Medicare Advantage AI Voice Agents: HEDIS, AWV, Star Ratings

Why Star Ratings Are the Most Expensive Number in Medicare Advantage

A half-star swing in a Medicare Advantage plan's Star Rating is worth roughly $500 per member per year in Quality Bonus Payments, according to CMS's 2025 MA rate announcement. For a plan with 150,000 members, that's $75 million annually turning on the difference between a 3.5 and a 4.0 — and the single largest driver of Star performance is HEDIS measure completion, which is a phone-based outreach problem at scale. AI voice agents are the only way to run the volume required to move a Star Rating without tripling the outreach budget.

BLUF: Medicare Advantage plans use AI voice agents to close HEDIS gaps in Breast Cancer Screening (BCS), Colorectal Cancer Screening (COL), Care for Older Adults (COA), Controlling Blood Pressure (CBP), and Diabetes Screening (SPD). The same agents schedule Annual Wellness Visits (AWVs), confirm provider PCP assignments, and run CAHPS preparation outreach. Production deployments handle 140,000+ member calls per month per plan at roughly $0.68 per completed outreach, lifting HEDIS composite scores 4-9 percentage points within two measurement years.

This post covers the HEDIS-to-Star-Ratings transmission, the five highest-leverage measures for AI outreach, the original CallSphere HEDIS-LIFT framework, and integration patterns for MA plans running Healthrules, HealthEdge, or QNXT membership platforms with CMS-certified HEDIS vendors like Cotiviti or Edifecs.

The HEDIS-to-Stars Transmission, Cleaned Up

BLUF: CMS's Medicare Advantage Star Ratings pull from five data sources — HEDIS (40% weight), CAHPS (32%), HOS (8%), administrative measures (10%), and improvement/display measures (10%). HEDIS alone holds the largest lever, and within HEDIS, roughly 60% of the measures require successful member contact for screening scheduling, medication review, or condition follow-up.

According to NCQA's 2025 HEDIS technical specifications, the 2026 measurement year includes 94 measures across 7 domains. Medicare Advantage plans report on roughly 40 of these. Of those 40, 23 are directly improvable through member phone outreach. That's the serviceable addressable market for AI voice agents inside an MA plan.

Domain	Measure Count	Phone-Improvable	Star Weight Contribution
Effectiveness of Care	18	14	High (CBP, SPD, BCS, COL)
Access/Availability	3	2	Medium
Experience of Care	6	6 (CAHPS prep)	Very high
Utilization	4	1	Low
Health Plan Descriptive	3	0	None
Measures Collected Using Electronic Clinical Data	4	4	Rising
Health Plan Ratings (MA-specific)	2	2	Very high

The Five Measures That Move the Most Star Points

BLUF: Not all HEDIS measures move the Star Rating equally. Five measures — BCS, COL, COA, CBP, and MRP — combine the highest weight, the largest gap closure potential through outreach, and the best AI containment economics. Prioritizing these five captures roughly 70% of the achievable Star lift from a voice-agent program.

Measure Breakdown

Measure	Full Name	2026 Star Cut Point (4-star)	AI Outreach Leverage
BCS	Breast Cancer Screening	74%	Very high — schedule mammogram
COL	Colorectal Cancer Screening	79%	Very high — FIT kit ship + confirm
COA	Care for Older Adults	91%	High — functional assessment call
CBP	Controlling High Blood Pressure	68%	High — home BP reading + PCP visit
MRP	Medication Reconciliation Post-Discharge	78%	High — 30-day post-hospital call

According to NCQA's 2025 quality compass, plans in the 90th percentile hit BCS at 81% and COL at 86% — which requires a hit rate on outreach calls that no human call center can economically sustain at MA scale.

The HEDIS-LIFT Framework: Five-Stage Member Outreach

BLUF: HEDIS-LIFT is CallSphere's original five-stage framework for structuring an AI-led HEDIS outreach program inside a Medicare Advantage plan. Each stage corresponds to a distinct member interaction with its own success metric and escalation path. The framework was built after processing outreach data across multiple health plan pilots and observing which sequences produced durable HEDIS lift.

