Based on past experience, but will be done carefully in future.

1. Essential to know what users really do, and why they do it.
    
2. Very ad hoc integration and analysis, software used once and thrown away. Usually done by grad students and post docs.
    
3. Labor intensive and slow.
    
4. Complex integration tasks beyond most scientists, hard even for IT professionals.
    
5. Analysis by statistical packages, old FORTRAN programs, maybe some new ad hoc programs.
    
6. Data may be from web services, but more likely from formatted ascii files or spreadsheets, especially for intermediate results.
    
7. Perl (or Python) scripts for re-formatting and collating data: the "duct tape of the internet".
    
8. Light weight!