Service quality assurance is of capital importance in modern cloud and network infrastructures, especially in multi-domain scenarios, where multiple operators collaborate to provide end-to-end (E2E) services. However, due to the dynamics of the multiple infrastructures and deployed services, it may be difficult to identify which domains need to perform (re-)configuration operations, named actuations, to keep the quality of the E2E services. In this regard, Machine Learning (ML) techniques appear as an interesting solution for guiding the actuation systems in multi-domain scenarios. With this in mind, we present a novel approach for self-optimised multi-domain service provisioning, leveraging the capacities of Deep Reinforcement Learning (DRL), with a focus on E2E service quality assurance. Running away from traditional approaches, the presented proposal tries to minimize the number of domains that need to actuate rather than determining the exact domain-specific actuations. We compare our proposal to existing strategies in terms of performance, scalability and applicability in real scenarios.