HCC Risk Adjustment and AI: How Accurate Coding Drives Capit...

A single missed HCC code on a Medicare Advantage patient can cost an organization $8,000 to $25,000 in annual capitated revenue. Multiply that across a panel of 10,000 attributed lives, and a 10% gap in HCC capture translates to $15 million or more in revenue left on the table every year — not because care wasn't delivered, but because documentation didn't reflect the conditions that were actually treated.

Risk adjustment under capitated contracts is not a coding exercise. It is the revenue model. In fee-for-service, revenue flows from claims. Under capitation, revenue flows from the documented acuity of your patient population. The more completely and accurately you document the chronic conditions, complications, and comorbidities your patients actually have, the higher your risk-adjusted capitation rate — and the more resources you have to care for those patients.

The problem is that risk adjustment coding is extraordinarily difficult to do well at scale. Conditions must be re-documented every calendar year. Documentation must meet specificity thresholds that go beyond routine clinical needs. And the gap between what physicians know about their patients and what ends up in the medical record is vast.

This guide covers how HCC risk adjustment works, why accuracy matters at the dollar level, where organizations lose revenue, and how AI is closing the gap between documented acuity and actual acuity.

How Risk Adjustment Works in Medicare Advantage and Commercial Capitation

Risk adjustment exists to solve a fundamental problem in capitated payment: patients are not equally expensive to care for. Without risk adjustment, health plans and provider organizations bearing capitated risk would be financially penalized for enrolling sicker patients and rewarded for avoiding them.

The CMS-HCC Model

The Centers for Medicare and Medicaid Services uses the Hierarchical Condition Category (HCC) model to risk-adjust payments to Medicare Advantage plans. The model works in three steps:

Diagnosis collection: ICD-10-CM diagnosis codes from all face-to-face encounters during the data collection period (the prior calendar year) are submitted to CMS.
HCC mapping: ICD-10-CM codes are mapped to HCC categories. Not every diagnosis code maps to an HCC — only conditions that predictably increase future healthcare costs. Approximately 9,500 ICD-10-CM codes map to roughly 86 HCC categories in the CMS-HCC V28 model.
RAF score calculation: Each HCC category carries a coefficient (a relative cost weight). These coefficients are combined with demographic factors (age, sex, Medicaid eligibility, institutional status) to produce a Risk Adjustment Factor (RAF) score for each beneficiary.

The RAF score directly multiplies the base capitation rate. A beneficiary with a RAF score of 1.0 generates the average per-member payment. A beneficiary with a RAF score of 2.0 generates double the average payment. A beneficiary with a RAF score of 0.5 generates half.

Commercial Risk Adjustment

Commercial payers increasingly use risk adjustment models for capitated and value-based contracts, though the specific models vary:

HHS-HCC model: Used for ACA marketplace risk adjustment, with its own set of condition categories and coefficients
Proprietary models: Major payers including UnitedHealthcare, Aetna, and Anthem use proprietary risk adjustment models for commercial capitation contracts
DxCG and ACG models: Third-party risk adjustment models used in various commercial arrangements

The principles are identical across all models: documented conditions drive risk scores, risk scores drive payment, and underdocumented conditions mean underpayment.

The RAF Score: What It Is and How It's Calculated

Understanding the mechanics of RAF score calculation is essential for understanding where revenue leakage occurs.

Demographic Baseline

Every Medicare Advantage beneficiary starts with a demographic baseline score based on:

Factor	Example Impact on RAF
Age band (65-69, 70-74, etc.)	+0.09 to +0.60 per band
Sex	Male baseline slightly higher
Medicaid dual-eligible status	+0.08 to +0.19
Institutional status	Significant increase for LTI beneficiaries
Originally disabled	+0.03 to +0.12

A community-dwelling 72-year-old male with no Medicaid eligibility might have a demographic baseline RAF of approximately 0.35.

