AI for HCC risk adjustment coding under V28.
V28 changed the math. Roughly 2,300 ICD-10 codes were re-weighted and over 200 removed entirely from the risk model. Across the medical groups we audit, average RAF dropped 9.3 percent on the same patient panel. Our HCC AI is built for what comes next, not what worked under V24.
The condition map changed. The math followed.
CMS finished phasing in the V28 risk model this year. The change is not a tune-up. It is a structural shift in which conditions carry weight and which do not. Several high-volume V24 capture categories now contribute less, or nothing, to your RAF score. The chart shows where the weight moved across eight common HCC categories.
AI trained on V24 surfaces opportunities that no longer pay. The right HCC AI for 2026 carries the V28 condition map native, with a V24 comparison view for finance leadership reconciling prior-year capture against current.
The HCC Coding AI console.
The actual operator view your coders, CDI leads, and finance teams work inside every day. V28 condition map, MEAT enforcement, RAF lift attribution, and full LLM Gateway audit on every suggested code.
What moves the needle in HCC coding.
Generic AI medical billing tools rarely lift RAF. The six capabilities below are tuned for the operating reality of risk adjustment coding under V28. Each runs as part of our managed service, not as a tool you wire in alone.
V28-current suspect generation.
Our NLP reads chart documentation, problem lists, medication lists, recent labs, prior-year claims, and hospital discharge summaries. It surfaces suspected HCCs with the V28 condition map applied. Conditions that lost weight under V28 are de-prioritized so coder attention sits where the RAF actually lives. Outputs are ranked by combined RAF impact and documentation confidence so the coder works highest-value first.
- V28 condition map, refreshed quarterly
- Handles negation, uncertainty, history
- Suspect list ranked by RAF impact
- Hospital discharge data integrated
Recapture campaign management.
A suspect list without a workflow is just a more sophisticated way of missing the same revenue twice. Our model identifies attributed members whose chronic conditions were captured last year but not yet reaffirmed this year, then drives pre-visit prep, Annual Wellness Visit outreach, and clinician huddles ahead of those visits. Campaign progress is tracked at the panel level, with measurement at the patient, provider, and pod level.
- Patient-level recapture queue
- AWV scheduling integration
- Provider-level recapture scorecards
- Year-end glidepath dashboard
Clinical documentation review.
Every AI-surfaced code is paired with the documentation snippet that supports it, dated and signed. MEAT criteria (Monitored, Evaluated, Assessed, Treated) are extracted from the note and flagged when implied but not explicit. The coder sees a missing-MEAT prompt before the chart closes. Problem-list copy-forwards are explicitly rejected as standalone evidence. The output is documentation that reads as audit-ready before the encounter is locked.
- MEAT extraction per code
- Pre-close MEAT-gap alerts
- Re-affirmation enforcement per year
- No-copy-forward discipline
RAF lift attribution.
Leadership needs to know where lift came from. Our dashboard breaks RAF gains into four buckets: suspect-driven new captures, recapture-driven re-affirmations, gap closures from documentation prompts, and new diagnoses from clinical encounters. This lets the CFO defend the program to the board, the Chief Medical Officer evaluate clinician training impact, and operations attribute lift to the workflow change that produced it.
- Four-bucket lift attribution
- Provider and pod scorecards
- Trend tracking against baseline
- Executive monthly readout
RADV audit defense.
The CMS RADV final rule is now in force. Audit samples extrapolate across the full membership, so a 200-chart finding becomes a multi-million-dollar clawback. We log every AI-assisted capture: the source documentation, the MEAT signal, the date and clinician of record, the coder confirmation step, and the version of the AI model that suggested it. The result is an audit trail that defends each code on its own merits, end to end.
- Per-code RADV audit log
- MEAT evidence packaged
- Model version traceability
- Appeal-ready documentation
Pattern analytics across the panel.
Beyond per-patient coding, the model surfaces patterns. Under-coded condition categories. Comorbidities consistently missed by clinician or pod. Regional variations across multi-site groups. These patterns feed clinician education, CDI program design, and pre-visit huddle priorities. The result is a continuously improving documentation system, not a one-time bump.
- Cohort-level miss analysis
- Provider education feeds
- CDI roadmap recommendations
- Regional benchmarking
Three workflows. One operating system.
The highest-leverage HCC AI deployments do not bolt onto a single moment in the chart. They run across the encounter lifecycle, with the right action at the right time. Here is how ours sequences.
