Various services are now available in the Cloud, ranging from turnkey databases and application servers to high-level services such as continuous integration or source version control. To stand out of this diversity, robustness of service compositions is an important selling argument, but which remains difficult to understand and estimate as it does not only depend on services but also on the underlying platform and infrastructure. Yet, choosing a specific service composition may fail to deliver the expected robustness, but reverting early choices may jeopardise the success of any Cloud project. Inspired by existing models used in Biology to quantify the robustness of ecosystems, we show how to tailor them to obtain early indicators of robustness for cloud-based deployments. This technique helps identify weakest services in the overall architecture and in turn mitigates the risk of having to revert key architectural choices. We illustrate our approach by comparing the robustness of four alternative deployments of the Sens App application, which includes a Mongo DB database, four REST services and a graphical web-front end.