We model the time and network element failure dynamics of network operators' service level agreement (SLA) risks. This involves identification of events interrupting service and stochastic modeling of failure events. The key concepts are modeling how a service builds from simple network components and the approximation of interval availability, when failures are rare and the SLA tracking period is in month/year scale. The dynamic SLA-risk gives rise to a component importance measure, which prioritizes repairs or indicates the importance of operability of a component in terms of SLA-risk in the current network and SLA state. Our work can be used in network planning and OAM.