Financial infrastructure
Exactly-once settlement at financial scale
An event-driven microservices platform for high-value settlement, where a duplicated or dropped message is a financial loss rather than a retry.
- 40%
- faster end-to-end settlement
- 0
- reconciliation breaks attributable to the messaging layer
- Exactly-once
- delivery semantics across service boundaries
The platform moved high-value settlements between counterparties. The existing design treated messaging as transport — publish, hope, reconcile later — and the reconciliation team absorbed the cost of that assumption every morning.
The actual problem
Settlement latency was the symptom. The cause was that transactional consistency stopped at each service boundary. A transfer could be committed locally and lost in transit, or applied twice, and nothing in the architecture made either case impossible — only unlikely, and only until load rose.
What we built
An event-driven architecture where the log is the source of truth rather than a side channel:
- Transactional outbox in each bounded context, so a state change and the event announcing it commit or fail together. There is no window in which one exists without the other.
- Idempotent consumers keyed on a deterministic settlement identifier, which makes redelivery a no-op instead of a double-spend.
- Bounded contexts drawn on money movement, not on the org chart — the boundaries follow where consistency is genuinely required.
Outcome
End-to-end settlement time fell 40%. The more valuable result is the one that does not appear on a dashboard: the class of bug where money moves twice was designed out rather than monitored for.