The Iceberg Review

Most of the people on your health plan cost very little, and almost all of the money sits with a small number of them. That is not news. What the Iceberg Review checks is whether it is true of your plan, and by how much, because the answer changes what is worth arguing about at the next renewal.

What does the iceberg actually show?

Two running totals, both climbing from left to right as claims get larger. One counts members. One counts dollars. By $10,000 a year the member line has reached 85% and the dollar line has reached 24%. The other 15% of members carry the remaining 76% of the dollars.

The vertical distance between the two lines is the argument. It is widest between roughly $5,000 and $10,000, which is also about where most benefits committee agendas stop looking.

The shape of that curve, and what actually sits in the expensive end of it, is the subject of what drives employer health plan cost.

Which three numbers do you need?

Claimant counts and dollars above three lines: $10,000, $50,000 and $100,000. Plus your covered member count, meaning employees and dependents rather than employees alone, and your total claims for the year.

Pull them on an allowed basis, not plan paid. Allowed is what the plan and the member together owe after network pricing. Paid is the plan’s share after the deductible and coinsurance come out.

The distinction matters because member cost sharing is concentrated at the small end. Strip it out and you remove proportionally more money from the low cost members than from the high cost ones. So a paid basis report read against this benchmark shows fewer claimants and a larger share of dollars than the plan really has, which reads as a dramatic concentration story and is an artifact of the basis.

What does the gap tell you?

Start with what the curve cannot do. The shape tells you where to look first, not what you will find when you get there. A plan can be concentrated because its hospital prices are bad rather than because its clinical management is weak, and a dispersed plan can still have real clinical opportunity sitting untouched. The curve narrows the search. It does not name the cause.

Given that, there are three outcomes and each one points somewhere different.

Above the benchmark is a concentration problem, so the first place to look is clinical and contractual rather than plan design. Below the benchmark is a volume problem, so site of care, steerage and unit price are the likelier ground. On the benchmark is an ordinary shape, in which case the largest band, $10,000 and up, is where three quarters of the money already is.

Download the worksheet

Can you have AI do the arithmetic?

Yes, and the prompt below is written for it. Two cautions before you paste anything.

Use aggregate figures only. Counts and dollar totals, nothing else. No member names, no dates of birth, no member IDs, no claim numbers, no diagnoses, and no claim-level file. Use whichever assistant your organization already permits rather than whatever is open in another tab.

I am the plan sponsor of a US employer-sponsored group health plan. Help me read my claims distribution for one plan year. Work only from the aggregate figures below. Do not ask me for member names, dates of birth, member IDs, claim numbers or diagnoses, and if I paste anything like that, tell me to remove it before continuing.

MY FIGURES

- Plan year:
- Covered members, meaning employees plus covered dependents, not employees alone:
- Are the dollar figures below ALLOWED amounts or PLAN PAID amounts? (answer allowed or paid)
- Total claims for the year, on that same basis:
- Claimants over $10,000, count: ____ dollars for those claimants: ____
- Claimants over $50,000, count: ____ dollars for those claimants: ____
- Claimants over $100,000, count: ____ dollars for those claimants: ____

DO THIS, IN ORDER

1. Check my inputs before you calculate anything. Tell me if my counts or dollars do not decline as the threshold rises, if my claimant counts exceed my covered members, or if my dollars above a line exceed my total. Stop and ask rather than proceeding.

2. If I answered PAID above, tell me plainly that the benchmark in step 4 is stated on allowed amounts and that the two are not the same quantity. Allowed is what the plan and the member together owe after network pricing; paid is the plan's share after the member's deductible and coinsurance. Tell me my member percentages will read low and my dollar percentages will read high against this benchmark, and ask whether I can get the report reissued on an allowed basis before we continue.

3. Calculate, for each of the three thresholds, my percentage of members and my percentage of dollars. Show the arithmetic.

4. Compare each one to this commercial benchmark: over $10,000, 15.0% of members and 75.9% of dollars. Over $50,000, 1.7% and 36.9%. Over $100,000, 0.6% and 25.9%. For each line, tell me whether I am more concentrated, less concentrated, or about the same, and by how many points.

5. Give me your read in plain language. More concentrated than the benchmark means my dollars sit with a handful of people. Less concentrated means my cost is spread across many members. Say which one I am, and say it in one sentence.

6. Based on that read, tell me what kinds of questions are worth my time at the next renewal and what kinds are probably not. Talk about categories of work, not products. Do not name or recommend any vendor, carrier, TPA or point solution.

7. Tell me what you cannot see from these figures. Be specific about what would change your read if you had it.

CONSTRAINTS

- Do not estimate savings. Do not produce a dollar figure for what any change would be worth.
- If I ask you to project next year's cost from this, decline and explain that three thresholds from one year cannot support a projection.
- Flag the small group problem: if my covered members number is under roughly 1,000, tell me that the $100,000 line will swing by several points year to year because a single claimant moves it, and that one year of data is not a trend.

The assistant can do the arithmetic and give you the read. It cannot tell you who owns the follow-up or by when, which are the two fields on page two that decide whether anything changes.

What does this not account for?

Your industry, your geography, the age of your workforce, your plan design and your network. The benchmark is a national commercial distribution, so it describes the shape of employer claims generally rather than the shape of yours.

It also has a small group problem worth knowing about. Under roughly a thousand covered members, the $100,000 line moves by several points from one year to the next because a single claimant carries it. One year against the benchmark is a conversation starter. Three years is a pattern.

Fringe Theory is independent and unaffiliated. Views expressed are my own and do not represent those of my employer. Nothing here is legal, tax, medical, or investment advice. Figures described as modeled or derived are my own arithmetic from the public sources listed, not disclosed data.