Process mining has been studied for many years but has not been so widely adopted in real business practices. In this study, we propose a practical approach to process mining. This approach has three characteristics. Firstly, we make use of transaction databases of business systems that don't necessarily have an identifier throughout a process instance, rather than workflow logs. Secondly, we visualize and analyze what really happened without model inference. Thirdly, we also analyze exceptional processes as well as typical processes. This approach is implemented in a tool called BPM-E and was successfully adopted in real business systems in Japan, North America, and Europe. We show two case studies of BPM-E in this paper to explain how it works after the technical description.