The disclosure relates to a method, network node and service instance, the method being executed by a service provider, for optimizing resource allocation according to the needs of at least one client and according to an SLA between the client and the service provider. The method comprises obtaining a statistical behavior of a workload forecast function associated with a service of the client, obtaining a forecast value for a future workload from the workload forecast function associated with the service of the client, obtaining at least one service-level objective associated with the SLA between the client and the service provider, obtaining a modified forecast value based on the forecast value, the statistical behavior of the workload forecast function and service-level objective associated with the SLA, and adjusting resources allocated to the service based on the modified forecast thereby optimizing resource allocation for the at least one client.