HCC Condition Coefficients

Each documented HCC category adds its coefficient to the demographic baseline. Under the CMS-HCC V28 model (phased in 2024-2026), representative coefficients include:

HCC Category	Example Conditions	Approximate Coefficient
HCC 18: Diabetes with chronic complications	Diabetic nephropathy, retinopathy, neuropathy	0.302
HCC 85: Congestive heart failure	Systolic, diastolic, combined heart failure	0.331
HCC 96: Specified heart arrhythmias	Atrial fibrillation, ventricular tachycardia	0.257
HCC 111: COPD	Chronic obstructive pulmonary disease	0.276
HCC 48: Coagulation defects and other specified hematological disorders	Hypercoagulable states, thrombocytopenia	0.194
HCC 22: Morbid obesity	BMI 40+	0.250
HCC 35: Major depressive disorder	Severe, recurrent major depression	0.155
HCC 115: Aspiration and specified bacterial pneumonias	Aspiration pneumonia, Klebsiella pneumonia	0.451
HCC 138: Chronic kidney disease, Stage 4	CKD Stage 4	0.237

Interaction Terms

The CMS-HCC model includes interaction terms that add additional coefficients when certain condition combinations are present. For example:

CHF + COPD interaction: Adds an additional coefficient beyond the individual HCC values
Diabetes + CHF interaction: Recognized as a higher-cost combination than either condition alone
Multiple high-severity conditions: Disease count interactions add incremental value when five or more payment HCCs are present

RAF Score Example

Consider a 74-year-old female Medicare Advantage member with diabetes with chronic complications, congestive heart failure, atrial fibrillation, and COPD:

Component	Coefficient
Demographic baseline (74F, community)	0.39
HCC 18: Diabetes with chronic complications	0.302
HCC 85: Congestive heart failure	0.331
HCC 96: Specified heart arrhythmias	0.257
HCC 111: COPD	0.276
CHF + COPD interaction	0.064
Diabetes + CHF interaction	0.058
Disease count interaction (4 payment HCCs)	0.039
Total RAF score	1.717

With a 2026 Medicare Advantage base rate of approximately $11,600 per member per year (national average), this member generates roughly $19,917 in annual capitation revenue. If the CHF and COPD are not documented in the current calendar year — even though the patient still has both conditions — the RAF score drops to approximately 0.949, and capitation revenue falls to $11,008. That is an $8,909 annual revenue loss for a single member due to two missed HCC categories.

Why Risk Adjustment Accuracy Matters: Revenue Impact Per Point

The financial stakes of HCC capture accuracy become clear when you model the revenue impact across a patient population.

Revenue per RAF Point

At the national average Medicare Advantage base rate:

Base Rate (2026 est.)	RAF Score Change	Annual Revenue Impact per Member
$11,600	+0.10	+$1,160
$11,600	+0.25	+$2,900
$11,600	+0.50	+$5,800
$11,600	+1.00	+$11,600

For organizations in higher-cost counties (parts of Florida, New York, California, Texas), the base rate can exceed $14,000, making each RAF point worth proportionally more.

Population-Level Revenue Impact

The aggregate impact scales dramatically with panel size:

Panel Size	HCC Capture Improvement	Average RAF Increase	Annual Revenue Impact
5,000 members	+5% of capturable HCCs	+0.08	+$4.6M
10,000 members	+5% of capturable HCCs	+0.08	+$9.3M
25,000 members	+5% of capturable HCCs	+0.08	+$23.2M
50,000 members	+5% of capturable HCCs	+0.08	+$46.4M

Even modest improvements in HCC capture accuracy — moving from 70% to 75% of conditions accurately documented — can represent tens of millions of dollars in annual revenue for mid-to-large Medicare Advantage organizations.

