Many tech leads live under the assumption that – if they acquire enough data to train their model, problems with quality will simply work themselves out. Wow! Magic. This assumption often operates under an additional (but false) assumption: There is only a negligible percentage of impure data.