An SLA often starts as a set of numbers: availability, latency, response time. But the picture is more complicated when a platform is the foundation for other products rather than an end-user service.
Initially, we measured availability and response time like everyone else. Customers wanted numbers, but those numbers said little about the service they actually received. Applications failed because of networks, virtualisation or customer architecture decisions, while the platform team was told: “You breached your SLA.”
We tried new metrics and separated our responsibilities from those of other teams. Yet we were still doing arithmetic around the problem. The breakthrough came from asking: what do we actually provide?
The platform’s value was a predictable, stable and manageable environment. We shifted from defending ourselves against failures outside our control to measures and practices that could improve reliability.
This became a service-level manifesto: principles for working with infrastructure, network and virtualisation dependencies; investigating incidents; and using SLIs to discover problems rather than merely justify penalties.
The SLA became a tool for understanding causes, proposing solutions and improving business operations — backed by the ability to act, rather than formal compliance alone.