That shouldn’t be the case. Could you show us the code snippet you’re using for extraction? Most common problem is that people rely on tsvLine.split("\t") instead of tsvLine.split("\t", -1). Another possible problem is that you might use very old Kinesis Enrich that produced fewer columns (but it should be really old, ~ end 2016)
Regardless of that, I’d recommend you to use our Scala Analytics SDK for reading enriched data. It is type-safe and battle-tested. I haven’t tried using it from Java, but Scala should be entirely compatible and maybe just add several extra dependencies.