Government agencies deploy AI to cut appeal processing time in half, but accuracy drops 11 points

A system using AI can sort citizen complaints, applications, and proposals three times faster than a person, dropping processing time from 20 minutes to 9 minutes per appeal. The trade-off is real: the AI gets 78% of classifications right, compared to the 67% humans achieve, which means it sorts most appeals correctly but creates new errors at a different rate.

Government agencies are drowning in citizen submissions and have been processing them slowly and inaccurately for years. This shows a concrete option: deploy machine learning to move things faster, even if the accuracy floor changes shape. The real question is whether agencies will accept 11 extra percentage points of misclassification in exchange for 54% faster throughput, or whether they'll keep humans in the loop as a check. Either way, this is what the first wave of AI deployment in government bureaucracy probably looks like: faster, not necessarily better, with tradeoffs nobody wanted to make explicit.