Currently insert is inefficient because we're constructing an aggregation pipeline that is expensive to parse. It would be beneficial to avoid the overhead of parsing the BSON aggregation pipeline which is fairly stable in its structure.