Health Plan Funding Simulator Methodology and Limitations

1. What the tool does & who it’s for

Choosing how to fund a group health plan is one of the highest-stakes decisions a small or mid-sized employer makes each year. The same benefits can be delivered through very different financial structures — and each structure carries a different mix of predictable cost, upside in a good claims year, and exposure in a bad one.

The Benefitra Funding Simulator exists to make that trade-off visible. You describe your group — its size, state, the ages of its members, and what you pay today — and the tool estimates what each funding path is likely to cost you, not just next year but across a five-year horizon, including the unlucky years. It is built as a renewal-trajectory tool: you already know roughly what you pay now; what you can’t see is how each funding type’s cost evolves over five renewals, and how volatile that path is.

Who it’s for and the benefits

• Small and mid-sized businesses (roughly 5 to 500+ employees) weighing whether to stay fully-insured or move toward level-funded, self-funded, captive, or pooled arrangements.
• Brokers and advisors who want a transparent, sharable way to frame the funding conversation with a client, improving understanding of the trade-off between protection and control.
• Anyone who wants to pressure-test a carrier quote or a renewal against a modeled range before signing.

The health plans and funding arrangements it compares

The engine prices seven funding arrangements with the full Monte Carlo simulation, and adds an eighth by a button — a deterministic Comparable-ICHRA benchmark. Each of the seven is a genuinely distinct arrangement with its own rating rule and its own renewal behavior; the tool does not collapse any two into one. The eight are:

These are distinct arrangements, not relabelings of each other, and each renews on its own rules. Level-Funded and Self-Funded are two different ways to self-fund: Level-Funded is self-funding inside a carrier package (bundled stop-loss and a TPA, fixed monthly payments that offer predictable monthly expenses, part of a good year’s surplus refunded) carrying its own aggregate corridor billed monthly, while Self-Funded is traditional self-funding where the group funds claims directly — they price and renew differently (Section 3 explains how). Unbundled self-funded arrangements can also let an employer choose best-in-class providers instead of taking a single carrier package. Taft-Hartley and MEWA are two different pooled-trust presentations: the Taft-Hartley pooled trust renews on a bounded distribution around a selected average renewal (the same pooled result for everyone), while a MEWA can be experience-rated on each group’s own claims, so they do not renew alike. Separately, the tool can add an eighth arrangement by a button — a Comparable-ICHRA employer-cost overlay, a deterministic benchmark against the individual-market premium rather than a full simulated path — so you can size an ICHRA against the seven modeled arrangements.

The tool respects state law: where a state’s statutes restrict small-group stop-loss (and therefore self-funding), those paths are automatically removed from your comparison rather than shown as if they were available to buy. The self-funded captive is also withheld for the very smallest groups, below the program’s enrollment floor.

See also: the simulator itself · how the funding types differ & which fits your risk · stop-loss & self-funding rules by state.

2. How the health plan funding simulator methodology works

The engine is a Monte Carlo simulation. The methodology follows sequential modeling steps that turn your inputs into five-year financial projections. Rather than assume one “expected” claims year and multiply it out, it plays your group’s next five years forward thousands of times, drawing a different, plausible claims experience each time. The tool auto-scales the number of runs to your group size — between 2,000 and 10,000 Monte Carlo runs: about 10,000 for groups up to 50 lives, 5,000 up to 200, and 2,000 beyond, because the small groups are where the tail is lumpiest and needs the most runs to resolve. (The run count is also user-adjustable, and the simulator can rerun new scenarios in seconds when a variable is changed.) The spread of those runs is the answer; a single-number estimate throws that spread away.

The condition-incidence catastrophe engine

The most important thing to understand about this model is where the risk comes from. Older calculators assume a single blended “catastrophe” number and stretch it with a volatility knob. This tool instead simulates the actual events. Each year, for each member, it draws 17 named catastrophic conditions — solid and blood cancers, ESRD / dialysis, organ transplant, cardiac events, congestive heart failure, NICU / prematurity, hemophilia, MS, autoimmune disease, cystic fibrosis, one-time gene therapies, major joint and spine surgery, stroke, and sepsis — plus a catch-all “other catastrophic” bucket for high-cost events outside that list (18 condition keys in all), each at that member’s age- and health-adjusted incidence rate.

