How to investigate a falling payment success rate: device, amount and failure-stage segments
When a payment success rate falls, establish a consistent denominator before changing channels. A failed create request, an unpaid checkout, a delayed confirmation and a failed balance update are different outcomes. This guide shows how to separate them in a merchant report. All numerical examples are illustrative and are not 稳付pay performance claims.
Continue by topic
Define the cohort, observation window and denominator
Keep the merchant order ID, payment-attempt ID, creation time, payment method, channel configuration version and final state. One purchase may have several attempts after a channel change. Report accepted creates, confirmed payments, pending cases, explicit failures and business credits separately. A query timeout alone must not convert a payment to failed.
Locate the affected device and amount segments
Compare the WeChat browser, other mobile browsers and desktop first, then compare channels and stable amount bands within each group. Show the sample count and use cohorts with equal observation time. A decline on one device suggests checking checkout compatibility and navigation errors; a decline across devices also calls for channel, release and network checks.
Record the stage as well as the reason
Useful stages include order creation, checkout display, customer payment, confirmation, notification receipt, signature validation and balance credit. A confirmed payment without a callback is a notification problem; a verified callback followed by a failed database transaction is a fulfillment problem. Keep redacted error codes, timestamps and internal correlation IDs instead of a single generic failed flag.
Show pending cases beside the percentage
In an illustrative cohort of 100 orders, suppose 90 succeed, five fail and five remain pending. 90/95 is the success share among completed outcomes; 90/100 is the currently confirmed share of the full cohort. Neither is the final full-cohort rate yet. Display the pending count and cutoff time, then recalculate after reconciliation. Small samples also need their absolute counts.
Validate a new cohort without rewriting history
Preserve the original device, channel and configuration snapshot when applying a fix to new orders. Resolve paid-but-unfulfilled orders through audited, idempotent compensation rather than asking customers to pay again. Record the affected scope, failing stage, pending list and validation cohort. Actual 稳付pay channel availability, pricing and configuration are confirmed during onboarding.
Frequently asked questions
Does a slow callback mean payment failed?
No. Verify the original order through the agreed query and notification protocol. Track payment and notification states separately instead of treating a browser timeout as failure.
How many orders prove a channel is stable?
There is no universal count. Compare similar device, amount and time cohorts with meaningful samples, and inspect pending cases and concentrated failure reasons.