Why it matters
Raw cost doesn't tell you whether a group is well run. A group with more members managing diabetes, heart disease or cancer will cost more than a younger, healthier group with the same plan, no matter how well either one is managed. Comparing their PMPM side by side rewards the healthy group and blames the sick one.
Risk adjustment fixes the comparison. It estimates what each group should cost given who's in it, then compares actual spending with that expectation. The same idea sets value-based care targets, shapes benchmark reports your broker or carrier may show you, and moves billions of dollars a year between insurers in the ACA markets.
How a risk score works
A risk score is a single number that says how much care a person or group is expected to need, relative to an average.
- 1.0 is average. A score of 1.20 means about 20% more expected cost than average; 0.90 means about 10% less. "Average" is the population the model was normalized to, so a score only means something next to that population (American Academy of Actuaries).
- Inputs. Age, sex and diagnoses are the core. Diagnoses come from claims, grouped into condition categories. Some models also use prescriptions or prior cost.
- Models. Common ones include the federal HHS-HCC model used in the ACA markets, Milliman MARA, Cotiviti DxCG and Johns Hopkins ACG. Scores from different models, or different versions of the same model, aren't comparable.
- Prospective or concurrent. A prospective score uses this year's data to predict next year's cost; it's the kind used to set targets. A concurrent score uses this year's diagnoses to explain this year's cost; it's the kind used to judge how a year went. Ask which one you're looking at.
Real benchmarks adjust for more than health. Area (prices differ by market) and plan design (richer plans get used more) are usually adjusted separately, so make sure a comparison isn't mixing a rich plan in one city with a lean plan in another.
Worked example: same bill, opposite stories
Two illustrative employer groups with the same plan design and area, scored by the same model. An average member (score 1.0) costs $600 PMPM.
| Measure | Group A | Group B |
|---|---|---|
| Actual cost PMPM | $600 | $600 |
| Risk score (1.0 = average) | 0.90 | 1.20 |
| Expected cost ($600 × score) | $540 | $720 |
| Actual minus expected | +$60 (11% over) | −$120 (17% under) |
| Risk-adjusted cost (actual ÷ score) | $667 | $500 |
On raw cost the two groups look identical. Adjusted for risk, Group A, the healthier group, is the one spending more than expected, and Group B is spending less. The last row says the same thing another way: divide actual cost by the score and Group A costs $667 for an average-risk member, against Group B's $500.
Read the transcript
Two employers each spend six hundred dollars per member per month. Same cost. But one has more members with diabetes, heart disease and cancer. Which one is spending more than it should? Risk adjustment is how you tell.
A risk score predicts how much care a group will need, based on age, sex and diagnoses. One point oh is average. Group A scores point nine. Group B scores one point two.
If an average member costs six hundred, Group A was expected to cost five hundred forty. It spent sixty dollars more. Group B was expected to cost seven hundred twenty. It spent a hundred twenty less. Same bill, opposite stories.
The same math sets value-based care targets, so a doctor group that takes sicker patients gets a higher target, not a penalty. And it moves real money: in twenty twenty-five, small group insurers with healthier members paid about one point three billion dollars to insurers with sicker ones.
The catch: a risk score only sees what's written down. More complete coding raises the score even if no one got sicker, and costs look better against it. A rising score isn't always a sicker group.
So ask your broker three things. What's our risk score, and which model produced it? Are we compared to benchmarks on a risk-adjusted basis? And how much of our score change is coding, not health? Find out more at actuariallysound dot com.
Reading the result carefully
Under expected isn't automatically good. Group B might be managing care well, or it might have members putting off care, or it might just have had a quiet year. A small group's one-year results are noisy, which is why credibility matters: the fewer members, the less a single year proves. Look at more than one year before drawing conclusions, and look at what's driving the gap.
Where risk adjustment moves money
Value-based care targets. In a value-based care contract, a provider group earns a bonus for coming in under a spending target. Risk adjustment raises the target for patients expected to need more care, so a doctor group that takes on sicker patients gets a higher target, not a penalty.
ACA small group insurers. In the ACA individual and small group markets, insurers whose members have lower-than-average risk pay into a pool, and insurers with higher-than-average risk are paid from it. For the 2025 benefit year, small group transfers totaled $2.68 billion in absolute value, about 4% of small group premium. Because transfers net to zero, that's about $1.34 billion paid by lower-risk insurers to higher-risk ones (CMS, June 30, 2026). It's one reason an ACA small group insurer has less to gain from attracting only healthy groups.
These transfers apply to fully insured ACA small group plans. Self-funded and level-funded plans aren't part of them, though your carrier or stop-loss insurer may still use risk scores to price your renewal or benchmark your results.
The catch: a score only sees what's coded
A risk score is built from diagnoses written on claims. If doctors start documenting conditions more completely, the score goes up even though no one got sicker, and costs look better against the higher expectation. CMS says as much in its own report: year-over-year risk score changes can come from model changes, enrollment, population health and coding practices, so a change in the score doesn't necessarily mean a change in how sick people are.
This matters most where a higher score means more money, as in value-based targets. Medicare's Shared Savings Program limits how much risk score growth can raise an accountable care organization's benchmark. For agreement periods starting in 2024 or later, the cap is the growth in demographic risk scores plus 3 percentage points (42 CFR 425.610). Commercial contracts may or may not have a similar limit.
When a group's risk score jumps from one year to the next, ask how much came from new members, how much from aging, and how much from new diagnoses on existing members. Only part of that last piece is a real change in health.
Common misunderstandings
Questions to ask your broker
- What's our risk score, and which model produced it? Ask for the model name and version, whether the score is prospective or concurrent, and which year's data it uses. Without that, the number can't be compared with anything.
- Are we compared to benchmarks on a risk-adjusted basis? If a report shows your PMPM against a "norm," ask whether the norm is adjusted for your members' risk, age and sex, your area and your plan design, and whether both sides were scored by the same model.
- How much of our score change is coding, not health? Ask to split the change into new and departing members, aging and demographics, and new diagnoses on members who were already there.