Performance improvement

I am trying to find ways to improve performance with existing PL/SQL block of code which is an equivalent of a MERGE statement. Could you please review following code snippet and suggest best practices or updates to improve the performance? I am thinking in the lines of using collections, FORALL, and BULCOLLECT.

Truncate a staging table

Insert into staging SELECT from external table. The external table may contain millions of rows. A hint has been used in the insert (--+append).

Gather table stats on the staging table : dbms_stats.gather_table_stats - cascade => TRUE

for loop (select * from staging)
         update target_table(20 million rows) , 45 columns out of a total of 54 columns with a where condition on two columns. Two of the columns that gets updated are using SYSDATE.
        if sql%notfound then
         insert into with one of the columns generating a sequence, and two other column values are defaulted to SYSDATE