The term covers two layers. Infrastructure failover happens inside a provider: a PSP runs duplicate data centres and switches traffic when one fails, invisibly to merchants. Provider failover happens on the merchant's side: a merchant integrated with two PSPs sends traffic to the second when the first is unavailable. Merchants with meaningful online volume increasingly want the second kind, because provider-wide outages do happen and last long enough to hurt.

The cost is easy to estimate. A retailer processing €2.4 million a day sees roughly €100,000 of attempted sales per hour. A 90-minute outage with no alternative route loses a large part of €150,000, and some of those customers do not come back.

A working setup has four parts:

1. Detection: provider health monitoring tracks error rates, latency and timeouts per provider in near real time.
2. Decision: when thresholds are crossed, for example five consecutive timeouts or an error rate above a set percentage over one minute, the routing layer marks the provider as degraded.
3. Rerouting: new transactions go to the backup provider, while in-flight ones are resolved first.
4. Recovery: the system probes the primary with a small share of traffic and restores it gradually once it is stable.

The hard case is a request that timed out. The first provider may have approved the payment and simply not answered in time. Sending the same card to a second provider risks charging the customer twice. A sensible timeout strategy resolves the status of the first attempt, or voids it, before retrying elsewhere, and an idempotency key prevents duplicates within any one provider.

Card data portability is the other requirement. If cards are tokenised in the first PSP's vault, the backup cannot charge them. Merchants solve this with network tokens, a third-party vault, or a payment orchestration layer that holds tokens independently of any single processor.

Failover reacts to technical failure: timeouts, 5xx errors, an unreachable endpoint. Cascading reacts to a business outcome: a decline from one acquirer, retried through another. The two are often configured in the same routing engine but should follow different rules, since cascading a hard decline breaks card scheme retry rules.

For a PSP being acquired, buyers ask for uptime history, incident reports and evidence of tested failover between data centres. Regulators ask similar questions under operational resilience rules, including DORA for EU financial entities.