The Story
The Blue Cross Blue Shield Association has estimated that hospitals' use of AI-assisted medical coding contributed around $942 million in additional costs to its health plans between 2023 and 2025.
The analysis found a sharp rise in patients being documented with complex conditions without evidence of a corresponding change in the care delivered. Roughly $653 million, about 70 per cent of the billing identified, was tied to additional diagnoses unaccompanied by any change in treatment, Luke Chalker, BCBSA's senior vice president of product and data science, told CNBC. He stopped short of attributing the whole increase to AI.
The association notes that the growth in what it calls complex coding came during a period in which 60 per cent of hospital systems began using AI coding tools. Much of the increase came from secondary conditions, those arising from an illness separate to the one being treated.
The American Hospital Association has pushed back, saying the argument is not supported by evidence.
Coding vendors have rejected the framing. Hamid Tabatabaie, chief executive of Codametrix, which automates coding for more than 500 hospitals and health systems, does not dispute that costs rose but objects to the blame being placed on AI, arguing the systems secure payment providers are already owed under existing reimbursement rules.
Shiv Rao, founder of the AI company Abridge, has described a possible future of bots fighting bots and agents fighting agents, while suggesting the technology could also reduce friction and cost.
Why It Matters
The dispute cannot be settled on the facts both sides agree on, which is what makes it worth understanding.
BCBSA says documentation of complex conditions rose without any corresponding change in treatment. Coding vendors do not dispute the rise. Their explanation is that hospitals have historically under-coded, failing to bill for conditions they legitimately treated because a human coder reading a long chart misses things a model does not. On that reading, $942 million is not inflation; it is the correction of years of money left on the table.
Both accounts fit the same evidence. If a patient's anaemia was always present and always treated but rarely coded, then billing for it now is accurate and also more expensive. The insurer sees costs rising for identical care. The hospital sees itself finally being paid properly for care it was already delivering.
Resolving that requires knowing whether the underlying coding rules were being applied correctly before, and neither party has an incentive to establish it.
What is not in dispute is where this goes. Insurers are deploying AI to review claims, providers are deploying it to contest denials, and the technology is expensive on both sides. Abridge's founder named the destination as bots fighting bots.
Every rupee or dollar spent on that contest is administrative. It buys no treatment, and it is recovered from premiums at one end and charges at the other.
Luke Chalker, senior vice president of product and data science at BCBSA, told CNBC that roughly $653 million of the identified billing was tied to additional diagnoses that were not accompanied by a change in care. He stopped short of attributing the entire increase to AI.
The Strategic Read
India is building towards exactly this configuration, and has a narrow window to design around it.
IRDAI's proposed Public Insurance Registry would make policy and claims data portable across insurers, and its consultation closed on 30 September. Health is the fastest growing non-life segment here, with standalone health insurers up 27.1 per cent year on year in September. Hospitals are consolidating, insurers are automating, and the National Health Claims Exchange is standardising the interface between them.
Standardised data is what makes automated coding possible on both sides. Once claims move through a common protocol, the vendor who can extract maximum legitimate reimbursement from it has a product, and so does the vendor who can contest it.
The American experience suggests the administrative layer grows regardless of who is winning. Neither side's spending improves care, and both recover it through premiums or charges.
The design question worth asking while the rails are still being built is whether the standard itself should encode coding rules tightly enough to remove most of the discretion both sides are now automating. A protocol that permits broad interpretation invites an arms race. A narrower one is harder to agree and leaves less room for either party to optimise.
India has the advantage of watching this happen elsewhere first. It is a rare position for a regulator, and the registry consultation was the moment to use it.
For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.