Redundancy always looks like a discretionary cost until the first outage makes it look like the cheapest decision the team ever made. Every client conversation about "do we really need failover" happens before an incident, never after one.
What we were doing before we learned this
Early on, we scoped infrastructure to handle expected load with a comfortable margin, and treated full multi-region failover as a premium add-on for clients who explicitly asked for it. It seemed like the responsible, cost-conscious default.
What changed our mind
A client's single-region compute cluster went down during a regional cloud provider incident that had nothing to do with anything they'd built. Total downtime was under two hours. Total cost — lost transactions, support load, and the emergency engineering time to migrate under pressure — was more than triple what a standing multi-region setup would have cost for the entire year.
- The failover infrastructure itself was cheap; it was idle capacity most of the time
- The expensive part was never the redundancy — it was staffing an emergency response while revenue was actively being lost
Redundancy is a fixed, predictable cost. An outage is an unpredictable one, and it's always larger than it looks from the outside.
What we do differently now
Multi-region failover is now the default in every infrastructure proposal, not an upsell. When a client wants to skip it to save cost, we show them the same math: the standing cost of redundancy against the realistic cost of a few hours of downtime at their transaction volume. That comparison usually settles the conversation on its own.