What DeepScribe Announced About Automated Visit Level Coding
DeepScribe said on 3 March 2025 that it added automated evaluation and management coding to its Ambient Operating System. The company says the software recommends the appropriate visit level code as part of the ambient scribing process, selects the code in real time from the conversation between patient and clinician, carries coding rules inside the AI, produces evidence backed documentation supporting the code it chose, shows quick reference pop ups explaining the parameters behind each code, and writes into records systems. DeepScribe says the product covers visit level codes, ICD-10 diagnosis codes and hierarchical condition categories. It cites Medicare data from 2022 putting visit level coding behind more than 1 billion dollars of errors, and a 2024 Medicare report attributing 564 million dollars of improper payments to a single code, 99214, with 63 percent improperly coded and 37 percent carrying insufficient or no documentation. The announcement also names Comprehensive Cancer Centers of Nevada and reports 90 percent clinician adoption there.
Why DeepScribe Automated Coding Matters to Practice and Revenue Leaders
The feature that counts is not the code, it is the receipt. A code with the supporting words from the note attached is defensible in an audit, and a code without one is an argument you lose. That is what makes this different from a coding tool bolted on afterwards: the software picking the code is the same software that heard the visit. The caution is the obvious one. Faster coding that leans high is a bigger problem than slow coding, and the figures DeepScribe cites describe how often coding goes wrong across Medicare, not how its own product performs. Nobody here reports an accuracy rate or a denial rate, and those are the only two numbers that settle it.
Where the DeepScribe Update Comes From
DeepScribe is the original record behind this update. It tells us what the company published. This brief adds the market context and the method we would use to test the development against other evidence.
How We Would Research DeepScribe Automated Coding
The source gives us the starting point. This is how we would build the next layer of research around it.
- We would ask DeepScribe for its coding accuracy against a certified human coder on the same set of visits, which is the only fair comparison.
- Then we would ask whether the suggested codes sit higher or lower than the practice has historically billed, because drift in either direction is a finding.
- We would ask what the denial rate did in the first six months at a named practice, since that is where wrong codes actually show up.