Our Take
Health plans are naming a specific technology as a cost culprit, but the mechanism (whether AI coding inflates legitimate billing or simply surfaces underbilling) remains unexamined.
Why it matters
If AI documentation tools are genuinely enabling providers to code more thoroughly for services already delivered, cost inflation may reflect accuracy, not fraud. Regulators and insurers need to distinguish between the two before policy hardens around the technology.
Do this week
Healthcare compliance officers: request your vendor's audit logs showing which code selections AI recommends versus which providers manually override, so you can assess whether inflation tracks legitimate service capture or systematic upcoding.
The survey finding
Nearly 70% of health plans surveyed by PwC identified AI-powered documentation and coding tools as a top-three driver of commercial healthcare cost inflation in the coming year (per PwC). The survey reflects insurer concern that provider-side AI is contributing to rising claims costs, though the report does not specify the scale of the inflation attributed to these tools or the methodology underlying the ranking.
The missing distinction
Health plans frame AI coding as a cost problem, but the underlying mechanism is not clear from the available reporting. AI documentation tools can produce cost inflation in at least two structurally different ways: by enabling providers to code services more completely and accurately (capturing legitimate revenue for care already delivered), or by encouraging providers to code at higher severity levels or bill for services not rendered. The PwC survey does not distinguish between accuracy gains and upcoding.
If AI is simply making provider billing more thorough, the inflation reflects improved documentation practices, not fraud or market failure. If AI is systematically recommending higher-severity codes or redundant procedures, that is a different problem entirely, one that requires auditing of model recommendations and provider override patterns.
Regulators and payers alike are watching this category closely. CMS and private insurers have begun scrutinizing AI-assisted coding in audits, but without transparency into how these tools recommend codes and how often providers accept or reject those recommendations, the cost signal alone is not actionable intelligence.
What to do now
Health plan compliance teams should request detailed audit reports from their network providers showing AI recommendation acceptance rates, override rates, and the severity distribution of AI-recommended versus manually selected codes. Providers deploying these tools should pre-emptively log recommendation patterns and comparative coding behavior before and after adoption. This evidence will be critical if regulators or auditors move to restrict AI coding tools or demand refunds for claims believed to stem from systematic upcoding.