Stephen Planck explains the value (and limitations) of filtered indexes:
When a table holds millions of rows yet most queries touch only a small, well-defined subset, a traditional non-clustered index feels like using a searchlight to find something sitting under a desk lamp: the index still stores an entry for every row even though the workload rarely needs most of them. SQL Server’s answer is the filtered index—introduced in SQL Server 2008 and still under-used today—allowing you to index just the rows that match a predicate you supply in a
WHEREclause.
I always thought I would get more value out of filtered indexes. Instead, they’re very useful for specific circumstances, but not the all-purpose performance dynamos you’d want them to be, particularly because of the limitations.