Our Take
Healthcare has solved data plumbing but not semantic agreement; without a shared framework for what clinical context means, AI will scale fragmentation faster than interoperability can fix it.
Why it matters
Payers, providers, and clinicians are each training AI against different local standards for the same patient conditions. The result: conflicting coding, denied claims, manual audits, and AI systems that reproduce inconsistency at scale. This matters now because documentation automation and prior authorization AI are moving into production without that alignment.
Do this week
Healthcare IT teams: audit your EHR documentation templates and coding workflows against a peer health system in your network to surface semantic gaps before deploying AI-assisted coding.
BlueCross Report Exposes the Coding Gap
BlueCross BlueShield Association identified $663 million in additional inpatient spending flagged to AI-driven coding intensity (company-reported). The industry response split predictably: are new technologies introducing errors, or finally capturing patient complexity providers previously omitted? The answer is upstream of both camps.
The real problem is not connectivity. APIs and standards already move data between systems at scale. The gap is semantic. A patient with diabetes and chronic kidney disease may be coded minimally to meet payer medical necessity in one health system, and coded with full specificity in another. Both are defensible. Neither reveals which one aligns with actual care or drives accurate reimbursement and research outcomes.
This variation starts before any data moves. Health systems configure EHRs with different documentation templates, order sets, and revenue cycle mappings. Providers document care narratively and contextually; payers abstract it for billing. By the time clinical data becomes a claim code, the same patient story produces semantically different outputs across systems. Even experienced, certified coders agree on accuracy only 50 percent of the time (per the article).
Automation Scales the Problem, Not the Solution
Interoperability alone cannot fix subjectivity. When documentation automation platforms and payer AI systems begin generating codes without longitudinal patient context, they will produce conflicting representations of the same condition across the health system. Each will technically be correct under local definitions. None will align.
This fractures trust. Payers suspect coding inflation. Providers defend specificity. Clinicians bear the burden of reconciling conflicting requirements across multiple payer systems. Denied claims that are later reversed, manual audits, and prior authorization delays persist not because data doesn't move, but because stakeholders cannot agree on what the data means.
The clock is ticking. Documentation AI, clinical workflow automation, and prior authorization bots are entering production now. Without a shared framework for how clinical context translates into compliant, accurate codes, these systems will embed fragmentation into the automation layer itself. The volume of data and speed of processing will only mask the misalignment longer.
Build Compliance Frameworks Before Deploying AI
The fix is not better pipes. It is a compliance layer above interoperability that establishes objective definitions of context and quality. This framework does not eliminate variation in how providers document care. It normalizes variation, ensuring that the same clinical narrative produces consistent codes across all use cases: billing, analytics, research, care decisions.
Such a framework acts as a compliance engine. Codes become more than billing artifacts; they become reliable representations of patient history that can move across systems with trust intact. Over time, this reduces audits, reversals, and friction in prior authorization.
Health systems should begin now by mapping their local coding standards against peers and payers, surface conflicts, and draft a single source of truth for how clinical specificity maps to code assignment. Documentation automation and AI-assisted coding vendors should be required to publish how their systems handle local variation before contracts are signed. Without that grounding, the automation will inherit and scale today's fragmentation.