The Compounding Effect

Risk adjustment revenue impacts compound in multiple ways:

Year-over-year: CMS uses prior-year diagnoses to set current-year payment, so a gap in one year creates a revenue shortfall in the next
Benchmark setting: In shared savings programs (MSSP, commercial ACOs), risk adjustment accuracy sets the cost benchmark. Underdocumented acuity produces an artificially low benchmark, making it harder to generate shared savings
Contract negotiations: Payers use historical RAF scores in capitation rate negotiations. Chronically underdocumented panels produce chronically underfunded contracts

Common Risk Adjustment Problems

Despite the clear revenue implications, most organizations significantly underperform on HCC capture. Industry data consistently shows that 15-25% of capturable HCC conditions go undocumented in any given measurement year.

Problem 1: Undercoding of Chronic Conditions

Physicians document for clinical purposes, not revenue cycle purposes. A cardiologist managing a stable heart failure patient may focus the encounter note on medication titration and symptom assessment without explicitly restating the diagnosis of systolic heart failure with sufficient specificity for HCC mapping.

Common undercoding scenarios:

Diabetes documented as "diabetes" without complication specificity. ICD-10 code E11.9 (Type 2 diabetes without complications) does not map to any HCC. E11.65 (Type 2 diabetes with hyperglycemia) maps to HCC 19 (Diabetes without complication) — a lower-value HCC. E11.22 (Type 2 diabetes with diabetic chronic kidney disease) maps to HCC 18 — a significantly higher-value category.
Heart failure documented without type specification. I50.9 (Heart failure, unspecified) maps to HCC 85 but may trigger coding queries. Documentation of "systolic heart failure" or "diastolic heart failure" with acuity (chronic, acute, acute-on-chronic) supports the most specific mapping.
CKD documented without stage. N18.9 (Chronic kidney disease, unspecified) maps to a lower-value HCC than N18.4 (CKD, Stage 4), which maps to HCC 138 with a coefficient of approximately 0.237.

Problem 2: Documentation Specificity Gaps

The CMS-HCC model rewards specificity. A diagnosis documented at a high level of generality may fail to map to an HCC entirely, or may map to a lower-value category than the patient's actual condition warrants.

Specificity gap examples:

Documentation	ICD-10 Code	HCC Mapping	Revenue Impact
"Malnutrition"	E46 (unspecified)	HCC 21 (if mapped)	Lower coefficient
"Severe protein-calorie malnutrition"	E43	HCC 21: Protein-calorie malnutrition	Full coefficient (~0.545)
"Lung disease"	J98.4	No HCC mapping	$0
"COPD with acute exacerbation"	J44.1	HCC 111	~$3,200/year
"Depression"	F32.9	May not map to payment HCC	Lower or $0
"Major depressive disorder, recurrent, severe"	F33.2	HCC 155	~$1,800/year

Problem 3: Missing Conditions Entirely

Many patients have chronic conditions documented in their problem list or historical records that are never addressed or re-documented during the measurement year. This is distinct from undercoding — the physician simply doesn't mention the condition in any encounter during the year.

Commonly missed conditions:

Secondary diagnoses during specialist visits (the cardiologist doesn't document the patient's COPD)
Stable conditions that don't require active management changes
Conditions managed by other providers in the patient's care team
Mental health conditions in primary care settings
Obesity, which is frequently present but not documented as a diagnosis

Problem 4: Coding Staff Limitations

Even when documentation supports accurate HCC coding, operational constraints limit capture:

Volume pressure: Coders processing 20+ charts per hour cannot perform the deep documentation review needed for complete HCC capture
HCC training gaps: Not all coders are trained in risk adjustment coding, which requires different skills than fee-for-service coding
Specialty knowledge: Accurately coding HCCs requires understanding the clinical context — knowing that a patient on insulin with an eGFR of 28 likely has diabetic nephropathy, even if the physician didn't explicitly state the connection
Retrospective timing: By the time coding reviews happen, the encounter is over, and querying the physician for documentation clarification introduces delays

The Annual Recapture Challenge

Risk adjustment has a feature that no other revenue cycle function shares: every HCC condition must be re-documented every calendar year. A patient who was coded with HCC 85 (congestive heart failure) in 2025 must have that condition documented again from a face-to-face encounter in 2026, or it drops off the RAF score entirely.

