Figure out the unhappy paths through the application. Prioritise those failure cases according to common-ness / likelihood. Make sure the key failure cases are accounted for. Pay attention also to errors that could arise from remote deployment, e.g. backend being down temporarily