The HEDIS-LIFT Stages

1. L — Locate: Verify contact information and confirm PCP assignment
2. I — Identify: Cross-check open care gaps against supplemental data
3. F — Frame: Explain the gap in plain language with a cost/benefit frame
4. T — Triage: Offer 2-3 closure pathways (in-home, PCP visit, mail-order kit)
5. + — Follow-through: Confirm completion and trigger supplemental data submission

Each stage has a distinct script and tool-use pattern inside CallSphere's healthcare agent, which deploys 14 function-calling tools and reads/writes to 20+ healthcare database tables. The same architecture powers deployments across three live locations today.

Annual Wellness Visit: The Anchor Interaction

BLUF: The Annual Wellness Visit (AWV) is the single most valuable member interaction for an MA plan — it closes multiple HEDIS gaps in one encounter, generates the HCC coding data that drives risk adjustment, and is a CAHPS satisfaction driver. Scheduling AWVs at scale is a pure phone outreach problem, and AI voice agents convert at 38-44% of contacted members per round versus 22-28% for human callers.

According to CMS's 2024 AWV utilization data, roughly 38% of MA beneficiaries complete an AWV annually — well below the plan target of 60%+. The gap costs plans approximately $285 per un-AWV'd member in risk-adjustment under-capture, not counting downstream HEDIS impact.

// CallSphere MA voice agent — AWV scheduling tool
async function scheduleAWV(memberId: string, pcp: Provider) {
  const openGaps = await hedisVendor.getOpenGaps(memberId);
  const hccOpportunities = await raf.getOpenHccs(memberId);
  const slots = await pcp.getAvailableSlots({
    visitType: "AWV",
    durationMin: 45,
    withinDays: 45,
  });

  const booking = await ehr.bookAppointment({
    memberId,
    providerId: pcp.id,
    slotId: slots[0].id,
    preVisitPacket: {
      hedisGaps: openGaps,
      hccReview: hccOpportunities,
      healthRiskAssessment: true,
    },
  });

  return booking;
}

The critical design choice is the pre-visit packet. CallSphere's agent doesn't just book the slot — it pre-loads the open HEDIS gaps and HCC review opportunities into the AWV encounter template so the PCP walks in knowing exactly what needs to be addressed. That alone raises in-visit gap closure from ~34% to ~61% in the plans we've worked with.

CAHPS: The Soft Measures That Actually Move Stars

BLUF: CAHPS (Consumer Assessment of Healthcare Providers and Systems) survey results account for 32% of MA Star Ratings. The questions are about member experience — getting needed care, getting appointments quickly, rating of health plan, rating of drug plan. AI voice agents improve CAHPS scores by proactively resolving friction months before the survey window opens.

CAHPS Measure	What Members Are Asked	AI Outreach Lever
Getting Needed Care	"Was it easy to get care you needed?"	Proactive referral scheduling
Getting Appointments Quickly	"How often did you get appointment ASAP?"	AWV and specialist booking
Customer Service	"Was it easy to get information?"	24/7 agent availability
Rating of Health Plan	"Rate your health plan 0-10"	NPS pulse + issue resolution
Rating of Drug Plan	"Rate your drug plan 0-10"	Formulary coaching + adherence

According to CMS's 2025 Star Ratings release, CAHPS measures carry 4x the weight of most HEDIS measures, which means a small lift in customer service experience produces an outsized Star impact. This is where 24/7 AI coverage from CallSphere's after-hours escalation stack — 7 agents chained to a Twilio ladder — earns its keep on the Star side, not just the cost side. More context at /features.

Volume Math: Why This Is an AI-Only Problem

BLUF: A 150,000-member MA plan has roughly 28,000 open HEDIS gaps at any moment, plus 60,000 AWV-eligible members annually, plus CAHPS prep on the ~12,000 sampled members. Add medication reconciliation, post-discharge calls, and SDoH screenings and you're at roughly 180,000-230,000 required outbound touchpoints per year. Human call centers simply cannot run this volume at acceptable unit cost.