The draw is a Poisson process over each member’s per-year probability, so a 30-life group either has a catastrophic case this year or it doesn’t — there is no smooth average that hides the lumpiness. When a condition fires, its cost is sampled from a right-skewed distribution (most cancers are expensive, a few are catastrophic), and the case is carried forward with multi-year persistence: a year-1 dialysis, cancer, or transplant loads years 2 through 5 the way it would in a real book. The heavy upper tail then emerges from the arithmetic — small pools are lumpy and dangerous, large pools smooth out — rather than being dialed in by hand. (The tool reports that tail out to the 95th percentile; the extreme 99th-percentile edge is generated internally but not charted, because at the run counts above it is too noisy to read as a decision number.)

The three-component claims model

Underneath, each simulated year’s claims are the sum of three parts, each behaving differently:

1. A deterministic baseline — the floor of routine, always-present spend that keeps even a good year from collapsing to zero.
2. A chronic / routine component — moderate, correlated year-to-year variation from the ordinary ebb and flow of a covered population’s health, modeled with memory so a bad chronic year tends to be followed by another.
3. A catastrophe component — the heavy-tailed part, now produced entirely by the condition-incidence engine above. This carries most of the year-to-year variance. It is what the adverse-year and catastrophic-year outputs are really measuring.

The three components are balanced so that, on average, they reproduce the group’s expected claims, but the structure is built to help identify which component is driving volatility and to calculate the average claims level separately from the tail risk. That is a deliberate, evidence-based choice: in real self-funded books, roughly 1% of members drive about a third of total spend, so the risk in health-plan funding lives overwhelmingly in that tail.

An honest word on what the condition engine does and doesn't move. Because the model pins each run’s average catastrophe cost to the group’s expected budget, the per-condition detail does not swing the headline (typical-year) number much. What it changes is the shape of the bad-year tail (the upper 75th–95th-percentile outcomes) and the read on whether your group is healthier or sicker than the community pool — which is exactly where the funding decision is actually won or lost. This is a deliberate modeling trade-off, and we call it out so you know what the engine is and isn’t precise about.

The 5-year risk-adjusted health outcomes recommendation

The tool’s “most likely to perform well over five years” pick is explicitly risk-adjusted, meant to help decision-makers evaluate options rather than just pick the cheapest average path. Each eligible path is scored on its expected five-year cost, penalized by its downside (how much worse a bad-luck run runs than expected at your group’s size), and aligned with the structural signal from Section 3 — whether your own risk sits above or below the community pool. In plain terms: it favors the path that pairs a low expected cost with a tight, contained downside, not the one that merely looks cheapest in an average year while hiding a fat tail. A healthy 300-life group scores strongly toward self-funding; a healthy 15-life group toward level-funded (where stop-loss caps the volatility); a sicker small group toward the shelter of community-rated fully-insured, which can lead an employer toward more protection or more control depending on the risk picture.

3. Community rating & funding-type pricing

This is the part that makes the tool different from a generic cost calculator, and it is worth reading slowly. Simulation can also support benefit-design work by projecting enrollment changes when plan adjustments are made across different categories. In one simulation study, modeling projected 3,700 new members for a plan and generated $6.4 million in incremental profit, showing how scenario work can influence funding decisions. The same simulated claims experience produces very different employer costs depending on the funding path — because each path responds to your group’s own health through an entirely different mechanism. Pricing every path off one blended cost model would be actuarially wrong. The tool prices each one by its real rating rule.

Community-rated fully-insured is blind to your group’s health

Under the ACA’s adjusted community rating, a small-group fully-insured premium may vary only by age (capped at 3:1 across adults), geographic area, tobacco use (up to 1.5:1), and family tier — and nothing else. For employers, the appeal of this structure is that monthly expenses are predictable, even though control and transparency are more limited. Your renewal is driven by the carrier’s index-rate change (the whole pool’s experience plus medical trend) and one year of your group aging, not by what your members actually cost. In the model, the condition-incidence engine still runs for a community-rated group — but only to compute your own expected claims for the comparison; it never prices this path.

