The Machine That Saves Billions

The setup

$57.3B

premium dollars MA plans kept out of claims in 2023

85¢

minimum of every premium dollar the law makes them spend on care

$40B

of that kept by just the 65 largest contracts

The pitch says AI will make the payer stop saying no. The number a CFO actually watches says otherwise. Medical loss ratio (MLR) is the share of premium a plan pays out as care; 1 − MLR is what it keeps for admin and profit. Prior authorization is one of the biggest levers on the claims side of that ratio — every denied or deferred service is a claim not paid. Below, each dot is one MA contract: enrollment runs left→right (log scale), MLR runs bottom→top, dot size is enrollment. Watch where the giants land.

The explorer

571 contracts, one federal floor, and the plans that live right on it

Critical lens — hide contracts with fewer than 0 enrollees571 of 571 shown

↑ The wild dots — a plan “keeping 25¢ on the dollar,” or one paying out 160% — sit where enrollment is thinnest. Drag right and they dissolve: those were small samples, not strategy. The median barely moves, and what's left is a tight band hugging the 85% floor — the disciplined machine, tuned to keep exactly what the law allows.

MA contract kept ≥ $1B (billion-dollar valve)

571

Contracts shown

—

Median MLR

—

Total kept out of claims

The critical lens

Why a low MLR isn't a denial rate — and the giants tell the real story

Be careful what you read into a single dot. A low MLR is not a prior-authorization denial rate: it's also shaped by aggressive risk-adjustment coding (which inflates the revenue that sits in the denominator), the mix of supplemental benefits, reinsurance, and how a plan books reserves. And the alarming outliers — a plan “keeping a quarter of every dollar” — are almost all tiny contracts, where a few thousand members make the ratio swing wildly. Those are small-n artifacts, not villains. Drag the lens and they vanish.

The number that survives the lens. Keep only the large contracts and the story sharpens instead of disappearing: the biggest MA plans cluster in a narrow band just above the 85% floor — not at 70%, not at 95%, but tuned to the minimum the law allows. That's the tell. A denial machine calibrated this precisely, on this much money, is not a bug someone will let software switch off. It's the point of the product. MLR is the clearest public shadow of that machine; the actual prior-auth ledger — 50 million determinations a year — sits behind it.

Note on the data: MLR here is the CMS-reported adjusted_mlr for each MA (H-prefix) contract in the 2023 reporting year; “kept out of claims” is total reported revenue minus total incurred claims for that contract. A contract is not a company — large insurers run many. Values above 1.0 (paying out more than premium) are real and mostly small or new contracts. Illustrative of industry structure, not an audit of any one plan.

Why it matters for builders

The Big Thing's line — “the market for ‘we made the payer stop saying no’ is zero” — is this whole chart. You cannot sell a CFO a tool that shrinks the $57.3 billion machine their own board is measured on. What you can sell is the number on the other side: “we raise your first-pass approval rate by checking the payer's own published criteria before you ever submit.” Same data layer, opposite customer. Build the tool that wins the fight faster — not the one that pretends to end it.

Read the piece — “Prior Auth Is a Fight and AI Won't End It” → Or read today's newsletter → · Explore the raw CMS MLR data →