If you pull data from a database you're doing the Extract part.
As soon as you do something to that data, for example rename a field or calculate the year from a date field, then your doing the Transform part. If your load script finishes you've finished the Load part. That's the basics.
You can add intermediate steps where you for example first extract the data from the database and store that data unchanged in qvd files, one for each extracted table. In a second stage you use these first stage qvd files to transform the data and apply business rules to create a new set of second stage qvd files. These could for example be for dimension and fact tables that can be shared between more dashboard qlikview documents that the end users will interact with. The final stage in this case are the dashboard workbooks that merely load data from the second stage qvd files that are needed for the dashboard.
You can make this as complex as necessary. Perhaps it makes sense to have more transformation layers to implement chains of transformations that are necessary to create the final qvd's that are necessary for the dashboard workbooks.