The Recapture Problem by the Numbers

Metric	Typical Performance
HCC conditions requiring annual recapture	100% of prior-year HCCs
Average recapture rate without intervention	65-75%
Revenue at risk from failed recapture (10,000 MA members)	$12M-$20M annually
Conditions most frequently missed in recapture	Stable chronic conditions, secondary diagnoses, mental health conditions

Why Recapture Fails

Calendar year reset: A patient with CHF, COPD, and diabetes seen in October 2025 who isn't seen again until March 2026 may have a three-month gap where those conditions could be missed if the encounter in March focuses on an acute complaint.

Visit pattern misalignment: Patients who are seen frequently may have their chronic conditions documented at routine visits. Patients who are seen rarely — or who are seen primarily by specialists rather than primary care — may not have annual wellness visits or comprehensive encounters where all chronic conditions are addressed.

Provider rotation: In large groups, patients may see different providers. The new provider may not be aware of all documented HCCs from prior years, particularly conditions managed by other specialists.

Documentation fatigue: Physicians who see the same patient repeatedly may not restate known chronic conditions because they feel redundant. From a clinical perspective, they are. From a risk adjustment perspective, that omission costs thousands of dollars.

The Recapture Timeline

CMS collects diagnoses on a calendar-year basis, but the payment impact is lagged:

January - December 2025: Diagnosis data collection period
Early 2026: CMS processes 2025 encounter data
Mid 2026: Risk scores calculated from 2025 data applied to 2027 payment rates
January 2027: Capitation payments reflect 2025 diagnosis data

This lag means that a missed HCC in 2025 doesn't show up as a revenue shortfall until 2027 — making it difficult for organizations to connect documentation gaps to financial outcomes in real time.

Chart Review vs. Prospective Coding: Two Approaches to HCC Capture

Organizations use two fundamentally different strategies for HCC capture, each with distinct advantages and limitations.

Retrospective Chart Review

How it works: After encounters occur, trained HCC coders or clinical reviewers audit charts to identify documented conditions that were not coded as HCCs. When conditions are identified, coders add the HCC codes and submit them on retrospective claims or encounter data corrections.

Advantages:

Captures conditions that coders missed during initial coding
Allows dedicated HCC specialists to review documentation in depth
Can target high-value conditions and high-risk patients
Familiar workflow for organizations with established coding operations

Limitations:

Cost: Retrospective review typically costs $15-$40 per chart reviewed. Reviewing 50,000 charts per year at $25 each represents a $1.25 million annual investment.
Timing: Reviews happen weeks or months after the encounter, creating lag between care delivery and HCC capture
Documentation ceiling: If the physician's documentation doesn't support the HCC, retrospective review cannot fix it. The condition may exist, but without adequate documentation, it cannot be coded.
Scalability: Requires trained HCC reviewers, who are scarce and expensive. Many organizations outsource retrospective review to vendors, adding cost and reducing internal visibility.
Declining returns: As chart review programs mature, each additional pass yields fewer incremental HCC captures. The easy wins get found first; the remaining gaps require increasingly specialized review.

Prospective Coding (Pre-Visit and Point-of-Care)

How it works: Before or during the patient encounter, the provider is presented with a list of HCC conditions that were documented in prior years but have not yet been recaptured in the current year. The provider is prompted to assess each condition and document it if still clinically active.

Advantages:

Addresses the documentation problem at its source — during the encounter
Closes recapture gaps in real time rather than retrospectively
Enables providers to document conditions with full clinical specificity
Generally higher capture rates than retrospective review alone

Limitations:

Provider burden: Presenting physicians with long lists of conditions to re-document can feel like administrative work rather than clinical care
Alert fatigue: If the system flags too many conditions, providers begin ignoring the prompts
Workflow integration: Prospective prompts must be integrated into the EHR workflow without disrupting clinical operations
Clinical judgment tension: Providers may resist documenting conditions they don't consider clinically relevant to the current visit

The Optimal Approach: Both

The highest-performing risk adjustment programs combine prospective and retrospective strategies:

Pre-visit: Identify recapture gaps and present them to providers before or during the encounter
Point-of-care: AI-powered documentation tools ensure that discussed conditions are documented with HCC-appropriate specificity
Post-visit: Retrospective review catches any remaining gaps that prospective approaches missed

Organizations running both strategies typically achieve 85-92% HCC capture rates, compared to 65-75% with retrospective-only and 75-85% with prospective-only approaches.