Everything else is priced to your own risk

Large-group fully-insured, PEO, and an experience-rated MEWA are experience-rated, giving the employer more ability to shape plan design than community-rated fully-insured: your own claims move the renewal, damped toward the manual rate by a size-based credibility weight (a 50-life group’s single year is only modestly credible; a 6,000-life group’s is fully credible). This is why an experience-rated MEWA does not renew like the Taft-Hartley pooled trust, which behaves like a community pool and renews on a bounded distribution around a selected average renewal (default ~3%) shared by everyone. Level-funded and self-funded are two distinct ways to self-fund and renew on different bases: Level-Funded is self-funding inside a carrier package with stop-loss, medically underwritten at bind — which is exactly why a healthy small group can be offered a level-funded rate below community-rated fully-insured — and it carries its own aggregate corridor (default ~120%) billed monthly, so it renews on the package’s own basis; some employers partner with a major carrier in these bundled arrangements, while others retain more direct control in traditional self-funding. Self-Funded plans fund their own claims directly up to a per-person stop-loss deductible, plus fixed costs and stop-loss premium, with an aggregate attachment capping the worst-year total, and renew on their own basis (not the level-funded corridor); the Self-Funded Captive adds a partly-refundable shared-risk layer. In a good year, level-funded, self-funded, and captive groups keep part of what they don’t spend in claims — a surplus the pooled and fully-insured paths never return. For large-group fully-insured, PEO, and MEWA structures, the employer is relying on administrators or carriers to administer the plan under the chosen model. For all of these, the incidence engine drives both the level and the volatility.

“Level-Funded” and “Self-Funded” are distinct arrangements that renew differently

Both are ways to self-fund, but they are distinct arrangements and the tool prices and renews them separately; traditional self-funded arrangements can be beneficial for employers that want more customization and vendor choice. Level-Funded is self-funding inside a carrier package: bundled stop-loss and a TPA, fixed monthly payments, and part of a good year’s surplus refunded. It carries its own aggregate corridor (default ~120%, billed monthly) and, because that convenience is priced, typically costs on the order of 3–10% over traditional self-funding. Self-Funded is traditional self-funding: the group funds claims directly, retains claims only up to the specific stop-loss deductible with the aggregate attachment capping the year, and is priced on its own basis — not the level-funded corridor. At entry the ladder runs roughly Self-Funded ≤ Self-Funded Captive ≤ Level-Funded (the carrier-packaged convenience of level-funded costs a little more up front), and each then renews on its own rules. Because the level-funded package is how smaller employers most often self-fund in practice, the tool defaults to showing Level-Funded for groups of 100 or fewer lives and Self-Funded above that — but that is only a default; you can toggle to compare the two side by side.

The insight the tool is built to surface

Because community-rated fully-insured charges every small group the pool’s average, a healthier-than-average group is overcharged — it subsidizes the pool, and can often price below that premium by self-funding or level-funding to its own low risk. A sicker-than-average group is undercharged — the pool subsidizes it, so community-rated fully-insured is a genuine hedge worth keeping while the group is eligible. In short: healthy groups tend to self-fund; sicker groups tend to shelter in community-rated fully-insured. The tool measures which side of that line your group is on and says so plainly — that is the single decision it exists to inform.

Which side you land on depends on your group’s real epidemiology, which is why the census upload matters (Section 4). Without a census, the tool estimates your position from age and a health selector; with one, it measures it member by member. State law can override the recommendation: where self-funding isn’t legally placeable at your size, even a very healthy group is pointed to the arrangements it can actually buy.

4. Your inputs — census & claims

The tool is designed to run on your numbers, not a synthetic average. Accurate inputs are critical because the simulator is built around your organization’s actual funding situation rather than a generic average. Defaults exist only as clearly-labeled fallbacks (“Modeled — edit to make it yours”) for the illustrative, no-input case. The more you enter, the sharper the anchor becomes.

