Dementia is one of the leading causes of disability and dependence among older people, currently affecting more than 55 million people and estimated to increase to 139 million by 2050. A growing number of technologies are being developed to assist people with dementia in their daily lives, but assistance with toileting remains a neglected area. In this work, we present RITA, a system to automatically guide people with mild dementia in the toilet. The system detects activities that the user performs and compares them to a predefined model that maps out the correct toilet procedure. If problems are detected, such as wandering around the toilet or sitting on the bowl for longer than a certain time, instructions are given on what to do next. As only depth images are used, the privacy of the users is guaranteed at all times, a crucial factor in such an intimate context as toilet-going. For an automated assistance system to be effective, the adequate design of its interaction with the user is essential. For this reason, our design is not only based on the available technological solutions, but also on the conclusions derived from focus groups with healthcare professionals. We report these findings with the aim of contributing to reducing the information gap in interaction design for people with dementia.