How AI Improves Risk Adjustment Accuracy

AI transforms HCC capture from a labor-intensive, retrospective exercise into a real-time, integrated capability. The technology addresses risk adjustment gaps at multiple points in the revenue cycle.

NLP-Powered Documentation Mining

Natural language processing analyzes clinical documentation — physician notes, specialist consultations, hospital discharge summaries, lab results — to identify conditions that are present in the documentation but were not coded.

What AI NLP detects that human reviewers often miss:

Implied conditions: A medication list that includes metformin, lisinopril, and atorvastatin alongside lab results showing eGFR of 42 and A1c of 8.2% strongly implies Type 2 diabetes with chronic kidney disease — a higher-value HCC than diabetes alone. AI connects these clinical data points across the chart.
Conditions documented in narrative text: A physician note that states "her chronic systolic heart failure remains stable on current medications" contains an HCC-mappable diagnosis that may not appear in the structured problem list or diagnosis fields.
Specificity indicators: Documentation that references "diabetic retinopathy" in an ophthalmology note linked to a patient coded only with E11.9 (diabetes without complications) identifies a specificity upgrade opportunity.
Cross-encounter patterns: A patient with multiple encounters documenting "shortness of breath," "lower extremity edema," and "BNP elevated at 890" across different providers may have undiagnosed or underdocumented heart failure.

Automated Gap Identification

AI systems compare documented HCC conditions against historical patterns and expected prevalence to identify probable gaps:

Recapture gap lists: Conditions documented in prior years that have not yet been recaptured in the current year, prioritized by RAF coefficient value
Suspect conditions: Conditions that clinical data (labs, medications, vitals, imaging) suggest are present but have never been formally diagnosed and coded
Specificity upgrades: Conditions currently coded at a general level that documentation supports coding at a more specific, higher-value level
Missing interaction opportunities: Patient profiles where documenting one additional condition would trigger an HCC interaction term, adding incremental RAF value

Real-Time Documentation Support

AI-powered clinical documentation tools — like ambient AI scribes — improve risk adjustment accuracy at the point of care by ensuring that clinical conversations are fully captured in the medical record.

How this works in practice:

A physician sees a Medicare Advantage patient for a routine follow-up. During the visit, the physician discusses:

Current diabetes management (adjusting insulin dosage)
Recent ophthalmology report showing nonproliferative diabetic retinopathy
Ongoing depression management (continuing sertraline)
Weight management (BMI 42)

Without AI documentation support, the physician might generate a note focused on the diabetes medication adjustment, potentially omitting the retinopathy discussion, the depression management, and the obesity diagnosis. The encounter captures one HCC instead of four.

With AI documentation support, the system captures the full conversation, generates documentation that includes all four conditions with appropriate specificity, and flags any conditions that need additional documentation detail to support HCC coding:

Condition	Without AI Documentation	With AI Documentation
Diabetes with complications	E11.65 (HCC 19: ~0.105)	E11.311 (HCC 18: ~0.302)
Diabetic retinopathy	Not documented	E11.319 (captured via HCC 18 hierarchy)
Major depression	Not documented	F33.1 (HCC 155: ~0.155)
Morbid obesity	Not documented	E66.01 (HCC 22: ~0.250)
Total incremental RAF	0.105	0.707
Annual revenue impact	$1,218	$8,201

The revenue difference from a single encounter: $6,983. Extrapolate that across hundreds or thousands of encounters, and the financial impact of AI-supported documentation on risk adjustment becomes transformative.

