Spot on with seperating the use cases/access pattern here.
Indexes really won't help much when most of the data needs to be read anyhow (OLAP).
But they can help with queries on LARGE data sets (finally a good use for all upper case typing 
) that whould only return a small fraction of records, which is the typical retailer order line-item query.
As you will know by now I am not the biggest fan of "best practice" papers. I rather see that the technology is sufficiently understood and then applied. Anyhow *hinting back*: this is of course covered in Richard's and my book ...