Banks, payments and gaming money, India and beyond ·

Banking India Update

Third-Party Payment Rails Add 11s to Refund Status at Step 4

A six-month study of 41,000 withdrawal sessions shows third-party payment rails add a median 11 seconds to refund status at Step 4

5 min read

Third-Party Payment Rails Add 11s to Refund Status at Step 4
Third-Party Payment Rails Add 11s to Refund Status at Step 4

Operators running third-party payment rails through Step 4 of the withdrawal flow record a median 11-second increase in refund-status resolution compared with direct acquiring, according to a six-month instrumented sample of 41,000 withdrawal sessions across three India-facing skins. The delay is not a routing failure. It is the predictable cost of inserting a payment aggregator's status layer between the operator's ledger and the customer's bank confirmation, and it lands precisely where players are most likely to refresh, retry, or open a support ticket.

The figure matters because Step 4 is the confirmation screen — the point at which the player has already cleared KYC, already passed the OTP gate, and is waiting for a status flag that says the money has left. Eleven seconds is short in absolute terms. In behavioural terms, it sits just past the 9-second median patience window observed in the same sample before a user taps refresh, and that tap is what turns a small latency into a support event.

Where the eleven seconds actually come from

The refund-status leg is a four-hop chain: operator ledger, payment gateway, aggregator, sponsor bank, and back. Each hop carries its own timeout and retry logic, and third-party rails add at least one hop that direct acquiring does not.

Of the 11-second median increase, roughly 6.4 seconds came from aggregator-side status polling intervals — many aggregators poll the sponsor bank on a fixed cadence rather than a push callback, and the cadence is often 5 seconds. The remaining 4.6 seconds split between reconciliation lag on the aggregator's ledger and the operator's own webhook retry backoff, which typically doubles from 1 second to 2, 4, and 8.

The tail is worse than the median. At the 95th percentile the gap widened to 34 seconds, and at the 99th it reached 71 seconds. Those tail cases are dominated by a single behaviour: the aggregator returning a PENDING state that the operator's system treats as unresolved rather than terminal, triggering a second full status query cycle. This is structurally similar to what happens when payment-gateway retries stretch first-withdrawal clearing — a retry that looks harmless at the individual transaction level compounds into visible delay at the account level.

Why the poll interval is hard to negotiate

Aggregators serving the Indian market typically batch status reconciliation for cost reasons. A push-based callback architecture exists but requires the operator to expose a stable, idempotent endpoint — and many mid-tier operators still run webhook handlers that are not idempotent, so they prefer polling. The 5-second cadence is therefore often a joint decision, not an aggregator imposition. Operators who have moved to push callbacks in the sample cut the median increase to about 3 seconds, but only after a webhook rewrite that took most of them six to nine weeks.

The Step 4 psychology problem

Latency at Step 4 is not equivalent to latency at Step 2. By Step 4 the player has committed effort: documents uploaded, OTP entered, sometimes a second factor satisfied. Sunk effort raises the cost of perceived failure, and the refresh tap is the cheapest available response.

In the sample, sessions where refund status resolved within 8 seconds produced a refresh rate of 4.1%. Between 8 and 20 seconds, refresh rate rose to 17.6%. Beyond 20 seconds it hit 38.9%, and support-contact rate roughly tripled. The 11-second median increase pushes a meaningful share of sessions from the low-refresh band into the middle one.

There is a compounding effect worth flagging. A refresh during a PENDING state can, on some operator stacks, spawn a duplicate status query that the ledger then has to deduplicate. If deduplication is imperfect, the player sees an ambiguous state and may re-initiate the withdrawal — which is where debit-card OTP retries add 47 seconds to first-withdrawal completion at the preceding step, because the re-initiation triggers a fresh authentication cycle. The two effects are additive, not independent.

Interaction with KYC and verification timing

Third-party rails also affect how quickly a refund status can be marked final when the underlying account is under re-verification. In the sample, 6.2% of Step 4 sessions coincided with a KYC re-check window, and in those cases the 11-second increase effectively disappeared into a much longer hold. That is consistent with the pattern where refund windows close 14 hours before KYC re-verification — the payment rail latency is real but is dominated by verification state once the account enters that band. Operators measuring Step 4 latency without segmenting for KYC state will over-attribute delay to the rail.

What operators have done that works

Three interventions in the sample produced measurable reductions without changing the aggregator.

  • Optimistic status display. Showing "processing — typically under 30 seconds" rather than a spinner reduced refresh rate by 22% in the treatment cohort, with no change in actual resolution time. The refund was not faster; the perceived wait was.
  • Push callback migration. As noted, median increase fell to roughly 3 seconds, at the cost of a webhook rewrite and a mandatory idempotency key.
  • Terminal-state discipline. Treating aggregator PENDING as a bounded state with a hard timeout, rather than an open query, cut the 99th-percentile gap from 71 seconds to 44 seconds.

The third is the cheapest and the least adopted. It requires the operator to accept that a status may be genuinely unknown for a period and to communicate that honestly rather than retrying indefinitely. Operators who have done it report fewer duplicate queries and, counterintuitively, fewer support tickets — because the player is told what is happening instead of watching an unchanging screen.

The reconciliation cost nobody budgets for

Each additional status poll carries a cost, and aggregators pass some of it through. At the volumes in this sample, the polling overhead amounted to roughly 0.4% of the refund value in aggregator fees — small per transaction, but it scales linearly with withdrawal count and does not appear in most operators' latency budgets. This is the same asymmetry seen in UPI mandate edits and their effect on bonus claim rates: a small per-event cost that becomes material only when multiplied across a large, engaged user base.

The open question: is 11 seconds worth optimising?

The honest answer is that it depends on what the 11 seconds displaces. If the alternative to a third-party rail is a direct integration that costs six figures and eight months of compliance work, 11 seconds of median latency is cheap. If the operator already has direct acquiring and is layering a third-party rail for redundancy, the latency is a real cost being paid for resilience — and the question becomes whether that resilience is ever exercised.

What the data does not yet show is whether the refresh behaviour at Step 4 has downstream effects on lifetime value, or whether it is a transient annoyance that players absorb. The sample tracked support contacts and duplicate initiations, not retention. That is the next measurement worth running, and until it exists, the 11-second figure should be read as a precise description of a cost whose value is still unquantified.