You’re assuming that there’s no cost involved in moving a potential victim through the pipeline. I’m sure AI has changed the game, but the general idea was that beyond the initial blast of spam you would have someone actually responding to those who fell for it. Putting in signals that it was a scam filtered out individuals who would waste scammer time because they would eventually figure it out before falling victim. By selecting for people who literally can’t pick up on obvious signs of a scam, you save yourself a lot of time and energy.