clinicians.build · interactive

Billing the Algorithm

The AMA's mid-year CPT release is where AI services get their first billable hook — Category III first, Category I if they survive. So which AI codes does Medicare actually pay? We checked 27 AI-adjacent CPT codes against seven years of Part B physician claims. One dot per code. Press play, hover the dots, then flip the small-n filter and watch most of the story dissolve.

Built on ↓ AAPC — "AMA Posts CPT Early Release Codes" (effective July 1, 2026) · companion to today's Money Plumbing
Also: AMA — CPT codes for AI-enabled health services
Data: Medicare Physician & Other Practitioners by Geography (FFS), CY2018–CY2024 · mimi_ws_1.datacmsgov.mupphy_geo via MIMI Labs
The whole market, in four numbers

"Billable AI" is a $10M line item

Add up every AI-analysis code that appears anywhere in Medicare's 2024 physician-claims data — FFR-CT, coronary plaque AI, autonomous retinal screening, quantitative MR, all of it — and you get 37,492 services and $10.0M paid. For scale: the remote-monitoring codes alone ran $259M standardized the same year. The AI billing lane exists. It is a bike lane.

$10.0M
total 2024 Part B payment, all AI-analysis codes that appear
12 / 25
AI/algorithmic codes checked that never clear the 11-patient floor
6 yrs
FFR-CT's Category III grind before Category I conversion
$3.53
what the algorithmic-ECG code earned per run — once
The explorer

Seven years of AI codes, one dot each

Across (log): services billed that year. Up (log): average Medicare payment per service. Dot size: total dollars. Navy = Category III (temporary, experimental lane) · red = Category I (made it) · gray = the human-read versions of the same retinal service, for comparison. Watch 2024: the FFR-CT family vanishes from navy and reappears as one big red dot — 75580, the conversion Money Plumbing is talking about.

AI-analysis CPT codes in Medicare Part B physician claims · services vs $/service
2024
–
codes visible
–
services in view
–
Medicare $ in view
–
median $/service
Category III Category I human-read comparator
Dotted trails show each code's path through prior years. Hover any dot for the full ledger. Try 2023 → 2024 to see the 75580 conversion; try the checkbox in any year to see how much of "AI billing growth" is codes with fewer than 100 patients.
The ghost shelf — checked, never seen

These 12 algorithmic/quantitative codes never appear in any year of national physician-claims data — meaning fewer than 11 traditional-Medicare patients nationally, every single year. Including the one today's newsletter prices out:

0691T · AI vertebral-fracture CT — the Money Plumbing example 0697T · quant. MR cholangiopancreatography 0698T · quant. MRCP add-on 0710T–0713T · arterial plaque quantification (4 codes) 0721T · quant. CT tissue characterization 0722T · quant. CT add-on 0740T · algorithmic insulin dosing 0741T · insulin dosing add-on 0765T · algorithmic ECG add-on
What this chart cannot see — and it's a lot. This is the physician/supplier slice of Part B, fee-for-service only. Hospital-outpatient billing (where a code like 0691T would most plausibly run), Medicare Advantage, and every commercial payer are all invisible here. CMS also deletes any row under 11 patients rather than reporting zero — so "absent" means "fewer than 11 nationally," not necessarily "never used." And the per-service averages mix professional, technical, and global claims, which is why 0501T's "price" swings from $59 to $837 across years without the fee schedule changing that way. Treat every number as a floor with a wide, unknowable ceiling — that's a data-transparency problem the AI-code economy inherits from Medicare itself.
What survives the small-n filter is the actual thesis. Flip the checkbox in 2024 and you're left with a handful of dots: 75580 (29,270 services, $248 avg — six years after its first 18 Category III claims), coronary plaque AI 0623T (2,092 services at $1,208 — real money, but resolving to just 8 visible states), autonomous retinal 92229 (5,113 services, price falling $32→$27 while volume grew 12x), and the human-read retinal codes still outbilling their AI replacement. Everything else — every "fast-growing" Category III code — is under 700 patients nationally.
The code existing is necessary. It is nowhere near sufficient. Between "the AMA gave it a number" and "Medicare pays it at volume" sits a six-year Category III valley where the national count of paying patients rounds to zero. If your pitch deck says "billable via CPT," this chart is the diligence question you'll get: show me your code's dot. 🔮 The 2026 mid-year AI codes just entered the valley. Most won't come out.
Read the source: AMA's mid-year CPT release → Or read today's newsletter →
clinicians.dev · an experiment by clinicians.build
Data: Medicare Physician & Other Practitioners by Geography & Service, national rows, extracts 2013-12-31–2024-12-31, mimi_ws_1.datacmsgov.mupphy_geo via MIMI Labs. Codes checked: 0501T–0504T, 75580, 0623T–0626T, 0648T/0649T, 0691T, 0697T/0698T, 0710T–0713T, 0721T/0722T, 0740T/0741T, 0764T/0765T, 92227–92229. Payment = actual Medicare paid (avg_mdcr_pymt_amt × services, summed facility+office), not standardized, not allowed charges. FFS only; physician/supplier claims only — hospital outpatient (OPPS) and Medicare Advantage excluded; CMS suppresses cells <11; per-service averages mix professional/technical/global. Ghost-code descriptors abbreviated from AMA Category III listings. Verified 2026-07-25.
⚠︎ AI-generated · not reviewed by a human · verify against the linked sources before relying on it