That better fit supports stronger financial decisions and, indirectly, better employee health outcomes.

What you can provide

• Current premium and funding path - anchors the cost level every path is measured against, in your own dollars.
• A census - the age (and, where provided, gender) of each member. This sets each member's exact incidence rates in the catastrophe engine and pins down whether your group is healthier or sicker than the community pool.
• Claims data - known paid-claims dollars, a loss ratio, or paid PMPM. This is the strongest anchor there is for the experience-rated paths: the tool scales its modeled distribution so its average matches your actual claims, keeping the epidemiological shape while your dollars set the level.
• Optional refinements - a stop-loss quote (specific deductible and aggregate attachment), last renewal percentage, tobacco share, family-tier mix, and exchange benchmark information for employers comparing an ICHRA-style alternative, each of which sharpens the relevant path.

Your data never leaves your browser. Census and claims files are parsed and simulated 100% on your device — they are never transmitted to Benefitra or any server, so no protected health information (PHI) ever leaves your computer. Nothing to store means nothing to breach. As an extra guard, if you upload a file that contains a personal-name column, the tool rejects it on the spot and asks you to remove it — the model only ever uses ages, coverage tiers, and claim amounts, never anyone’s name or identity. For a tool that touches health-adjacent employer data, that is a deliberate privacy and trust decision, not an afterthought.

On the roadmap - and how we'll keep the promise. We plan to add a server-side step that protects the model’s proprietary calibration math. It is not live yet, and it is being designed so the guarantee above still holds: only de-identified aggregate statistics — never a raw census, never PHI — would ever be sent to a server. If and when it ships, we’ll say so plainly here; until then, everything runs entirely in your browser.

Every default-derived number carries a visible “Modeled” tag; numbers you enter are highlighted as yours. Missing inputs degrade gracefully — claims, then premium × a labeled expected loss ratio, then a labeled community default — and the tool always tells you which one it used.

5. The cost-containment overlay

For the self-funded and captive paths, the tool offers an optional overlay to help address avoidable high-cost claims in self-funded arrangements by modeling clinical cost-containment strategies — the levers a self-funded employer can actually pull to bend the bad-year spike down. It is off by default; when you turn it on, you can explore how much thinner the catastrophic tail gets, and which conditions drive it. Because the tail is generated by real conditions (Section 2), each lever haircuts the realized condition mix of that scenario, so savings scale naturally — and reductions occur differently depending on which conditions actually show up, small for a healthy or large group, larger for a hot small group.

The overlay names no vendors and focuses on operational strategy rather than vendor-branded services. It models the mechanism — for example, “direct cash-pay” rather than any particular cash-pay network — so the estimate reflects the strategy, not a sales pitch. Two levers work differently from the rest: reference-based pricing and direct cash-pay reprice facility care plan-wide rather than trimming one claim type, so they are modeled against a broad slice of spend (on the order of 15–30% of total annual cost), not just the catastrophe tail. And because a whole-plan RBP or cash-pay model replaces the conventional PPO relationship, the tool applies exclusivity rules — it won’t stack two standalone strategies that couldn’t actually co-exist in one real plan design.

How it’s modeled — honestly

This feature does not invent savings. It works by re-attributing and estimating, in four disciplined steps:

1. Read the realized tail. For each simulated bad year, the engine already knows which conditions actually fired and what they cost — cancer, cardiac, MSK, specialty Rx, NICU, dialysis, transplant, autoimmune, and the rest. No fixed share table; the mix is whatever that run drew.
2. Reach only what a lever can touch. Each strategy affects only the addressable fraction of the conditions it actually reaches — COE the surgical slice, specialty-Rx management the drug slice, reference-based pricing and direct cash-pay the facility slice, and so on, with careful consideration of what is truly addressable; some interventions also work partly through bulk purchasing, which can reduce direct costs for certain plan inputs or consumable items.
3. Apply a sampled, ranged haircut. The saving on that reachable slice is drawn from a min–mode–max (PERT / triangular) distribution, not a fixed number, so the uncertainty flows into the output bands instead of producing false precision.
4. Stack on the residual. Strategies apply sequentially to what’s left after the prior lever, so overlapping levers don’t double-count. There is no arbitrary savings ceiling: the total is bounded only by each strategy’s evidence-based rate and by how much of that scenario’s cost is genuinely containable — a share that emerges from the realized mix. In a healthy year it is a small slice; in a bad year for a small group dominated by one catastrophic claim it can be much larger, because that is genuinely how concentrated and containable those dollars are.

