How would you choose a composite index for a filtered, ordered query?

Start with the actual predicates, sort order, selected columns, and data distribution. For a query filtering customer_id by equality and ordering by created_at and id, an index beginning with customer_id followed by the ordering keys may help. Check the query plan and measured workload rather than relying only on a rule of thumb. Indexes consume storage and add write overhead. Range predicates, selectivity, collation, and database capabilities affect which columns can efficiently support filtering or ordering. Recheck plans after meaningful data growth.