It's not a binary. If you expect to receive 90 meritorious complaints and 10 bad ones, a review procedure with no false negatives and 10% false positives will give you the wrong outcome 1% of the time. You might be OK with that depending on the cost of a false positive. But if the rise of AI means you get 100 meritorious complaints and 100 bad ones, and you apply the same process, now your processing costs have doubled and your error rate has quintupled.