In broader health-program planning, diverse revenue streams can improve financial sustainability, even though this tool models employer plan funding rather than program revenue.

With every applicable lever turned on for a mid-age group, the modeled reduction in catastrophic-tail dollars lands in roughly the 15–22% range, consistent with published, de-overlapped estimates. The strongest evidence sits behind Centers of Excellence (RAND and peer-reviewed bundled-surgery studies) and the reference-based-pricing / direct cash-pay price gap (facility prices well below commercial benchmarks). Softer, more vendor-reported evidence sits behind care navigation and aggressive specialty-Rx or high-cost steering — which also carry patient-access and legal caveats.

Read this as ranges, not promises. Each lever’s saving is uncertain and depends on your vendors, your population, and how the strategy is actually implemented. Some levers rest on softer or vendor-reported evidence and carry legal and access caveats. The overlay is a way to explore what containment could do to your tail — not a projection of what it will do. Modeled savings are ranges, not guarantees.

6. Reading the results — what you get

A distribution of five-year costs is only useful if you can act on it. This build wraps the engine in an interpretation layer designed to make the trade-off legible without hiding the uncertainty behind it.

The percentiles you see

For each funding path the tool reports the distribution of simulated costs at five percentiles: the 5th, 25th, 50th (median), 75th, and 95th. On the chart the median is the navy line, the 25th–75th percentiles form the darker inner band (the middle half of outcomes), and the 5th–95th percentiles form the lighter outer band — a 90% planning range. Read the 5th percentile as a good-luck year, the 95th as a bad-luck / catastrophic year, and the median as the honest “typical” outcome. We deliberately stop the reported range at the 5th and 95th percentiles: the 1st- and 99th-percentile extremes are generated by the engine but are too noisy at these run counts to show as decision numbers, so charting them would imply a precision the model doesn’t have.

Trend to renewal, not just a five-year total

For each funding path the tool shows a projected average annual renewal percentage alongside the five-year totals — so you can see the trajectory that separates the arrangements, not just where they land in year five. A path that starts cheap but renews at 12% a year can cost more by year three than one that starts higher and renews at 6%; the renewal view makes that visible.

A plain-language “why this wins”

The risk-adjusted recommendation comes with a written, plain-English explanation of why a given path is favored for your group — your health-relative position, your size, and your tail exposure — rather than a bare ranking you have to reverse-engineer.

Two ways the numbers can be framed — conservative vs. documented client results

The tool can present its comparison on one of two bases, and it labels which one is showing:

• Conservative model (the default). Every alternative path is priced by the data model described above, anchored to your own premium. This is the cautious read - it does not assume you achieve best-in-class results, and it is the basis you should treat as the honest planning range.
• Documented client results (optional). A toggle re-frames the alternatives to the average outcomes Benefitra has actually documented across its placements, developed from real placement experience rather than assumed best-case savings - roughly a PEO cost about 28% below fully-insured (applied only at 50 or fewer enrolled, where that field evidence sits), a captive about 18% below fully-insured, and self-funded about 1% under the captive. It is a way to show what actively-managed groups have reached, not a promise that yours will.

In both modes, any real quote or your current-path premium you enter always takes precedence over the modeled figure — the mode only re-frames the paths you haven’t priced with your own numbers. The conservative basis is what you should rely on for planning; the client-results view is context, clearly marked as such.

Sensitivity & what-if — see what actually drives the answer

Because no model input is certain, the tool includes a sensitivity view to help determine which assumptions most affect the recommendation: a tornado chart ranking the drivers that move your result most (group size, how healthy your group is, medical trend, stop-loss terms, containment), plus live what-if sliders that let you nudge an assumption, evaluate new scenarios immediately, and watch the recommendation and the cost bands respond in real time. If the answer flips the moment you touch one slider, that fragility is something you should see — so we show it.

