The number one factor is frequency. If a query or stored proc that would use this index is run hundreds or even thousands of times a day, an index that will give even a small amount of speed advantage, is worthwhile.

Another factor is amount of advantage. If you have a query that's run just once a month, like for an end of month report, that takes 12 hours to complete and runs into the next workday, I've seen the right index change that to a few seconds, and is worthwhile.

Think size - a small table of a few hundred rows that won't be growing doesn't need as much indexing attention as compared to a table with millions of records. Small frequently accessed tables might already be loaded in memory and an index won't help at all.

Yet another factor, toward the other direction, is that you don't want to over-index a table - it slows performance when performing inserts, updates and deletes, requires maintenance and takes up space. In general, even a highly used table shouldn't have more than 10 indexes. Unless it's a data warehouse table or strictly used for reporting (i.e., there's very little addition and no change to the data) - then you might find it beneficial to have lots of indexes.

These are some of the major factors - enough for a start. Give that one you found a try!