WORLD AFFILIATE GUIDEINDEPENDENT · INTERNATIONAL · PRACTICAL

Rules and integration

Make conversion and refund events reliable

Prevent duplicate commissions and missed updates with stable identifiers, retry testing and business reconciliation.

Illustration: screens monitoring event flows
Editorial image for Conversion and refund events

Before you begin

A received event is not yet a correct commission. Messages can repeat, refunds can arrive before processing finishes and a service can respond while business recording fails. This method helps programme and integration owners define necessary evidence. Confirm transport properties in your provider’s documentation; Stripe is used here as a documented technical example.

Separate three identifiers

Distinguish message ID, business-operation ID and partner ID. The first identifies another delivery of the same message, the second connects orders, invoices, cancellations and exports, and the third carries attribution. Add event type, version and business timestamp to the test record. Do not infer a reliable order identifier from a monetary amount and customer address.

Verify before processing

Use the provider’s documented authenticity mechanism, including signature verification where available. Do not change secrets or their configuration as part of an editorial acceptance test. Define what is recorded after an invalid signature and avoid personal information in public logs. A successful HTTP response does not prove authenticity or successful business processing.

Make retries safe from duplicate effects

Specify what happens when an event returns: retrieve the existing result rather than create another commission. Keep processing evidence linked to the business operation. API idempotency keys and inbound-event deduplication address different problems. Stripe documents them separately; do not assume one automatically protects the other.

Handle events arriving out of order

Transport may not guarantee the expected sequence. Describe a refund received before its order exists locally. Prepare controlled waiting or retrieval from the source system where supported. An older update should not overwrite a newer state without an explicit rule. Preserve occurrence and receipt times to explain chronology.

Test failure and recovery

In an authorised environment, simulate unavailability, delayed responses and repeated receipt. Check the provider’s retry limits and timing. Manual replay may coexist with automatic retry; understand their interaction. Define who can replay an operation, required evidence and post-processing checks. The goal is a single business effect, not merely a successful message status.

Reconcile with business records

Regularly compare source orders and invoices with eligible platform operations. Distinguish normal delay, documented rejection and actual loss. Reconcile counts and amounts by currency: one monetary total can conceal duplicates offset by missing records. Preserve end-to-end identifier samples for purchases, partial refunds and cancellations.

Prepare useful incident evidence

For each discrepancy, collect business reference, event type, timestamps, version, expected state and observed state without exposing secrets. Identify the first divergent step. Link the record to the tracking incident method and assign an owner and resolution deadline. After correction, replay neighbouring cases as well: fixing refunds must not duplicate purchases.

Test a retry with the same identifier

Fictional case: an €80 sale is sent twice following a network interruption. The expected outcome remains one recognised sale, followed by an explainable status change if cancelled. Compare business identifier, timestamp, recipient response and reconciliation state. A successful technical response does not establish correct attribution or the payable amount.

Recover after partial success

Simulate a successful business write followed by an interrupted response. On redelivery, check that the system finds the existing write, avoids a second commission and resumes only the unfinished step. Record business and delivery outcomes separately: retrying must not erase evidence of the first effect.

Replay an order from purchase to refund

A reliable conversion has a stable business identifier. Retrying delivery must not create another sale. Run these cases in a test environment and compare the merchant system with the event recipient.

Illustrative sequenceExpected business stateFailure to detect
Order T-104 confirmed for €100One €100 saleEvent sent before confirmation
Same notification received twiceStill only one saleDuplicate counting after a retry
Partial refund of €30€70 remaining valueIncorrect full cancellation
Remaining €70 refunded€0 remaining valueNegative balance or missing refund

GA4 purchase and refund events use transaction_id to identify the transaction; item data can describe a partial refund. This is an analytics contract. Commission validation still depends on affiliate terms and the merchant ledger: an analytics event should not automatically become a payment instruction.

Practical answers

Questions to settle before signing

Does HTTP 200 mean a commission is correct?

It indicates a response from the receiving endpoint. Processing, business reference, eligibility and final state still need verification. Acknowledgement semantics depend on the integration.

Does API idempotency remove every duplicate?

No. It concerns the operation and scope defined by the API. Repeated event delivery needs its own controls. Test both paths separately.

Sources and further reading

EDITORIAL NOTE

This guide is designed to help you ask better questions and organise a practical plan. It is educational and does not constitute legal, tax or financial advice. Platform rules, market conditions, technical capabilities and eligibility requirements can change, sometimes with little notice. Confirm material decisions with current official sources and, where the consequences matter, with qualified professionals who understand your market and organisation.

Next step

Turn the framework into a shortlist.

Write down the outcome you want, the limits you cannot ignore and the evidence that would make a test successful. Then compare only the platforms that fit those conditions.

Compare platforms →Open the glossary

Documented insight

A refund must find its original sale

Google Analytics documents transaction identifiers for purchase deduplication in web streams and refund processing. Keep the reference stable across store, tracking and partner systems. A refund without a stable reference may correct a financial total without correcting the relevant commission. Do not assume web-stream deduplication applies identically to app streams.

Test duplicate purchase delivery, full refund, partial refund and a refund received before final purchase confirmation. Check identifiers, items, amounts and final states in both systems. Use an identifier unique to each order, without customer names or email addresses; an empty identifier can incorrectly deduplicate different purchases.

For partial refunds, match the refunded items and corrected value. Distinguish event receipt, acceptance and final accounting state. Retain the order reference and reason for manual corrections so that later synchronisation does not reintroduce the discrepancy.

Identifier
Unique per order, stable and without personal data.
Events
Test duplicates, partial refunds and full refunds.
Reconciliation
Compare the final state across systems.

Can all real sales use the same test identifier?

No. It may incorrectly deduplicate purchases. Use a test environment and fictional orders for trials.

Try the reasoning

A partially refunded order retains its full commission. What reconciliation is missing?

Reasoning: Connect the refunded line to its sale and commission rule. A total refund amount does not identify each affected event.

Primary source : Google Analytics — Identifiants de transaction et déduplication