Share
CMS is moving Star Ratings money from how plans answer the phone to whether members stay, get care and stay out of the hospital. The same member conversations decide all three.
The service measures CMS is retiring are mostly ones Medicare Advantage plans already do well. Complaints about the plan, for example, averaged 4.0 stars in the new ratings. Many of the measures that remain are where the industry is weakest, such as diabetes medication adherence and follow-up after an emergency visit, which both average below 3 stars. And the member-reported measures of physical and mental health now count three times as much.
That shift lands in a hard year. Only about 37% of contracts with drug coverage reached four stars, down from 44%, and those ratings decide which plans earn quality bonuses in 2028.
It is tempting to read the removals, most of which take effect with the 2029 ratings based on 2027 performance, as permission to spend less on member experience. I read them differently. CMS hasn’t stopped paying for member experience. It has moved the payment to the point where experience turns into retention and health.
A member who can’t get a straight answer still calls 1-800-MEDICARE. A new enrollee who is disappointed in January can still switch plans through March 31. Grievance and appeal obligations are unchanged.
What changes is which scorecard records the failure. It stops showing up in the stars and starts showing up in the financials: in early-tenure disenrollment, in acquisition spend that never pays back, in avoidable admissions, and in the member-reported measures that remain, each carrying more weight in a smaller measure set.
The first 90 days
Retention is won or lost early. Researchers tracking members who left a newly chosen Medicare Advantage plan between January and March found that rate more than tripled from 2017 to 2022, to 12.2%, and reached 20% among dual-eligible members. They could not say why. But service teams hear the likely triggers every January: a doctor who turns out to be out of network, a drug that costs more than expected, an allowance card that never got activated, three calls to get one answer. Each is visible before the member leaves, and each is cheaper to fix than a member is to replace. The fix works best in the language the member actually prefers, captured at enrollment and carried into every call, text and letter.
The week after the emergency room
A member with heart failure and diabetes is treated in the ED and sent home. Whether the member returns depends partly on ordinary things: a follow-up visit booked, a ride arranged, a prescription filled. Clinical care belongs to the member’s doctors and the plan’s care managers. But when admission and discharge alerts flow to a worklist that service and care management share, trained advocates can handle the logistics and bring in a nurse when something sounds wrong. That supports a measure where the industry averages 2.5 stars, and it focuses effort on the members whose avoidable admissions cost the most.
The flex card declined at the register
Supplemental benefit cards are now part of how many members judge their plan, especially in special needs plans for dual-eligible members. For 2027, CMS requires these cards to verify eligible items in real time at the point of sale and to expire with the plan year. That protects the program, and it also means some members will learn at checkout that an item doesn’t qualify. CMS also dropped the requirement to send mid-year notices about unused supplemental benefits, so whether members understand their allowance is now up to the plan.
The opportunity is easy to overlook. A benefits call is one of the few conversations a hard-to-reach member starts on their own. Once the problem is solved, and only then, it can be the moment to book an overdue eye exam or medication review. A frustrated member gets their answer and nothing else.
No plan can afford a skilled advocate on every balance question and every discharge. Voice AI can answer routine allowance, card-status and eligible-item questions at any hour and in the member’s language. To earn trust, it needs controls a compliance team can audit. It tells members they are speaking with AI. It answers only from approved benefit content. It hands off to a person for anything clinical, uncertain or complaint-related. And every conversation is scored, with people reviewing the exceptions.
Language is where this pays off fastest. The interpreter and TTY measure leaves the Star Ratings, but the obligation to serve members in their own language does not, and the members who need it often include dual-eligible members, who leave new plans at the highest rates. A three-way interpreter call adds hold time and loses nuance. Real-time voice translation lets an advocate speak with a member directly in her preferred language, and lets voice AI answer routine questions the same way. For one Medicare and Medicaid plan, bilingual teams supported by real-time translation raised CSAT 35% and cut handle time 10%.
The economics should be explicit. Start with your own numbers: new members, times your early-disenrollment rate, times what each cost to acquire plus the margin you expected to earn. Add the cost of repeat contacts and of avoidable admissions among your highest-risk members. Then decide up front how much of the capacity AI frees becomes savings and how much funds outreach to new and complex members. Both are legitimate. Leaving the split undecided is how AI programs disappoint.
Two cautions keep this honest. First, not all retention is good retention. The goal is to keep the members a plan can serve sustainably and to lose fewer of them to preventable friction; a satisfied member is not automatically a profitable one. Second, a contact center cannot fix a thin network, a benefit design or a utilization management policy. A skilled advocate can resolve the call. Only the team that owns the cause can resolve the cause.
That is why we built RubrAIk, our healthcare payer AI framework, around three jobs. Augment gives advocates real-time guidance and next-best-action prompts, so a resolved benefits call can surface the open care gap. Automate handles routine questions and follow-through through voice and digital AI, with oversight. Anticipate analyzes every interaction, not a sample, and routes complaint drivers and emerging friction to the network, benefits or clinical owner while there is still time to fix them.
Anticipate is the part that answers the second caution. Most plans already hear their friction. It arrives as call reasons, complaint codes and survey comments, usually sampled, summarized weeks later and handed to a team that cannot change the cause. Anticipate works the other way. It reviews every interaction, groups what members are saying by cause, and sends each cause to the team that owns it. A rise in out-of-network surprises goes to network management. A cluster of formulary questions goes to pharmacy. A pattern of declined flex card purchases goes to the benefits team before it turns into complaints to Medicare. The advocate still resolves the call. Anticipate makes sure the cause reaches someone who can remove it.
The evidence so far is operational. Rebuilding training around member personas and top call reasons cut escalations 54% for a large managed Medicare plan. Omnichannel outreach for a Fortune 50 Medicare Advantage plan increased scheduled screening appointments 2.8 times over two years while costs fell 70%. Neither result proves a Star Ratings outcome on its own. Voice AI is newer and should meet the same standard of measured results before it scales.
Defend first, then rebuild. Four moves I would make now:
- Name the five biggest drivers of repeat contact from this Enrollment Season, including flex card questions, and give each a cause owner outside the contact center, with a date.
- Replace the scripted welcome call with a personalized outreach triggered by risk: a doctor out of network, a drug off formulary, a card not activated, a second call in two weeks. Deliver it in the member’s preferred language.
- In one market, connect admission and discharge alerts to a shared service and care management worklist for members with multiple chronic conditions, and measure follow-up rates.
- Move routine benefit workflows to voice AI with the controls above, measured on accuracy, resolution and handoff.
For 2028, the goal is an operating model in which what members tell the plan informs benefit design, network and quality decisions before the next bid is filed.
My working thesis is this: the plans that perform best in the next cycle won’t be the ones that spend the most on service. They will be the ones that keep the members they can serve well, reach complex members at the moments that matter, and do it at a cost the business can sustain. Every plan should test that against its own data.
If you want to test it with us, pick one regulated workflow. Give us 90 days. We will measure the result independently and tie our fee to it.
See how ResultsCX helps Medicare Advantage plans find member friction early and remove it.