Formative assessment is a practice aimed to enhance teaching and learning and to develop self-regulation. Among the strategies for activating formative assessment, the provision of feedback is one of the most studied: good feedback helps close the gap between actual and desired performance and it promotes self-regulation. Automatic Assessment Systems can improve the practice of formative assessment, especially for the immediacy of feedback and the flexible solutions for questioning. In this context, the Department of Mathematics of the University of Turin has developed a model for automatic formative assessment using an automatic assessment system suitable for STEM disciplines. The key features of the model, grounded on the models from the literature, are: availability, algorithmic questions, open answers, immediate feedback, interactive feedback, and contextualization. These are detailed and discussed through evidence gained from three experimentations, involving the total number of 553 students. In particular, it emerges that the use of interactive feedback ensures that students process the information from the feedback and use it to improve their performance, thus solving the major problem raised in the literature that feedback are not useful when students do not read them.