The medical loss ratio has a reputation as the rule that stops Medicare Advantage plans from keeping too much money. Here is the number that complicates the story: the largest MLR remittance year on record is CY2020, when 118 MA contracts fell under the 85% floor and sent $1.42 billion back to CMS, and it happened because pandemic care deferral meant plans spent too little. In CY2023, the latest reported year, only 3.7% of MA contracts failed the floor, and outside the pandemic year the annual failure rate has run between roughly 2% and 9%. The aggregate regulatory MA MLR has sat near 90% for years, five points above the line.
Meanwhile, between 2022 and 2025, UnitedHealth's reported medical care ratio went from 82.0% to 89.1%, and CVS's Health Care Benefits ratio ran from 83.8% in 2022 to a 92.5% peak in 2024, an 870-basis-point swing in two years, before easing to 91.2%. Humana is guiding its insurance-segment FY2026 benefit ratio to roughly 92.75%. (Those are GAAP figures under each company's own definition, some segment-level, not the regulatory MLR, which adds quality spend to the numerator and nets taxes out of the denominator; never chart the two together.) The direction is what matters: for most of the last fifteen years, everything written about MLR assumed the problem was plans spending too little on care. Since 2024 the live problem is the opposite. Both directions have a playbook, and they share one lever almost nobody pulls hard enough.
What the medical loss ratio is
The MLR is a ratio of what a plan spends on care to what it collects. The Medicare Advantage version, at 42 CFR Part 422, Subpart X, is:
(incurred claims + quality improvement activity spending) ÷ (premium revenue − taxes and licensing fees), with a credibility adjustment for small contracts, calculated per contract, per year.
Three structural points operators regularly get wrong:
- QIA lives in the numerator. Quality improvement spending counts as care, not administration. That single design choice is most of the strategy section below.
- MA has no multi-year averaging. The commercial ACA markets calculate MLR on a three-year average; MA does not. Each contract year stands alone.
- The floors differ by market. MA: 85%. Commercial: 80% individual and small group, 85% large group, under 45 CFR Part 158, with rebates going to policyholders. Medicaid managed care: an 85% target under 42 CFR 438.8, but remittance is a state option, and roughly a fifth of MCO states do not always require it.
What happens below 85%
The MA sanction ladder at 42 CFR 422.2410 is cumulative and per contract. Miss the floor and you remit the difference to CMS. Miss it three consecutive years and CMS freezes new enrollment. Five consecutive years and the contract is terminated. The remittance itself is excluded from every year's MLR math, so it is the one option on the menu that buys nothing.
The commercial market shows the same physics with rebates instead of remittances: rebate totals peaked in the 2019-2020 reporting years, again because the pandemic left insurers under-spent, and CMS's figures for reporting year 2024 show $1.64 billion paid out. Floors bite when utilization collapses. They do not brake anything when utilization spikes; when claims surge, the roughly 13 points of the bid a plan keeps for administration and margin are the entire shock absorber.
The four levers
MLR is a ratio, so there are exactly four moves:
The fourth, shrinking revenue, nobody chooses on purpose. Which direction you want depends entirely on which side of 85 you sit, and for many MA operators the answer flipped between 2022 and 2024.
If your MLR is running too high
Reduce avoidable utilization, and book the work as QIA while it pays for itself. The regulation is unusually well aligned with the clinical evidence here. The canonical transitional-care trials, Coleman's Care Transitions Intervention (30-day rehospitalization 8.3% vs 11.9%), Naylor's APN discharge planning (24-week readmission 20.3% vs 37.1%), and Project RED (utilization IRR 0.695), are exactly the "grounded in evidence-based medicine" footing the QIA tests demand, and post-discharge outreach is named at 42 CFR 422.2430(a)(2)(ii). The spend shrinks future claims and counts as care today. No other lever has that dual character.
Risk adjustment accuracy, with the compliance asterisk stated plainly. Complete, documented risk capture raises payment, which enlarges the denominator. But this is the most policed lever in the program: OIG's October 2024 report found diagnoses appearing only on health risk assessments and HRA-linked chart reviews drove an estimated $7.5 billion in 2023 payments, with 20 companies generating 80% of it. The durable version of this lever is documentation at real visits, MEAT-complete and RADV-defensible, not code mining.
Stars revenue, with the nuance most content misses. A 4-star plan gets a 5-percentage-point benchmark increase under 422.258(d)(7), doubled in qualifying counties. That is clean revenue and the better MLR lever. The rebate ladder (50% below 3.5 stars, up to 70% at 4.5+) is different: rebate dollars must fund supplemental benefits or premium buy-downs, so they enter the denominator and then return to the numerator when members use them. Closer to MLR-neutral by design.
Right-size supplemental benefits. Since contract year 2023, incurred claims must be reported with supplemental benefits broken out, so the line is visible to CMS. An under-used supplemental benefit is the worst of both worlds: it consumes rebate dollars, hits the numerator only when used, and turns into a CAHPS liability when members cannot access it.
Payment integrity, which MA treats better than any other market. In MA, fraud prevention, detection, and recovery activities all count as QIA under 422.2430(a)(4)(ii); commercial issuers get no such treatment. Recovered overpayments reduce incurred claims, so the net MLR effect depends on the mix, but the activity itself is not dead administrative weight.
Pharmacy: MTM is the frictionless dollar. A medication therapy management program meeting the Part D requirements is QIA by name under 422.2430(a)(4)(i), no four-part test to argue, and the program is mandatory anyway. Adherence outreach that moves PDC is the single most regulatorily comfortable quality spend in the program.
