Most practices judge their coding by what comes back. Claims go out, payments come in, and denials stay manageable, so the coding must be fine.
But a payment only tells you the claim was accepted. It doesn’t tell you the code matched the note.
Two mistakes that don’t look like mistakes
Undercoding gets paid. A provider manages several chronic conditions and adjusts a prescription, then picks a lower visit level out of habit. The payer pays it without a word. Nothing lands in a denial queue, so nobody goes back to look.
Overcoding gets paid too, until it doesn’t. A note lists several problems but doesn’t show what was decided about them, or copied-forward text makes a simple visit look complex. The claim still pays. The problem only surfaces if someone audits the chart.
So one error costs you money quietly, and the other builds up risk quietly. Your denial report won’t show you either one.
Missing codes are just as quiet
Sometimes the visit level is right and something is simply missing. At a one-provider primary care practice whose billing we run, the problem was an add-on code the visits already qualified for. Once it was being billed, it added about 140 codes a month and about $1,750 a month in payments, on average from January to June 2026.
Nobody had to see more patients for that. The care was already happening, and the claims just didn’t show it. That’s one practice’s result and not a promise, but it shows why “the claims are getting paid” is the wrong test.
Look both ways
If you only look for missed revenue, you’re running a revenue project, and an auditor won’t be impressed by a review that only ever corrected upward.
A useful review treats both directions the same way: read the note, apply the rule, and record why the code stands or changes.
Office visit levels are set by total time or by medical decision making. Decision making has three parts, and a visit has to meet two of them:
- Problems: how many were addressed, and how complex they were
- Data: what was reviewed or ordered
- Risk: what the treatment decisions involved, such as prescription drug management
The key word is addressed. A diagnosis that’s only listed in the note doesn’t count as managing it.
Run this check yourself
You can do a basic version without any software:
- Pull a random sample. Take a few dozen established patient visits from recent months. Don’t hand-pick them.
- Hide the billed code. Have a certified coder or a second person level each note first.
- Compare. Sort each visit into matched, coded too low or coded too high.
- Look for patterns. Break the results down by provider, visit type and payer.
- Compare your code mix. A big difference in visit levels between similar providers is a reason to read more notes. It isn’t proof of a problem on its own.
- Get advice on anything already paid. If a review suggests claims were overpaid, talk to your compliance advisor or attorney about repayment obligations.
Then do it again. One review shows you where you stand. A second shows whether your fix worked.
How we handle it
Reading notes is slow, and that’s why most offices rarely do this. Our coding pipeline drafts codes from the note, and each code carries the sentence that justifies it. Hard rules run in software instead of being left to the model.
A biller still makes every call. The software never submits a claim on its own. Its job is to make sure the evidence is in front of the person deciding.
General information, not legal, coding or compliance advice. Payer rules vary.