Every home health company says some version of "AI plus human review" right now. It shows up on every website, every one pager, every sales deck. And most of the time, it means almost nothing. Nobody explains what the AI actually does, what the human actually does, or where one hands off to the other.

So here is what it actually looks like inside a real chart review at Olli.

The chart comes in

A chart lands in the queue. It could be a start of care OASIS, a recert, or a full episode getting ready for a Plan of Care review. Before a person ever opens it, the AI runs a first pass.

This first pass is not guessing. It is checking the chart against patterns built from thousands of prior reviews: does the GG0130 scoring line up with what is documented in the clinical notes, does the primary diagnosis support the case mix assignment, are there M-item answers that contradict each other, does the documentation actually support the acuity being billed. The AI flags anything that looks off and tells the coder exactly why it flagged it including sharing the CMS justification.

This is the part people usually picture when they hear "AI coding." But it is only step one, and it is the smaller half of the work.

The coder validates, not rubber stamps

Every flag the AI raises goes to a certified coder. Not a reviewer skimming for typos, a certified coder trained specifically in OASIS-E2 and home health coding rules. Their job is to look at each flag and make a real clinical judgment call.

Sometimes the AI is right and the coder confirms the correction. Sometimes the AI flagged something that actually has a reasonable clinical explanation buried in the notes, and the coder overrides it. That override matters just as much as the catch. It is a human applying context an algorithm cannot: knowing that a patient's functional decline lines up with a documented fall, or that a diagnosis code that looks unusual actually fits a rare but real clinical picture.

This is the step that keeps AI honest. It is also the step that keeps coders from burning hours on charts that are already clean, since the AI has already narrowed down where to look.

Why the combination beats either one alone

Certified coders working without AI support catch the vast majority of what matters, but people get tired, patterns get missed across large chart volumes, and turnaround slows down under a stack of recerts. AI working without a certified coder catches consistency issues fast, but it does not understand context, and it will confidently flag things that are actually fine, or worse, miss things that need a trained eye.

Put them together, and the AI does the tireless pattern matching across every chart at the same level of attention, while the coder brings the judgment that no model has. The result agencies actually feel is fewer missed errors, fewer downstream GG0130 or M-item mismatches showing up in payment reviews, and charts moving through review faster because coders are spending their time on judgment calls instead of hunting for problems from scratch.

That is the whole idea: pairing the best of AI with the best certified coders/reviewers. Not AI replacing coders, and not coders working alone with no support. A workflow where each one does the part they are actually good at.

Want to see it instead of just read about it?

The easiest way to understand this is to watch it happen on a real chart. Book a call and we will walk you through a sample chart review from flag to final audit trail, so you can see exactly where the AI catches something and exactly where a coder's judgment takes over.