revive product reads historical funding patterns and schedules retries for when money is typically available (paydays, salary cycles, start of month). Chargebee now exposes a retry_engine field on invoices to track whether Chargebee, FlexPay, or GoCardless Success+ handled dunning for each invoice — indicating a third-party retry engine marketplace is emerging. (Chargebee)

Case studies (Chargebee):

• Zenchef recovered 60% of formerly unpaid accounts using Chargebee Smart Dunning (as-of unknown)

Braintree

Braintree does not natively handle dunning management, smart retries, revenue recovery from failed payments, or detailed subscription analytics — these require third-party tooling layered on top. (as-of 2025-07-20)

---

Cross-PSP and multi-gateway retries

Cross-PSP retry routing (sending a failed transaction to a different acquirer or processor than the original attempt) typically adds 2–5 percentage points of recovery beyond same-PSP retries, and is only possible inside a multi-PSP Payment Orchestration setup. Multi-gateway routing immediately recovers approximately 8% of failures by trying alternate processors, independent of AI timing optimisation. (as-of 2026-06-02) (GR4VY, 2026-06-02)

---

Contradictions

---

Gaps and open questions

• SCA/PSD2 interaction: How EU Strong Customer Authentication re-authentication requirements interact with ML retry scheduling for European markets — a live constraint for European retailers — is not covered in any retrieved source. PSD2/SCA
• One-time ecommerce payments: All sources focus on subscription/recurring billing; smart retry application to one-time CNP ecommerce transactions (failed orders at checkout) is not covered in depth.
• Adyen 2024–2026 update: No public Adyen source describing updated Auto Rescue mechanics or 2024–2026 benchmarks was found. The only technical paper is from 2020.
• Independent benchmarking: No third-party controlled study comparing recovery rates across Stripe, Adyen, Recurly, and Chargebee in equivalent conditions exists in public sources.
• Network token interaction: How Network Tokenisation reduces failure rates upstream (preventing the need for retries) is referenced but not quantified with primary sources.

Key terms

Term	Meaning
Soft decline	A temporary failure — insufficient funds, velocity limit, network timeout — where the card is valid and a retry may succeed
Hard decline	A permanent failure — stolen card, closed account, fraud block — where a retry cannot succeed without a new payment method
Involuntary churn	Subscription cancellations caused by payment failures rather than deliberate cancellation; see Involuntary Churn
Dunning	The broader process of recovering failed payments, including both automated retries and customer-facing outreach; see Dunning
Adaptive Acceptance	Real-time retry of false declines during the original transaction attempt; distinct from Smart Retries; see Adaptive Acceptance
AMP	Acquirer Monitoring Programme — Visa programme that monitors and fines merchants for excessive retry volumes