Evaluation and management coding sits at an uncomfortable intersection: undercode the visit and the practice loses revenue it earned; overcode it and the claim becomes an audit target. Both failure modes trace back to the same root cause — medical decision making (MDM) documentation that doesn’t clearly support the level billed. As payers lean harder on algorithmic claim review and CMS keeps E/M accuracy on its enforcement radar, that documentation gap is no longer a minor administrative nuisance. It’s a measurable compliance and revenue problem, and agentic AI is starting to close it at the point of care rather than after the fact.
The Two-Sided Risk of E/M Level Selection
Since the 2021 E/M guideline overhaul, physicians can select a visit level based on either time or medical decision making, whichever is higher. In practice, most coding errors happen on the MDM side, because MDM scoring depends on three variables — the number and complexity of problems addressed, the amount and complexity of data reviewed, and the risk of complications or morbidity — that rarely get spelled out explicitly in the note. A physician might mentally weigh a moderate-complexity problem and order a confirmatory test, but if the note doesn’t document that reasoning, a coder has no defensible basis for billing the higher level. The safe default is to code down, which is why undercoding is estimated to cost practices real revenue every year. The opposite failure — billing a level the note doesn’t support — is what shows up in payer audit findings and OIG work plan items.
Where MDM Documentation Breaks Down
The problem isn’t a lack of coding knowledge. It’s that MDM support and multi-procedure sequencing both depend on details that live in the clinician’s head, not necessarily in the chart.
The MDM Documentation Gap
Coders can only bill what the note supports, and notes are written for clinical continuity, not claims defense. A visit that clinically warranted a moderate-complexity E/M level can get coded as straightforward simply because the note never states which data points were reviewed or why a particular risk determination was made. Multiply that across a busy practice’s daily visit volume and the aggregate revenue leakage — plus the inconsistent documentation that makes any single claim look arbitrary to an auditor — becomes significant.
CPT Sequencing Under MPPR
A second, less visible problem sits in procedure sequencing. Under Medicare’s Multiple Procedure Payment Reduction (MPPR) policy, every procedure billed after the first on the same claim typically reimburses at a reduced rate, so the order codes are listed in directly determines how much a practice collects. Getting that sequence wrong — or failing to check it against CMS’s Correct Coding Initiative (CCI) edits before submission — either strands reimbursement on the table or creates a code combination that shouldn’t have been billed together at all.
How Agentic AI Closes the Gap
Agentic AI tools are increasingly built to catch both problems before a claim leaves the building, not after a payer flags it. OpenEvidence, the clinical decision-support platform widely used by U.S. physicians, launched a feature called Coding Intelligence on March 26, 2026, aimed squarely at this gap. According to the company’s announcement, the tool generates ICD-10 diagnosis codes, E/M level recommendations, and CPT suggestions automatically at the moment a clinical note is finished, and it writes the supporting MDM rationale directly into the record rather than leaving it implied. It also applies CMS’s CCI rules to validate code combinations and sequences multiple procedures to reflect MPPR discounting, so the highest-value code appears first when that’s clinically accurate.
The significance isn’t the specific vendor — it’s the shift in where the check happens. Instead of a compliance team sampling claims after the fact, or a coder guessing at what the physician meant, the documentation and the code recommendation get generated together, with the reasoning attached. That closes the gap between what happened clinically and what the claim says happened, which is the exact gap both revenue-cycle teams and auditors care about.
What This Means for Compliance and Audit Programs
Compliance officers evaluating agentic coding tools for E/M and procedure-heavy specialties should look for a few specific capabilities rather than taking “AI-assisted coding” at face value:
- Documented MDM rationale attached to every E/M level recommendation, not just the level itself
- Automatic CCI edit checking before a claim is submitted, not as a post-submission audit step
- Procedure sequencing logic that accounts for MPPR discounting rather than defaulting to chart order
- An audit trail that shows what data supported each code, so a reviewer can reconstruct the decision months later
- Human sign-off retained on every claim, since documentation quality still varies by clinician and specialty
None of this replaces a compliance program’s own auditing function. If anything, it gives that function better material to work with — a documented rationale to test, instead of a bare code with no paper trail behind it.
Getting Ahead of the Documentation Gap
E/M and procedure coding accuracy has been a stated audit priority for years, and that scrutiny isn’t loosening. What’s changed is that the tools available to close the underlying documentation gap have moved from theoretical to deployed, with named products now generating MDM rationale and sequencing logic automatically rather than leaving it to memory. Practices that adopt this kind of validation earlier are the ones that will have a defensible answer the next time a payer or OIG auditor asks why a claim was coded the way it was.
Medikode’s automated medical coding platform applies this same documentation-to-code validation across E/M, HCC, and procedural coding, giving compliance teams an audit trail built in from the start rather than reconstructed after the fact. Learn more at Medikode’s automated medical coding platform.