Benchmark costs context

The tool places your current cost per member against public benchmarks so you know whether you’re starting high, low, or typical. The reference point is the KFF Employer Health Benefits Survey 2025: the national average single-coverage premium was about $9,325/year, with small firms near $9,211 and large firms near $9,361. Seeing your PMPM against those bands frames the whole comparison before you weigh funding paths.

A guided first run, a shareable report, and a broker mode

• Guided onboarding. A short first-run walkthrough orients you to the inputs and to what the output bands mean, so the first result isn't a wall of numbers.
• PDF report export. You can export a shareable PDF report of your run (email-gated) to bring to your broker, CFO, or advisor.
• Broker version. A separate, access-gated build for advisors exposes editable per-condition calibration - so a broker can tune incidence and cost assumptions to a specific book. It's kept private and distinct from this public tool, so the default calibration you see here is never quietly altered underneath you.

7. Validation & accuracy

A model is only as trustworthy as its track record, and because validation is a critical part of any methodology used for health care funding decisions, we would rather show you an honest, partial one than a polished, overstated one. Here is exactly where the validation stands today.

A preliminary, out-of-sample backtest

We replayed the funding decision through public market data — the KFF Employer Health Benefits Survey and Peterson-KFF series, 2013–2025 — as if we were making the call at the time, then scored whether the path the tool would have favored for each size band went on to deliver the better multi-year outcome. The definition of “correct,” the out-of-sample rule, and the size-band segmentation were fixed before the run, so the result can’t be reverse-fit.

Preliminary result - directional

The tool’s size-band ordering is directionally consistent with what the public market data shows actually happened — strongest for large groups, solid for mid-size groups, and genuinely ambiguous for the smallest groups.

This is a directional signal, not a validated per-employer hit-rate. Aggregate public data cannot honestly produce a single “X% accurate” number — it shows trend gaps and migration, never two counterfactual cost paths for the same employer. We publish the ambiguous small-group result alongside the strong large-group one on purpose: showing where the tool is weak is the credibility.

Underwriter review

Beyond the backtest, the model’s pricing logic — how each funding path is rated, how the incidence engine feeds the tail, and how renewals are projected — has been reviewed and validated by underwriter Borka Skoric. That is a working underwriting review of the methodology, not a formal actuarial opinion: the methodology is documented so that it can be put in front of an independent actuary, and independent actuarial review remains a stated goal below rather than something we claim to have completed.

The Renewal Index and the path to peer review

The community-rated renewal branch of the model is driven by a transparent, versioned Benefitra Renewal Index — developed so users can see the basis for the community-rated branch, using state filed small-group renewal trends, decomposed and updated over time — rather than a black-box forecast. It is presented as a probabilistic prior with error bars, never a prediction of future healthcare costs. Beyond the preliminary backtest, the credibility roadmap is deliberately sequenced: independent actuarial review of the methodology, a firmer segmented hit-rate once per-employer data (state DOI filings, real claims) is available, with future review needing to account for changing regulations as additional data sources become available, and only then a peer-reviewed write-up. We will publish misses alongside hits at every stage. A funding tool that hides its failures is the one you should distrust.

The backtest is a fast, ongoing workstream; independent actuarial certification and a peer-reviewed publication are separate, slower efforts that will be linked here as they land. No accuracy figure will be presented as validated before it is genuinely earned.

8. Limitations — please read before you rely on this

We put this section front-and-center on purpose. The honesty here is the credibility. A funding decision is a health-and-money (YMYL) decision, and this tool is one input to it — not the answer.

