Why risk adjustment software is a 2026 problem, not a 2024 one
The CMS-HCC V28 model is fully effective for payment year 2026, completing a three-year phase-in. V28 cut the ICD-10-to-HCC crosswalk from roughly 9,797 valid codes under V24 to about 7,770, and CMS's actuaries projected an average MA risk score reduction of about 3.12%. Diagnoses that triggered an HCC under V24 may no longer do so. Read our V28 readiness playbook for the full picture.
The operational consequence is simple. If your software only works after claims settle, it will tell you what you lost after the window to fix it has closed. The capture decision now has to happen at or before the encounter. That single shift reorders what "best" means in this category.
What to look for in risk adjustment software
Prospective and retrospective in one place
Prospective capture happens at or before the visit, so the diagnosis is documented within the year it is clinically valid. Retrospective review catches what was missed and supports recovery and audit. Running them as two disconnected tools means the two views never reconcile against the same member record. The best software does both and reconciles them.
V28-aware trumping logic
V28 expanded to 115 condition categories and tightened constrained groups, where related conditions share a single coefficient. Documenting more conditions in a constrained group does not raise RAF: only the highest-weighted member counts. Software should apply this trumping logic live, so coders are not chasing diagnoses that will not pay, and so redundant capture does not create audit exposure.
RADV-defensible chain-of-custody
Every captured HCC should carry a retrievable evidence packet: the source clinical note with MEAT support, supporting labs or imaging, any provider attestation, and the coder review record. This should be a property of the system, not a packet assembled by hand the week an auditor calls. Our RADV-defensible HCC guide details the design choices that hold up under audit.
Full submission-lifecycle visibility
A diagnosis is not captured until CMS recognizes it. Good software tracks each one through the 277CA acknowledgement, the MAO-004 (which reports diagnosis-level risk eligibility, additions, and deletions), the MMR membership and payment report, and the MOR model output report. MAO-004 rejections are an early signal that documentation will not survive RADV. Seeing the whole chain in one place is the difference between catching a problem and discovering it in a payment shortfall.
Point-of-care, pre-claim flagging
The highest-leverage feature is surfacing the specific at-risk condition during the encounter, with the clinical criteria required. Generic prompts ("review chronic conditions") do not move capture. Specific prompts ("documented CKD Stage 3a in 2024; confirm staging today") do.
Vendor comparison
The vendors below are all credible in risk adjustment. They differ in where they focus: some at the point of care, some in retrospective chart review, some as broad data platforms. Categories are descriptive, not pejorative.
| Vendor | Focus (prospective / retrospective / both) | AI approach | RADV / audit support | Submission-lifecycle visibility | Best-fit org |
|---|---|---|---|---|---|
| Navina | Prospective, point-of-care | Clinician copilot; EHR-native condition detection | Evidence linked to the chart | Not its primary focus | Physician groups and ACOs working in the exam room |
| Reveleer | Retrospective-led; quality too | Evidence Validation Engine (EVE) for chart review | Manages CMS and RADV-IVA submissions | Submission and audit-focused | Plans running scaled chart review and audit |
| Apixio | Both; retrospective, prospective, concurrent | NLP/ML coding predictions plus coder review | HCC Auditor; compliance solutions | Coding and compliance-focused | MA and ACA organizations scaling coding accuracy |
| RAAPID | Both; retrospective and concurrent | Neuro-symbolic AI plus knowledge-graph NLP | MEAT-based audit trail; RADV solution | Coding and audit-focused | Plans and groups wanting defensible coding at scale |
| ForeSee Medical | Prospective, point-of-care | NLP disease-detection engine in the EHR | Source-document hyperlinking (InstaVu) | Not its primary focus | Practices and plans wanting EHR-embedded suspecting |
| Innovaccer | Both, via platform and Galaxy agents | Data platform plus specialized agents | Part of broader risk/quality workflows | Within the platform | Enterprise plans standardizing on one data platform |
| Arcadia | Retrospective-led, analytics | Data lakehouse and analytics | Risk analytics, not a coding engine | Analytics view | Systems and ACOs wanting a data foundation |
| Pelica Risk Adjustment Copilot | Both, real-time, on one shared record | Copilot plus action layer that executes the follow-through | Chain-of-custody per HCC; trumping applied live | 277CA through MOR, per diagnosis, in one view | Risk-bearing IPAs, ACOs, and plans replacing vendor sprawl |
Navina
Navina is a clinician-first copilot for prospective capture. It summarizes patient data from the EHR, HIE, and claims, surfaces suspected conditions at the point of care, and supports one-click documentation inside the chart. For a physician group whose main lever is the visit, Navina is well designed. (Source: Navina risk adjustment.)
