Summary
AI medical coding uses software to read clinical documentation and propose billing and diagnosis codes such as CPT and ICD-10. It takes 2 main forms. Autonomous coding products, such as Fathom and CodaMetrix, work on large volumes of charts for health systems and physician groups, and Fathom reports a 95.5% automation rate and 98.3% accuracy at 1 named customer. Clinician-assisted coding suggests codes at the point of care for the clinician to confirm, which is how Wai works, with suggested coding prepared during the consultation and signed by the clinician. Evidence supports keeping a person in the loop. In a 2024 NEJM AI benchmark, general large language models asked to produce codes from descriptions scored below 50% exact matches, so a coding tool earns trust through its method and its review step. Ask each vendor how codes are grounded in the documentation, how results are measured, and who confirms each code before it reaches a claim. The signing clinician remains accountable for what is billed.
Each rival detail comes from that company's own page, read 9 November 2026. 12
See it in your clinicYou can be running this week. No procurement, no minimum term. A 7-day trial, then one published price.
Request a demo| Approach | Example | Best fit | Who confirms the code |
|---|---|---|---|
| Autonomous coding at scale | Fathom | Health systems, physician groups, outpatient clinics | Set by the customer's coding workflow |
| Autonomous coding for health systems | CodaMetrix | Health systems, including inpatient professional fees | CMX Automate codes with no human intervention, per its site |
| Clinician-assisted at the point of care | Wai | Private clinics | The clinician signs each suggested code |
The evidence on general models
A Mount Sinai team tested GPT-3.5, GPT-4, Gemini Pro and Llama2-70b Chat on producing codes from descriptions, drawing on 12 months of codes from its own records.3 GPT-4 led with exact matches of 45.9% for ICD-9-CM, 33.9% for ICD-10-CM and 49.8% for CPT.4
The task was a benchmark of code lookup, separate from coding a full note. Its lesson for buyers is that purpose-built method and human review carry the accuracy, and the clinical coding entry sets out the basics.
Autonomous and assisted coding
Fathom says it serves health systems, physician groups and outpatient clinics.1 CodaMetrix describes coding encounters with no human intervention through its CMX Automate product, for health systems.2 Both suit organisations with coding teams and high volume, and vendor figures are worth confirming on your own charts.
Smaller clinics usually gain more at the point of care, where the clinician knows what happened. Accountability for AI coding sets out who answers for each code.
Coding beside the signature
Wai prepares suggested coding during the consultation, alongside the note and letters, for the clinician to review and sign.5 Studio then carries insurance eligibility and claims on the same record.6
Capturing the full value of each visit and coding and revenue integrity show how to measure the result in your own clinic.
Bring a consultation you code often. See the suggested codes beside the signed note.
Request a demoFrequently asked questions
Can AI do medical coding?
How accurate is AI medical coding?
It varies by product and task. Fathom reports 98.3% accuracy at 1 customer, which is a vendor figure worth testing on your own charts.1
Will AI replace medical coders?
Coders and clinicians keep the judgement on complex cases and accountability for each claim, and AI handles more of the routine reading.
Who is responsible for a code suggested by AI?
The person signing it. On Wai, suggested coding is a draft for the clinician to check and sign.5
Does Wai code during the consultation?
Yes. Wai prepares suggested coding during the consultation alongside the note and letters.5
How should we test an AI coding tool?
Run 1 month of your own charts, compare against a certified coder's review, and record first-pass claim acceptance.
Sources
- Fathom home page, read 9 November 2026
- CodaMetrix home page, read 9 November 2026
- Large Language Models Are Poor Medical Coders, Benchmarking of Medical Code Querying, read 9 November 2026
- LLMs are not ready to automate clinical coding, says Mount Sinai study, read 9 November 2026
- Wai for clinicians, read 9 November 2026
- Pricing, read 9 November 2026
Sign codes you can stand behind
Bring one real clinic day and we will walk through it on your own workflows. We reply within 24 hours.
Partner with us
See it on your own patients, then build it with us
Thirty minutes on your own patients, doctors and rules, in your market. We reply within one working day, and the clinics that partner with us early shape what we build next.
See what Wai does between the visits