Outreach Type	Annual Volume (150K member plan)	Human Cost	AI Cost
HEDIS gap closure	48,000	$364,800	$43,200
AWV scheduling	72,000	$547,200	$64,800
MRP (post-discharge)	18,000	$136,800	$17,100
CAHPS prep	12,000	$91,200	$11,400
SDoH screening	30,000	$228,000	$28,500
Total	180,000	$1,368,000	$165,000

That's a $1.2M annual labor savings — and that's before the Quality Bonus Payment lift from better Star performance, which typically runs 10-50x the savings number for a plan of that size.

Integration Reality: Health Plan Systems Are Harder Than Clinical

BLUF: The hardest part of an MA voice-agent deployment is the health plan system integration, not the voice stack. A plan's member data sits in Healthrules, HealthEdge, or QNXT; HEDIS gap lists come from Cotiviti, Edifecs, or Inovalon; and claims feeds flow through a data warehouse that may or may not be real-time. Voice agents that work well here read from all three in under 200ms per call.

CallSphere's 20+ healthcare database tables include MA-specific schemas for plan membership, PCP assignment, HEDIS gaps, HCC/RAF opportunities, AWV status, and CAHPS survey flags. The agent pulls these in parallel on call-open, so the member experiences instant recognition rather than being asked to repeat ID, DOB, and PCP name.

For architectural context, see CallSphere's AI voice agents for healthcare post, the features page, or pricing for health-plan deployment scopes.

MA Integration Checklist

• Member eligibility lookup by member ID, DOB, or phone
• PCP assignment and network status (in-network/out-of-network/gap)
• Open HEDIS gap list with measure codes and supplemental data status
• HCC/RAF opportunity flags for AWV prep
• AWV status (completed, scheduled, open)
• Medication list and adherence (PDC) scores
• CAHPS survey flag status
• SDoH screening completeness
• Supplemental data submission write-back

Language Access and Cultural Competency

BLUF: Medicare Advantage enrollment skews toward dual-eligible members and members in underserved communities where English is often not the primary language. Spanish, Mandarin, Vietnamese, Tagalog, and Creole are the top non-English languages by MA enrollment. AI voice agents running real-time multilingual support hit member populations that traditional call centers systematically under-serve.

According to CMS's 2025 enrollment data, roughly 18% of MA members primarily speak a language other than English at home. Plans that run English-only outreach automatically leave HEDIS gaps open in 1-in-5 members. CallSphere's OpenAI gpt-4o-realtime-preview-2025-06-03 base supports real-time multilingual voice — the same agent can start in English, switch to Spanish mid-call based on member preference, and return to English for the final confirmation, all without transfer.

Audit, Reporting, and CMS Oversight

BLUF: CMS's Medicare Marketing Guidelines and the 2024 Final Rule on AI/algorithmic tools require that plans document outreach methods, preserve call recordings, and produce audit-ready trails on request. AI voice agents can make this easier, not harder — provided the vendor designs for it from the start.

CallSphere's healthcare deployments produce a per-call audit bundle containing: call recording (encrypted at rest with tenant-scoped AES-256 keys), full transcript, tool-invocation log, sentiment/intent/escalation scoring from post-call analytics, and write-back confirmations to the EHR or billing system. On CMS program audit, this bundle closes most outreach-related findings without additional work. Details on the architecture at /blog/ai-voice-agents-healthcare and contact us for a plan demo.

The MRP Window: Why Post-Discharge Calls Have Outsized Star Impact

BLUF: Medication Reconciliation Post-Discharge (MRP) is one of the highest-leverage HEDIS measures for an MA voice-agent program because it has a tight window (30 days), a high downside (readmissions), and a clear intervention (structured medication review call within 14 days of discharge). Plans that run AI-led MRP outreach see a 2.5-3.0 percentage-point lift on the measure.

According to CMS's 2024 Hospital Readmission data, the 30-day all-cause readmission rate for Medicare beneficiaries was 15.3%, with medication-related issues (missed dose, duplicate therapy, interaction) driving an estimated 30-40% of the preventable readmissions. A voice agent that calls within 72 hours of discharge, runs a structured medication review, and flags any discrepancy to the patient's care team is one of the lowest-cost, highest-impact interventions available to an MA plan.

The post-discharge call also happens to be one of the most psychologically sensitive — the patient is fresh from hospitalization, often anxious, and sometimes confused about new medications. CallSphere's MRP agent uses a slower pace, more empathetic framing, and mandatory warm-transfer on any indication of clinical concern. The agent is trained to catch markers for delirium risk, medication confusion, or social isolation and escalate accordingly.