Coding Accuracy at Scale

AI coding engines process encounter documentation and assign HCC codes with consistency and specificity that manual coding cannot match at volume. Key capabilities include:

99%+ code mapping accuracy: AI matches documented conditions to the most specific ICD-10 code supported by documentation, ensuring maximum HCC capture
Hierarchical logic enforcement: The CMS-HCC model applies hierarchical rules (e.g., diabetes with complications supersedes diabetes without complications). AI automatically applies these hierarchies, preventing duplicative or conflicting coding
Guideline adherence: AI applies official coding guidelines consistently — no variation based on coder experience, fatigue, or training gaps
Throughput: AI processes charts in seconds rather than the 15-30 minutes required for manual HCC review, enabling comprehensive review of every encounter rather than sampling

Compliance Guardrails: Avoiding Upcoding in Risk Adjustment

Risk adjustment accuracy must flow in both directions. While undercoding leaves revenue on the table, overcoding — documenting or coding conditions not supported by clinical evidence — creates severe compliance risk.

The Regulatory Environment

CMS and the Department of Justice have intensified risk adjustment enforcement:

False Claims Act liability: Knowingly submitting unsupported HCC codes can trigger False Claims Act violations with treble damages and per-claim penalties
RADV audits: CMS conducts Risk Adjustment Data Validation audits, reviewing medical records to verify that submitted HCC codes are supported by documentation. Unsupported codes trigger payment recovery and potential extrapolation
OIG focus: The Office of Inspector General has identified risk adjustment as a priority enforcement area, with multiple ongoing investigations and settlements
2024-2026 enforcement escalation: CMS has expanded RADV audit scope and implemented chart-level validation requirements that increase documentation scrutiny

Common Compliance Risks in Risk Adjustment

Risk Area	Example	Consequence
Unsupported diagnoses	Coding HCC conditions based on historical data without current-year clinical documentation	RADV recovery, False Claims Act exposure
Inappropriate additions	Adding diagnoses during retrospective review that are not supported by the encounter documentation	Per-claim FCA penalties
Specificity inflation	Coding a more specific (higher-value) HCC than documentation supports	Audit recovery, compliance sanctions
One-way review	Chart review programs that only look for missed HCCs without also identifying overcoded conditions	Systematic upcoding pattern
Cloned documentation	Copy-forward of problem lists without clinical reassessment	RADV audit failures

How AI Maintains Compliance

Well-designed AI systems build compliance guardrails directly into the risk adjustment workflow:

Documentation linkage: Every suggested HCC code is linked to the specific documentation that supports it. Reviewers can verify the clinical basis for each code in seconds.
Bidirectional review: AI identifies both undercoded conditions (revenue opportunities) and overcoded conditions (compliance risks), ensuring accuracy in both directions
Audit trail: Every AI-generated code suggestion includes the documentation source, the mapping logic, the confidence level, and whether a human reviewer accepted or modified the suggestion
Confidence scoring: AI assigns confidence levels to each HCC suggestion. Low-confidence suggestions are routed to human reviewers rather than auto-coded, preventing speculative coding
Regulatory rule updates: AI systems incorporate CMS-HCC model updates, V28 transition rules, and RADV audit criteria, ensuring compliance with current requirements rather than outdated guidelines

The compliance principle is straightforward: AI should help organizations code everything that is documented — and only what is documented. It should never generate codes that documentation does not support, regardless of how likely a condition may be based on clinical inference.

ROI Model: Revenue Impact of Improved HCC Capture Rates

The following model demonstrates the financial impact of AI-improved risk adjustment for a representative Medicare Advantage organization.