If your MLR is running too low
Four options: enrich benefits, invest in QIA, raise provider quality incentives, or remit. The first three convert a would-be penalty into member value, quality performance, or network goodwill; quality bonuses to providers count as incurred claims and, under 422.208, quality-based payments are excluded from the substantial-financial-risk calculation, so they do not drag stop-loss obligations with them. The fourth converts the shortfall into nothing. Given the three-year enrollment freeze on the ladder, a contract sitting at 84 has every reason to spend rather than remit.
The cheapest lever: book quality spend as what it is
The most consistent finding in fifteen years of MLR data is that plans under-claim quality. GAO measured commercial QIA spending at 0.7% to 0.9% of premiums in 2011-2012, and all eight insurers it interviewed said the MLR rules had little or no effect on their quality spending. A decade later the MA picture is barely different: in CY2023, QIA was 1.54% of the MLR denominator while non-claims costs ran 8.03% of revenue. Five dollars of administration for every dollar of counted quality spend.
Spend that routinely sits in the admin bucket but qualifies as QIA: HEDIS chart review and CAHPS survey costs, NCQA accreditation fees, care coordination and chronic disease management, post-discharge outreach, medication adherence programs, patient self-management tools, and the reward dollars in member incentive programs. The required MTM program, the required chronic care improvement program, and the required CAHPS survey are all spend the regulation obliges and the MLR rewards; booking them as overhead is an unforced error.
The other direction is real too. CMS's commercial MLR audits found issuers stuffing QIA with overhead, marketing, vendor profits, even office wall art, and tightened 45 CFR 158.150 in 2022 to require expenses be "directly related" to quality. CMS proposed extending that language to MA and has not finalized it; assume it intends to. The discipline that survives audit in either direction is the same: every quality dollar instrumented to a named measure with verifiable results at the time it is spent, records kept ten years.
What not to do
Denial pressure. The Senate Permanent Subcommittee on Investigations' October 2024 report documented post-acute prior-authorization denial rates at the largest MA insurers rising sharply from 2020 to 2022, and June 2026 OIG reports found the three largest MA insurers denying roughly 65% of long-term acute care and 54% of inpatient-rehab prior-authorization requests, with nearly all appealed skilled-nursing denials overturned. Beyond the regulatory and reputational exposure, there is a clean irony in the MLR math itself: retrospective and concurrent utilization review is categorically excluded from QIA under 422.2430(b)(7). The apparatus that suppresses the numerator is itself administrative cost. Denial-led MLR management loses on the regulation's own terms.
QIA stuffing. The exclusion list at 422.2430(b) is broad, the catch-all puts the burden on the plan, and CMS can audit, recoup, and sanction. The test that separates legitimate from aggressive: can you name the measure and show the verifiable result?
The vertical-integration question, stated honestly. When an insurer owns the providers and pharmacies it pays, MLR "claims" are partly intercompany transfers, and neither Part 158 nor Subpart X caps what a plan may pay an affiliate in the numerator. CMS itself raised this in a 2024 request for information, writing that related-party payments "may, in some cases, be inflated" to meet MLR requirements. None of the proposed fixes were finalized, and MLR reporting has since appeared on a deregulatory candidate list. The peer-reviewed evidence is genuinely mixed: hospital-owned MA parents report higher MLRs, but with lower premiums and claims, and OIG's May 2026 look at Part D found vertically integrated sponsors paid their own pharmacies slightly less, not more. An open policy question, not established wrongdoing, and worth watching because the strongest reform proposals on the table are CMS's own, currently stalled.
If you are a delegated IPA or MSO
You do not have a federal MLR. Subpart X applies to MA organizations; a delegated IPA or MSO is a first-tier entity with no 85% floor, no remittance, no sanction ladder. What you manage instead is a medical expense ratio against your capitation, and its boundaries are set by the DOFR, which is why two groups with identical clinical performance can report wildly different ratios.
Two facts worth carrying into your next payer negotiation. First, every capitation dollar the plan pays you counts fully as incurred claims in the plan's MLR, so heavy delegation is mechanically an MLR-raising move for them. Second, quality-based incentive payments are excluded from the substantial-financial-risk calculation under 422.208 and count as incurred claims for the plan. Quality-linked dollars are the least regulatorily encumbered way for a plan to move money to your group. Ask for them.
Where AI care management spend fits
The QIA design tests at 422.2430(a)(3) ask four things: improves quality, objectively measurable with verifiable results, directed at enrollees or segments, grounded in evidence. The parallel commercial text explicitly contemplates telephonic and web-based interaction and vendor-delivered services the plan remains responsible for. AI-run outreach, adherence work, and post-discharge follow-up fit the categories as written, with one clean boundary: a service billed by a provider as a clinical claim cannot also be QIA. Pay a practice for TCM and it is a claim; run plan-side or group-side post-discharge outreach and it is quality spend.
This is the shape of what Pelica builds. The Copilots run care-gap, adherence, and post-discharge outreach member by member, and every touch is logged against the measure it serves, TRC, PDC, the HEDIS clinical set, which is precisely the "verifiable results" instrumentation the QIA tests and a ten-year record requirement ask for. Customers pass 70% on Transitions of Care within 30 days of going live and hold 96%+ adherence on the triple-weighted Part D measures. The work improves the measures; the documentation makes the spend count for what it is.
Sources and methodology
Regulatory text was read from the current eCFR (42 CFR Part 422 Subpart X, 45 CFR Part 158, 42 CFR 438.8), rule history from the Federal Register, MLR distributions from the CMS Part C MLR public use files (CY2023, the latest posted), rebate totals from CMS's MLR refund files, market trend from KFF's NAIC-based series and each carrier's 10-K, and the oversight record from GAO, HHS-OIG, MedPAC, and the Senate PSI report, all linked in context. GAAP medical-cost ratios and the regulatory MLR are different constructs and are never compared directly above.