The processor is not the source of truth

D2C, B2B, and subscriptions share one payments system. The processor is a choice. The ledger is not.

The failure mode is this: if the processor is the source of truth for the card, you do not have a platform. You have a vendor. Failover means re-collecting cards. A challenge is whatever that processor ships. A second market or a second product line becomes a second integration, then a third.

You orchestrate for more than lock-in. If a provider is down, you retry on another. You send a charge to the provider with the highest approval rate in that location. And you can move volume quickly, which is a lever when you talk rates: a PSP that knows you can leave will come to the table.

Own the payment. Vault the card once. Route each attempt. Keep the challenge independent of any one processor. Do not expand PCI scope to buy that flexibility.

Authorization, capture, settlement, and reconciliation are shared problems. Retries are the same payment. An unresolved attempt does not get a second dispatch. The product surface is not the money path.

Approval rate is a routing problem. Cost is a routing problem. PCI is an architecture constraint, not a checklist you visit after the integration works. If you own the payment, adding a processor is an adapter. If the processor owns the payment, adding a processor is a migration.

I have lived the migration version. At Natural Cycles a processor change bought 13.36 points of approval. I would rather route than migrate.