Reveleer
Reveleer is built for high-volume retrospective chart review. Its Evidence Validation Engine retrieves and parses records and populates abstraction fields for review, and the company manages CMS and RADV-IVA submissions. For plans that need scaled, on-time chart review and audit submission, Reveleer is a strong fit. (Source: Reveleer retrospective risk adjustment.)
Apixio
Apixio offers retrospective, prospective, and concurrent coding, combining NLP and machine-learning predictions with certified coder review. It also launched an AI-powered HCC auditing solution. For MA and ACA organizations focused on coding completeness and compliance across the full review cycle, Apixio is a credible engine. (Source: Apixio risk adjustment.)
RAAPID
RAAPID uses neuro-symbolic AI and knowledge-graph-infused clinical NLP to produce an evidence-backed, MEAT-based audit trail for every HCC. It supports both autonomous retrospective review and a RADV audit solution. For teams that prioritize defensible, traceable coding, RAAPID's design leans into audit transparency. (Source: RAAPID retrospective risk adjustment.)
ForeSee Medical
ForeSee Medical embeds NLP-driven disease detection directly in the EHR, presenting risk-adjustment suspects and HCC recommendations at the point of care, with InstaVu hyperlinking suspects back to the exact page of source documentation. For practices that want prospective suspecting inside their existing workflow, it fits. (Source: ForeSee Medical HCC coding.)
Innovaccer and Arcadia
Both are enterprise data platforms rather than dedicated coding engines. Innovaccer's Galaxy platform adds specialized agents for payer risk and quality workflows on top of its unified data model. Arcadia provides analytics and a longitudinal record across a large population. For organizations whose first need is a clean data foundation, either is reasonable; risk adjustment is one workflow within a larger platform. (Sources: Innovaccer DAP; Arcadia platform.)
How Pelica's Risk Adjustment Copilot differs
Most of the vendors above do one part of the job well: prospective suspecting, or retrospective review, or audit trails, or analytics. Pelica's Risk Adjustment Copilot is built to do them together, in real time, on a record that the rest of the organization shares.
- Real-time, both directions. At-risk HCCs surface before the encounter, and retrospective review reconciles against the same record, so prospective and retrospective never drift apart.
- V28 trumping applied live. Constrained-group trumping is computed as conditions are captured, so coders work the diagnoses that actually pay.
- Chain-of-custody by default. Every captured HCC carries its source note, supporting evidence, attestation, and coder review record, ready for RADV.
- Full lifecycle in one view. Each diagnosis is tracked from 277CA through MAO-004, MMR, and MOR, so rejections are caught early.
- One record across all teams. The same canonical record feeds quality, pharmacy, network, and care management, so a member with three at-risk HCCs gets one well-prepared encounter, not three separate calls.
At our flagship customer, a physician-led IPA in New York running risk on roughly 175,000 patients, the Risk Adjustment Copilot reached 100% team adoption across the platform. Across Pelica deployments, customers have lifted RAF by roughly +0.4 in two quarters with no new headcount.
The point is not a better coding engine. It is the same record under risk, quality, pharmacy, network, and care management, so capture is a coordinated operating cadence, not a siloed coding project.