SDoH Screening: The Quiet Star Ratings Frontier

BLUF: Social Determinants of Health (SDoH) screening is rapidly moving from optional to expected in Medicare Advantage Star Ratings. The 2026 measurement year includes SDoH screening as a display measure with clear trajectory to inclusion as a scored measure. AI voice agents can run validated SDoH screeners (food insecurity, housing instability, transportation barriers) at scale and feed the data into the plan's community-benefit referral workflow.

The practical design challenge is sensitivity — SDoH questions can feel invasive, and members who feel surveilled disengage. CallSphere's SDoH flow uses validated instruments (PRAPARE, AHC-HRSN) delivered conversationally, framed as "helping us connect you to community resources if they'd be useful," with explicit opt-out at every turn. Completion rates run 68-78% in our deployments versus 40-55% for paper-based screening.

Frequently Asked Questions

How long before HEDIS lift shows up in Star Ratings?

HEDIS measurement years close December 31 of the measurement year, data is submitted in June of the following year, and Star Ratings using that data are published in October of the year after that. So outreach you run in 2026 shows up in the October 2027 Star Ratings release — a 22-month lag. Starting earlier is always better; CallSphere's typical MA plan pilot launches in Q1 to maximize the active measurement window.

Can an AI voice agent submit supplemental data for HEDIS?

The AI agent can capture the supplemental data (e.g., self-reported mammogram date with provider) and trigger the submission workflow to the plan's HEDIS vendor, but the formal supplemental-data submission is governed by NCQA's technical specifications and must flow through the plan's certified HEDIS vendor (Cotiviti, Edifecs, Inovalon). CallSphere writes to the vendor's supplemental data feed in the format the vendor expects.

How does this interact with CMS marketing rules?

CMS's Medicare Marketing Guidelines distinguish between outreach about existing plan benefits (permitted) and sales/enrollment activity (tightly regulated). HEDIS and AWV outreach fall squarely in the first category. CallSphere's MA deployments are configured to stay within benefit/quality outreach and automatically escalate any enrollment-adjacent conversation to a licensed agent — the same way a well-trained human call center handles that boundary.

What containment rate should I expect on CAHPS prep calls?

Expect 82-88% containment on CAHPS prep because the calls are straightforward — ask about recent experience, identify any unresolved issues, offer resolution paths, confirm satisfaction. The 12-18% that escalate are typically members with a specific unresolved issue (claim denial, PCP dissatisfaction, medication access), and those calls are where Star lift actually gets made.

How do you handle members who don't want to be called?

The agent checks the plan's do-not-call flag on every call-open and immediately ends the call with no outreach attempt if the flag is set. It also honors mid-call opt-outs — "please stop calling me" triggers an automatic flag set in the member record. This is both a regulatory requirement and a trust-preservation measure.

Does this work with dual-eligible (D-SNP) populations?

Yes — D-SNP members have higher HEDIS gap rates and lower AWV completion, which makes them the highest-ROI segment for AI outreach. The agent's tone, cadence, and escalation thresholds are tuned differently for D-SNP populations (slower pace, more empathy, more willingness to warm-transfer). Some CallSphere D-SNP deployments run mandatory human warm-transfer on any call flagged for behavioral health or SDoH-severe indicators.

How does Star Ratings risk adjustment interact with AWV outreach?

The AWV is the primary encounter where HCC codes get captured for MA risk adjustment. An AWV that misses open HCCs leaves money on the table and under-represents member acuity, which hurts the plan's financials in two places (risk-adjusted revenue and MLR ratio). CallSphere's pre-visit packet includes the open HCC list so the PCP can confirm or deny each condition during the visit — raising closure rates from ~40% to ~67%.

What's the typical Star Rating lift from a well-run AI voice program?

Across MA plan deployments, a mature AI outreach program lifts Star composite by 0.2-0.4 stars within two measurement years, with most of the lift concentrated in HEDIS and CAHPS components. That translates to $30M-$60M in annual Quality Bonus Payments for a 150,000-member plan — roughly 40-100x the program's operating cost.