Lessons from running legacy managed-services operations, applied to AI-era rebuilds

Long before "AI automation" was a category, healthcare revenue-cycle operations were already running on a version of the same split. Claims processing, billing, collections, prior authorization — high volume, rule-governed, and yet full of cases that rules alone couldn't safely resolve: a denial that needed a judgment call, an authorization that didn't fit the standard criteria, a collections case that needed a human read on how to handle it.

Those operations worked when the volume side was handled systematically and the judgment side was staffed by people with real standing to decide — not when either one tried to do the other's job. Automation that tried to adjudicate edge cases produced bad outcomes at scale. Specialists handling routine volume by hand produced backlogs and burnout. The operations that actually worked kept the line explicit and revisited it as rules, regulations, and case mix shifted.

That's the direct lesson for AI-era rebuilds: the technology changed, but the structural problem didn't. Volume still needs to be absorbed systematically. Judgment still needs to sit with people who can actually exercise it. What's different now is how cheaply the volume side can be built — which makes it more tempting, not less important, to get the split right rather than automate everything because you finally can.