A new essay published in STAT by Darshak Sanghavi, a pediatric cardiologist and chief medical officer at Machinify, argues that the most consequential use of artificial intelligence in American healthcare is not clinical at all. It is the automated competition between hospitals and insurers to control how much a patient owes, a fight that helps explain why medical bills arrive so large and so contested before they ever reach a collection agency.
For ARM professionals working medical receivables, the piece illuminates the machinery upstream of every placement. Sanghavi describes how hospitals translate discharge records into billing codes, then use AI to identify secondary diagnoses, the complications and comorbidities that justify billing a patient as sicker and more expensive to treat. So-called ambient listening tools now record physician conversations and populate records automatically, surfacing diagnoses a busy clinician might never have logged.
The data he cites are pointed. Blue Cross Blue Shield analyzed tens of thousands of maternity admissions and found that at hospitals adopting AI coding tools fastest, the share of patients coded with serious post-delivery bleeding rose from roughly 4% to more than 12% between 2022 and early 2025, while transfusion rates stayed flat. At one hospital, fewer than 20% of coded cases met clinical criteria. BCBS estimated the nationwide effect could reach $2.3 billion. In Massachusetts, septicemia hospitalizations have more than tripled since 2010 without matching increases in length of stay, a pattern state officials attribute to coding behavior rather than population health.
On the other side, insurers deploy program integrity AI to flag suspected upcoding and deny claims automatically. Sanghavi calls it an arms race in which both systems learn and adapt continuously. The revenue cycle management industry was valued at roughly $65 billion in 2025, and health systems spend more than $140 billion a year on revenue cycle operations.
The consequence for patients, and eventually for the firms collecting on what they owe, is friction that compounds at every step. Denials, appeals, and disputed balances accumulate into debts that are harder to validate and harder to collect cleanly. Sanghavi proposes fusing provider and insurer AI into a single adjudication engine that issues one payment decision near the point of discharge, settling cost before a bill ever lands.
That vision faces real obstacles, including unprecedented data sharing between longtime adversaries and governance frameworks that do not yet exist. But for an industry that inherits the aftermath of these disputes, the essay is a useful reminder that the size and validity of a medical balance are increasingly determined by software long before a consumer is ever contacted.
