Telecentre implementation has been inundated with failure in developing countries. This has necessitated the need for evaluations to unearth reasons for such occurrences. Various methods have been proposed to evaluate IT projects. However, very few of these methods have been used to predict telecentre failure. We combine two approaches (ICT4D evaluation and Machine Learning) to predict the likelihood for failure of a telecentre. Through the use of a case study, this paper uses the well-established Case-Based Reasoning (CBR) methodology to predict failure of telecentres. We apply CBR on real life dataset to predict the Design-Reality Gap score (DRGS). We compare three CBR methods with a naïve benchmark using ArchANGEL. We demonstrate through our experiments that CBR can be used to predict DRGS. This gives a refreshing indication suggesting that it may be feasible to use CBR to evaluate ICT initiatives and to predict adequately outcome of an initiative. Through this mechanism, it may be possible for managers and owners of telecentres to pre-empt an outcome and have the advantage to take mitigating steps. This affords managers an opportunity for remedial action for sustainability.