Essays · AI & the economics of getting paid

Downcoding: The Quiet Tax on a Doctor's Visit

By Zack Spooner, founder of Remend Health and an AI engineer by trade. Second in a short series on how AI is reshaping the economics of getting paid in American healthcare.

A denial is loud. The claim comes back rejected, someone notices, and at least in theory there is a fight to be had. Downcoding is quiet. The claim gets paid. It just gets paid less than it should, and the difference is small enough, per visit, that almost no one bothers to contest it.

That is exactly why it works.

What downcoding is

When a physician sees a patient, the visit is billed at a level of complexity, the familiar 99213, 99214, 99215 codes for established patients. A higher level reflects more clinical work: a complicated history, harder decisions, more time. The level is supposed to be determined by what is documented in the chart.

Downcoding is when the insurer overrides that. The doctor submits a level 4 visit, and the payer, without ever reading the note, decides it looks more like a level 3 and pays the lower rate. The visit happened. The work happened. The documentation supports the original code. The payer just decides to value it less, and shifts the burden onto the practice to appeal if they disagree.

This is not theoretical, and it is not rare. In September 2025, Cigna rolled out a policy, effective October 1, that allows it to automatically downcode level 4 and 5 office visits by one level when, in its view, the primary diagnosis does not justify the complexity. The American Academy of Family Physicians objected on a basic point of coding: a single diagnosis code cannot tell you the level of a visit, because complexity is determined by the full picture of medical decision-making and time, which lives in the note the insurer is not reading. Cigna paused the policy in California after the state medical association raised it with regulators, which tells you something about how it holds up under direct scrutiny.

Cigna is not alone. UnitedHealthcare has been adjusting submitted office-visit codes rather than denying them since 2019. The American Society of Clinical Oncology has named Aetna, UnitedHealthcare, Humana, Cigna, Anthem, and several Blue Cross plans as operating downcoding programs on outpatient and emergency visits. The American Medical Association has a standing position against automatic, algorithm-driven downcoding done without clinical review or adequate notice. When that many payers converge on the same mechanic, it is not a coincidence. It is a strategy.

Why the quiet version is the effective version

The genius of downcoding, if you can call it that, is that it is calibrated to sit just below the threshold where fighting back is worth it.

The math is simple and brutal. The difference between a level 4 and a level 3 visit might be twenty or thirty dollars. Appealing it requires staff time, documentation, a portal submission, and follow-up. For a busy practice, the rational choice on any single claim is to let it go. So they let it go. Multiply that small, rational surrender across every provider, every payer, and millions of visits, and the aggregate is enormous. The payer collects the spread between what was owed and what practices have the patience to chase.

This is the same asymmetry that drives outright denials, which I wrote about in the first piece in this series. Issuing the adjustment is automated and near-free. Contesting it is manual and costly. The economics only point one way.

And it rides on top of an administrative system that is already at its limit. The AMA's 2024 survey of physicians found that doctors complete an average of 43 prior authorizations per week and spend, with their staff, about 12 hours a week on them. Ninety-four percent reported that prior authorization delays care. Twenty-four percent said it had led to a serious adverse event for a patient. This is the environment downcoding lands in. The people who would have to appeal are the same people already drowning in the paperwork the payers generate. Asking them to also audit every paid claim for quiet underpayment is asking for labor that does not exist.

The part that should change

Here is what I find genuinely interesting about downcoding as a problem. Unlike a denial, where there is at least a clear no to push against, a downcoded claim looks settled. It is marked paid. To even know you were underpaid, you have to compare what you submitted against what you received, line by line, across thousands of claims, and then decide which gaps are worth pursuing. That is a data problem before it is a billing problem. It is precisely the kind of work humans do badly and at great expense, and that software does well and cheaply.

For most of the last decade, the asymmetry held because one side had automation and the other side had people. Payers could adjust at machine scale, and practices could only respond at human scale. That is starting to change, and it is the thing I am watching most closely. When the cost of detecting and contesting a quiet underpayment falls toward zero, the entire logic of downcoding, the bet that the spread is too small to chase, stops being safe.

The quiet tax works because it is quiet. The question is how long it stays that way.