Often data processing is not implemented by a workflow system or an integration application but is performed manually by humans along the lines of a more or less specified procedure. Collecting provenance information in semi-structured processes can not be automated. Further, manual collection of provenance information is error prone and time consuming. Therefore, we propose to infer provenance information based on the file read and write access of users. The derived provenance information is complete, but has a low precision. Therefore, we propose further to introducing organizational guidelines in order to improve the precision of the inferred provenance information.