What this tool is - and isn't

• It is decision support, not advice. It does not create an advisor-client relationship, is not an insurance quote, and does not recommend a specific product for your specific circumstances. Treat every number as a modeled estimate to discuss with a licensed professional.
• It produces honest ranges, not forecasts. Every output is a distribution of possibilities. The tool cannot tell you what will happen - only what is plausible across many simulated futures.
• It cannot predict individual claims. A single unforeseeable catastrophic claim can land you well outside even the modeled bad-year band. The tail is modeled statistically, from population incidence - not clairvoyantly.
• Its assumptions may not fit your group. Incidence rates, condition mix, cost distributions, and renewal behavior are calibrated to population-level research and typical groups. An unusual demographic, industry, geography, claims history, internal departments, or workforce pattern can make the defaults a poor fit - which is why the tool lets you replace them with your own census and claims.
• The numbers are modeled estimates, not adjudicated dollars. Costs are commercial-billed figures reduced by a network-discount factor and thinned by severity gates; recurrence and persistence rules are the softest inputs and are tunable, not settled facts.
• The condition engine shapes the tail, not the headline. By design, the average catastrophe cost is pinned to your expected budget, so the per-condition detail sharpens the bad-year tail and the healthy-vs-sick read - it does not make the typical-year number more precise. That's a deliberate trade-off, and it means the value of the detailed engine is in the risk picture and the recommendation, not in the central estimate.
• Your data stays in your browser - and the rest is on you. Files are processed entirely on your device and never sent to a server, and name columns are rejected automatically; but client-side processing can't police everything, so keep identifiers out of the file, and treat the file on your own machine as your responsibility. A planned server-side step (for calibration math only) is not live yet and, when it ships, is designed to send only de-identified aggregates - never raw data.
• It is not a substitute for a licensed advisor or actuary. Real funding decisions require underwriting, real stop-loss quotes, plan-document review, and professional judgment the tool doesn't have.
• The cost-containment estimates are modeled ranges. They depend on vendor performance and implementation, some rest on softer evidence, and several carry legal and patient-access caveats. They are an exploration of potential, not a promise of savings.
• Data has lag and gaps. Public sources are periodic and lag the market; state rules change. Figures reflect the best public data available as of the last-updated date, not real-time conditions.
• The validation is preliminary. The backtest is directional and built on public aggregates, not a peer-reviewed, per-employer hit-rate. Do not read it as a certified accuracy figure.
• Structures carry non-modeled risks. For pooled arrangements (MEWA / Taft-Hartley), the modeled range holds only while the arrangement stays financially healthy; the tool cannot see a specific trust's solvency, operations, or financial health directly.

Bottom line. Use the simulator to frame the conversation and understand the shape of the trade-off — then take a real quote and a licensed advisor before you decide. If a number here ever seems too good to be true for your situation, it probably is; ask us to walk you through the assumptions.

10. About the author

Last updated: August 18, 2026

References

• Kaiser Family Foundation - Employer Health Benefits Survey (cost, renewal & funding-prevalence series).
• Society of Actuaries - health trend, age-curve & stop-loss research.
• U.S. Department of Labor - ERISA & self-funded health plans.
• NCI SEER and USRDS - cancer and ESRD incidence and cost, used for the condition-incidence engine.
• High-cost claims, per-case cost & small-group renewal trend - Sun Life, High-Cost Claims and Injectable Drug Trends (2023, 2025, 2026 editions) and Milliman, 2025 U.S. Organ and Tissue Transplant cost report (Feb. 2025), for claim concentration and per-case catastrophic cost; League RJ et al., "Assessment of Spending for Patients Initiating Dialysis Care,"JAMA Network Open 2022;5(10):e2239131; Peterson-KFF small-business briefs, 2026 and 2027, with state DOI / RateReview.HealthCare.gov filings, for small-group renewal trend; and the KFF Employer Health Benefits Survey 2025 for the premium benchmark.
• Cost-containment evidence - RAND Corporation analysis of a commercial Centers-of-Excellence / bundled-payment program (Carrum Health), published in Health Affairs, for bundled-surgery savings and conservative-care redirection; and Murray RC & Whaley CM, "Commercial prices relative to Medicare were lower under an employer-based reference pricing program,"Health Affairs Scholar 2026;4(7):qxag156, for reference-based-pricing savings. This evidence also informs methodology development where cited savings and care-delivery findings intersect with ongoing clinical practice.