Effective date: July 26, 2026
Coda uses artificial intelligence to help hospital teams produce accurate, well-documented billing from surgical documentation. This page explains what the AI does, what it cannot do, and the safeguards around it.
It analyzes an operative note and proposes billing codes, with cited reasoning and the supporting evidence behind each proposal.
It identifies documentation that is incomplete and asks the physician targeted questions to complete it.
It answers coders' and billers' questions about a specific case, grounded in that case and in coding references.
Every AI output in Coda is a proposal awaiting human review. Physicians confirm documentation, coders and billers confirm codes, and nothing is submitted to any billing system autonomously. There is no fully automated decision-making in the product.
Customer data, including protected health information, is never used to train, fine-tune, or improve any AI model, ours or anyone else's. AI processing operates under agreements requiring zero data retention: the model providers do not store inputs or outputs after responding.
By default, AI processing runs inside the hospital's own dedicated cloud environment, covered by our cloud provider's Business Associate Agreement. A secondary model path is covered by its own Business Associate Agreement with zero data retention. AI runs only when a user starts it on a specific case; there is no background scanning of records.
Audited like a person. Every AI access to clinical information is recorded in the same tamper-evident audit trail as human access, tagged as AI activity, and visible to the hospital's compliance staff.
Traceable decisions. Each analysis retains its inputs, steps, and outputs, so any proposal can be traced back to its reasoning after the fact.
Least privilege. The AI can use only a short, explicitly approved set of tools, scoped to the case at hand. It has no open-ended system or network access.
Grounded against fabrication. Proposals are grounded in the documentation itself and in exact-match coding reference data. The product's stance is clean, defensible documentation, never aggressive coding.
Quality gates. Changes to the AI must pass a scored benchmark before release, and production output is sampled by an automated review stage.
Off switches. AI features can be disabled for an individual hospital, and platform-level controls can stop AI processing within minutes without a software release.
AI systems can be wrong. That is precisely why Coda is built around human review: the AI drafts and cites, and people decide. Where the AI is uncertain, or the documentation is incomplete, the product raises a question rather than guessing.
Questions about Coda's AI: info@unira.io
We post updates here with a new effective date and notify hospital customers of material changes.