Exactly-once semantics is a cornerstone - and often unspoken - rule of many business requirements: we must not charge a user twice for the same product, deliver the same service twice, or lose a payment halfway through.

But the distributed systems we build on rarely provide exactly-once behavior end to end. As business rules grow more complex and architectures span services, queues, databases, and third-party APIs, getting this right becomes a really hard engineering problem.

In this talk, we’ll walk through a real fintech case from the initial product promise to the technical architecture behind it. We’ll bridge the gap between business expectations and the guarantees engineers must build, examine the most important failure modes, and arrive at a solution that delivers effectively-once behavior in practice.