Knowledge capture and reuse is a challenging task consisting of many steps. The knowledge discovery cycle presented by Fayyad offers a global overview of how these steps are combined together. By taking a step back and considering data as activity traces, we propose to change this knowledge discovery cycle. We consider that knowledge is built rather than discovered. Therefore human involvement to make sense out of the traces is paramount. We propose to use knowledge engineering techniques to benefit from the user's knowledge as early as possible. The use of modelled traces can solve this issue and change the knowledge discovery process to make it more interactive and knowledge-intensive. The proposed methodology will be illustrated with software applications we have developed which, combined together, support the whole process of knowledge discovery. A discussion of the proposed methodology and of the required tools is given.