Baseline Assumptions

Parameter	Value
Attributed MA lives	15,000
Average current RAF score	1.05
Estimated true RAF score (based on actual acuity)	1.18
RAF gap (underdocumented acuity)	0.13
MA base rate (national average, 2026)	$11,600
Current HCC capture rate	72%
Retrospective chart review cost per chart	$28
Annual retrospective reviews conducted	12,000

Current State Revenue Leakage

Annual revenue at current RAF (1.05): 15,000 x $11,600 x 1.05 = $182.7M

Annual revenue at true RAF (1.18): 15,000 x $11,600 x 1.18 = $205.3M

Annual revenue leakage from RAF gap: $22.6M

Cost of current retrospective review program: 12,000 x $28 = $336,000

AI-Enhanced Risk Adjustment Performance

Metric	Before AI	After AI	Change
HCC capture rate	72%	91%	+19 percentage points
Average RAF score	1.05	1.155	+0.105
Conditions recaptured annually	65%	89%	+24 percentage points
Time per chart review (retrospective)	22 minutes	4 minutes	-82%
Charts requiring manual review	100%	28%	-72%
Coding specificity accuracy	78%	96%	+18 percentage points

Revenue Impact Calculation

Revenue Component	Annual Impact
RAF score improvement (0.105 x 15,000 x $11,600)	+$18.3M
Reduced retrospective review costs	+$240,000
Reduced external chart review vendor spend	+$180,000
RADV audit risk reduction (estimated avoided recovery)	+$400,000
Total annual financial impact	+$19.1M

AI Platform Investment

Cost Component	Annual Cost
AI coding platform (QuickCode)	~$1.2M
AI documentation platform (QuickScribe)	~$900,000
Implementation and integration (Year 1, amortized)	~$200,000
Internal program management	~$150,000
Total annual investment	~$2.45M

ROI Summary

Metric	Value
Annual revenue impact	$19.1M
Annual investment	$2.45M
Net annual return	$16.65M
ROI	680%
Payback period	47 days

Even at conservative assumptions — capturing only 60% of the identified RAF gap rather than the full modeled improvement — the ROI exceeds 400%, with a payback period under 90 days.

Sensitivity Analysis

Scenario	RAF Improvement	Annual Revenue Impact	ROI
Conservative (60% of gap closed)	+0.063	$11.0M	349%
Base case (80% of gap closed)	+0.105	$18.3M	680%
Aggressive (95% of gap closed)	+0.124	$21.6M	782%
Smaller panel (5,000 members, base case)	+0.105	$6.1M	508%
Larger panel (50,000 members, base case)	+0.105	$60.9M	1,240%

Getting Started: Risk Adjustment Accuracy Assessment

Organizations considering AI for risk adjustment should begin with a structured assessment of their current state:

Measure your current HCC capture rate. Compare coded HCCs against clinical documentation to establish a baseline. Industry benchmarks suggest 70-75% is average; 85%+ is high-performing.
Quantify your RAF gap. Compare current average RAF scores against expected RAF based on patient demographics and clinical profiles. CMS provides plan-level data that can serve as benchmarks.
Audit recapture rates. Track what percentage of prior-year HCCs are re-documented in the current year. Anything below 80% represents significant revenue exposure.
Identify your highest-value gaps. Not all HCCs are equal. Focus first on high-coefficient conditions (CHF, COPD, diabetes with complications, CKD, vascular disease) where documentation improvements yield the most revenue per condition.
Evaluate documentation quality. Assess whether gaps are primarily a coding problem (conditions are documented but not coded) or a documentation problem (conditions exist but are not documented). The solution differs significantly for each.

The organizations that achieve the highest risk adjustment accuracy are the ones that treat it not as a retrospective coding exercise, but as an integrated capability spanning documentation, coding, quality, and compliance — with AI as the connective tissue that makes it work at scale.

Internal Link References

Value-Based Care Revenue Cycle - Comprehensive guide to VBC financial management
Fee-for-Service to Value-Based Care Transition - What changes in your revenue cycle
AI Medical Coding: Accuracy, Compliance, and ROI - How AI coding technology works
AI vs. Human Coding Accuracy - Accuracy comparison and benchmarks
AI Scribe Accuracy and Safety - Documentation quality evaluation
How to Calculate ROI of AI in RCM - ROI framework and methodology
Complete Guide to Medical Coding - Medical coding fundamentals
ICD-10 and CPT Code Guide - Code set reference