Protection in access networks may be cost prohibitive: the availability gain may not be able to justify the necessary investment. Even though several expected (near) future applications of optical access (e.g. mobile backhaul and fixed mobile convergence, or long reach PONs with extensive node consolidation) are often cited as rationale behind access network protection, the additional investment still needs to be minimised. A key to resolve the conflict between desirable availability and the prohibitive investment is the selective deployment of redundancy. Knowing not only the architecture, but also the geographical layout of the network offers the possibility to meet global availability requirements, while protecting just a minimal subset of network elements. In this paper we investigate how far a topology-dependent selective and partial protection scheme can reduce the cost of achieving given availability levels.