DEV Community

Executive Mobility
Executive Mobility

Posted on

PIX, Retries and Approval Rates: Phoenix365 on LatAm Payment Cascades

A player in Mexico tries to deposit through SPEI shortly before a major football match. The payment provider times out, the cashier keeps spinning, and the player leaves before placing a bet. For an ordinary retailer, a failed payment might mean a purchase postponed until tomorrow. In iGaming, it can mean losing the session and the customer. At Phoenix365, a B2B casino and sportsbook technology provider working with Latin American operators, payment reliability shapes the operator experience as much as it shapes the finance team’s numbers. Effective orchestration depends on provider routing, PIX-specific behaviour, safe retries, and honest measurement of approval rates.

Why One Payment Provider Isn't Enough in LatAm

Latin America is not one payment market. Consumers have different local rails, bank arrangements, card acceptance patterns, and wallet preferences. An operator using one international checkout configuration may offer a technically functioning cashier that still fails to serve important customer segments.

Common methods illustrate the variation:

  • PIX, Brazil: Instant account-to-account transfers initiated through banking apps.
  • SPEI, Mexico: An interbank transfer system operated by Banco de México.
  • PSE, Colombia: A bank-account payment option used for online purchases.
  • Local cards and e-wallets: Familiar alternatives whose coverage varies by country and provider.

A single payment processor may create several points of weakness:

  • Peak-time interruptions: Timeouts during popular matches can interrupt deposits precisely when demand rises.
  • Issuer differences: Card approval may vary significantly by issuing bank or acquisition route.
  • Coverage gaps: One connection may not support the required local method, currency, or customer segment.
  • Provider limits: Transaction caps, volume limits, and maintenance windows can unexpectedly restrict availability.

Our Payments Hub lets each operator build its own set of providers and methods for its market, while the operator chooses the providers and holds the contracts. Which methods an operator can offer for gambling depends on the rules in each market.

A payment cascade is an ordered sequence of eligible processors: a payment starts on the preferred route and can move to a fallback only when the earlier outcome makes switching safe and permitted.

How PIX Changes the Cascade

This section uses PIX as an example of how instant account-to-account payments change cascade design; the same pattern applies to similar rails in other markets.

PIX is the instant payment system created by Brazil's central bank. It works around the clock and lets customers pay through a QR code or a copy-and-paste payment code. For a European reader, the distinction from a card authorisation is important: the customer actively initiates a bank transfer, while with a card the issuer approves or declines the charge.

In a card cascade, routing can be revised around a declined authorisation, subject to network rules. With PIX, routing largely occurs before the payment request and QR code are generated. Once the customer sees the code, the request is tied to a particular payment flow. Silently creating a second request through another provider risks duplicate payment instructions and customer confusion.

Three failure points need distinct handling:

  • Request creation fails: The provider never returns a usable payment instruction, so a fallback may be considered after checking the original request's status.
  • Customer does not pay: The QR code expires or remains unpaid; the cashier should say the payment has not arrived and offer a clear, new payment option.
  • Payment settles but confirmation arrives late: The system must match the late confirmation to the original payment and credit it once.

PIX can also be used for payouts to registered recipient keys or eligible account details, and refunds should follow the applicable refund mechanism and the original payment record. A payout flow still needs account-ownership checks: a PIX key on its own does not verify who the recipient is.

Designing the Cascade: Routing Rules and Order

Good routing depends on the transaction in front of it. A route that performs well for one bank or transaction value may be inefficient for another. The decision engine should consider:

  • Payment method, country, and currency.
  • Transaction amount and applicable provider limits.
  • Issuing bank or card BIN, where relevant.
  • Provider health, recent errors, and response time.
  • Transaction fees and total processing cost.

The first provider should usually be the eligible route with the strongest reliable results for that segment, even when a cheaper one is available. Backup providers follow a defined order. Commercial decisions balance acceptance, latency, settlement reliability, fraud exposure, and fees.

A few examples of how routing signals can affect the cascade:

  • High recent timeout rate: Temporarily reduce or suspend traffic to the affected route.
  • Strong approval for a bank segment: Prefer that processor for matching card transactions.
  • PIX request not created: Check the status before issuing a new code through another provider.
  • Increased processing cost: Reassess the ranking without sacrificing legitimate approvals.

A circuit breaker helps automate provider health management: when error rates cross a threshold, the system pauses new traffic on that route and later tests recovery before returning it to service. Thresholds, cool-down periods, and partial recovery need deliberate configuration.

Retries Without Double Charges

A retry is not automatically safe merely because the customer has not seen a success message. Timeouts can conceal a transaction that was accepted downstream. Before switching providers, the cashier must establish whether the earlier attempt is definitively failed, still pending, or successfully settled.

Distinguish recoverable and definitive outcomes:

  • Potentially retryable: Confirmed temporary gateway unavailability, issuer downtime, or a soft decline when the relevant provider and network explicitly allow another attempt.
  • Do not retry: Fraud blocks, stolen or blocked cards, explicit “do not retry” responses, and unresolved transactions that may already have succeeded.

Every logical deposit should have a stable identifier and idempotency protection so repeated API calls do not create an additional charge or a second wallet credit. Individual processor attempts also need their own tracking references. Idempotency must work alongside reconciliation across providers; reusing one key on separate processors is not sufficient protection by itself.

Asynchronous confirmations complicate the flow. A deposit can remain pending while the provider processes the transfer or delivers a webhook. Wallet balance should increase only after verified success, and duplicate or delayed webhooks must be handled safely. When status is uncertain, query the provider and reconcile first; retry only once the earlier attempt is confirmed as failed.

Set a maximum number of attempts, minimum delays between retries, and stop conditions for hard declines. The player should see a helpful status message such as “Payment pending confirmation” with a way to refresh or contact support.

Measuring Approval Rates the Right Way

A single payment approval rate can hide more than it reveals. A cascade creates additional processor attempts, so dividing successful payments by all attempts produces a different number from dividing successful deposits by unique customer payment intents. PIX also needs a distinction between request creation and actual payment completion.

Track separate measures:

  • First-attempt approval and final completion after fallback.
  • Success rates by method, provider, bank, and relevant segment.
  • Time from initial request to verified confirmation.
  • Abandonment or failure at each stage of the deposit journey.
  • Total processing cost per successfully credited deposit.

Consider 100 unique deposit intents. Seventy succeed on the initial route; ten more succeed after allowed fallback attempts. Final completion is 80%, and first-attempt success is 70%. If the system made 140 processor attempts altogether, attempt-level approval is about 57%. All three figures are mathematically valid but answer different questions.

Risk controls must be analysed alongside payments performance. Tougher fraud screening may reduce approvals while preventing harmful transactions; relaxing checks solely to increase the headline payment approval rate can worsen outcomes. At Phoenix365, we use three-layer fraud analysis, while payment and risk teams should agree on shared definitions and escalation criteria. Compare new customers with returning players separately, because their verification status and checkout behaviour can differ substantially.

What to Get Right Before Building One

  1. Route by segment. Build routing rules from payment type, market, bank, amount and real provider performance.
  2. Handle PIX before showing the code. If creation fails, resolve the first attempt, then show the customer a clearly labelled replacement request.
  3. Retry only safe failures. Respect decline instructions, cap attempts, use idempotency keys, and reconcile pending transactions before modifying wallet balances.
  4. Measure both stages. Report initial approval independently from final unique-intent completion, with latency, fraud, and processing cost beside them.

A well-designed cascade can recover legitimate payments, but it cannot compensate for weak provider coverage, unclear status handling, or careless risk policies. What signals do you use to remove a provider from the cascade automatically?

Top comments (0)