Pre-visit planning.
AI surfaces the suspect HCC list for the upcoming Annual Wellness Visit so the clinician walks in knowing which chronic conditions need re-affirmation, which suspects need clinical assessment, and which prior-year captures are at risk if the visit closes without follow-up. Pre-visit huddles become structured, not improvised.
Concurrent prompts.
As the note is being written, our model prompts the clinician on MEAT gaps in real time. The prompts appear inside the EHR chart workflow, not in a side application. A suggested HCC requires the clinician to confirm or decline before the encounter closes. This puts the decision where the clinical judgment lives, when the chart is most fresh.
Retrospective sweep.
At month-end, AI re-sweeps every chart that closed without confirming open suspects. Each one is queued for clinician review, addendum, or outreach for a follow-up visit before the calendar year closes. This is the safety net that catches what pre-visit and concurrent prompts miss, with full audit trail.
| Condition | M | E | A | T |
|---|---|---|---|---|
| CHF | ||||
| CKD 4 | ||||
| COPD | ||||
| DM2 w/comp | ||||
| MDD recurrent | ||||
| Asthma severe | ||||
| A-fib | ||||
| Cancer breast | ||||
| Vascular periph | ||||
| CKD 3 w/HTN |
MEAT, every chart, every chronic condition.
Documentation defensibility now matters more than capture rate. The CMS RADV audit final rule extrapolates findings across the full membership. A 12 percent error rate on a 200-chart sample becomes a multi-million-dollar clawback for a practice with 8,000 attributed lives. The right defense is not retroactive appeal. It is documentation that satisfies MEAT at the point of capture.
Our model extracts MEAT signals from progress notes, problem lists, medication lists, and labs. Every AI-suggested code carries the MEAT evidence with it. The heatmap shows what audit-ready coverage looks like in practice: green where MEAT is fully documented, amber where one element is implied but soft, red where coverage is missing entirely. Coverage is scored at the chart, provider, and pod level with month-over-month trend tracking.
What is your panel leaving on the table?
The math behind HCC capture is unforgiving and surprisingly simple. Every missed HCC is roughly $1,800 in plan revenue, and the average high-utilization MA patient carries four to seven missed conditions when documentation discipline lapses. Multiply that across your attributed panel and you are looking at the real cost of a coding workflow that has not been updated for V28.
Adjust the sliders to your panel size and current capture rate. The calculator shows annual revenue at risk and the lift opportunity if you reach the top quartile. These are the same numbers our senior partners walk through on a 30-minute free audit call.
Outcomes from well-implemented engagements.
Measured across mid-sized multi-specialty groups, ACOs, and IPAs running our HCC AI for at least 90 days. Anonymized; individual results depend on panel size, payer mix, and current documentation discipline.
Numbers are typical ranges from anonymized engagements. The free 30-day audit gives you a tailored projection based on your panel size, payer mix, and current capture rate, with the math attached.
Built by senior operators, not generic AI vendors.
Most HCC AI on the market is software you license and wire in yourself. Ours is delivered as a managed service with senior partners accountable for outcomes. The differences below are the reasons CFOs choose us over the alternatives.
Service, not software.
Certified coders validate AI output. A named senior partner is accountable for your RAF target. You do not buy a tool and then staff it yourself.
SLA-bound outcomes.
RAF lift and recapture rate are in writing in your contract. We get paid when your captured revenue actually closes, not on AI usage.
RADV-defensible by design.
Every code carries source documentation, MEAT evidence, timestamps, and model version. Appeal-ready out of the box, not a feature roadmap.
V28-current, V24-aware.
Both models live in the dashboard. Finance leadership reconciles prior-year RAF runs against V28 capture without spreadsheet gymnastics.
Frequently asked questions: AI for HCC coding.
How is V28 different from V24 and why does it matter for our HCC AI?
How do you keep AI-suggested codes audit-defensible against RADV?
What outcomes can we expect in the first 90 days?
Does this work for groups still on V24 or running V24 and V28 in parallel?
How does AI handle MEAT criteria specifically?
What is the workflow? Concurrent, retrospective, or pre-visit?
How does it integrate with our EHR?
Is this software we license or a service ASP-RCM runs?
What does the free 30-day audit include?
Bring your panel data. We bring the V28 math.
A free 30-day audit. We take a 90-day encounter sample, run it through our V28 and V24 models, and return a four-page written report on where your RAF is leaking. No sales deck, no SDR triage. A senior partner on the call.