Nobody Has Measured What a Benefit Buys. The Bids Assume It Anyway.
Benefit design is where a health plan decides membership, revenue and cost of care in one meeting, seven months before the plan year, on a benchmark eight weeks old. It is also the least-evidenced room in the building. Six things from building an instrument for it.
1. The rebate is the product budget — and stars fund it. Bid below your benchmark and you keep 50, 65 or 70 percent of the savings by star band. Note the edge everyone misstates: four stars earns 65%, not 70% — the top band starts at 4.5. Average rebate in 2026: $222 per member per month. That money is every $0 premium, every flex card, every dental benefit. Fall one band and the budget shrinks before the bonus revenue moves an inch: the Stars cliff and the product budget are one mechanism wearing two names.
2. The clock is the constraint nobody models. Final rate lands the first Monday of April; bids lock the first Monday of June; one narrow reallocation window opens in late July; the market answers in October. And the Total Beneficiary Cost tether — a $40 per-member-per-month year-over-year threshold — chains each design to last year's. Product is where the rate cycle's whiplash becomes irreversible.
3. No one has ever measured enrollment per benefit dollar. Not once, causally, in the public literature — we checked. The strongest causal finding in Medicare Advantage choice is about price: demand jumps discontinuously at a $0 premium. Everything about benefit richness is associational — who enrolls where, never why.
4. What is measured is inertia. In the cleanest natural experiment available, more than 75% of beneficiaries stayed in traditional Medicare against a $0-premium plan that dominated it on every dimension — and people aging into the program enroll at roughly five times the rate of incumbents switching. Enrollment response lives in age-ins and active shoppers, a minority. A projection that does not split them is averaging two populations with a five-fold behavioral difference.
5. The benefits that sell plans went unmeasured — until the regulator demanded receipts. GAO found CMS could not see supplemental-benefit utilization; the best public numbers are survey self-reports. Then the rules turned: special supplemental benefits now require a bibliography of supporting evidence at bid submission, and every enrollee receives a mid-year letter listing the benefits they never used. An evidence file before the benefit, a measurement after it — the federal government now runs the discipline the industry skipped.
6. Check which broker rules actually exist. The 2025 compensation reforms were stayed before ever governing an enrollment period and vacated in court in 2025 — yet the official regulation text still displayed the dead language. Decks citing the fixed-fee regime cite a rule that never operated. Distribution is part of the product, and so is reading the docket.
We built this as an instrument: the rebate ladder with its stars coupling priced in the design's own dollars, the clock and the tether, the choice evidence in one pane, and a veto on enrollment lift booked from benefit richness — because a member count built on a lift coefficient is built on a number nobody has ever measured. Browser-only; nothing entered leaves the machine.
A benefit design is a bet placed in June on a market that answers in October — size the bet by the evidence, not the brochure.
In your last bid cycle, what did one star band down cost the benefit budget — and had anyone priced it before the June lock?