If you run benefits or finance at a company with a self-funded health plan, you have probably been handed a number called IBNR and asked to sign off on it. This article explains what that number is, why your plan has one, and how to read an estimate without being an actuary.
IBNR is an estimate of claims your plan already owes on care that has happened but that you have not yet paid. Some of those claims are sitting at your third-party administrator waiting to be adjudicated. Others have not been submitted yet. Your plan is on the hook for both, and at year end your auditor will want the number reflected in the financials.
The underlying concept is the same one used in property and casualty reserving. If you want the non-health foundational version first, read the plain-English IBNR explainer and come back for the translation to your plan.
Why self-funded health is not workers compensation
Most of the canonical reserving literature, including Jacqueline Friedland’s CAS study note Estimating Unpaid Claims Using Basic Techniques, uses workers compensation and auto liability as its primary worked examples. The exposure-base table that Friedland provides for self-insured entities lists payroll for workers compensation, vehicle counts for auto liability, and population or operating expenditures for public-entity general liability (Friedland, p. 132). Self-funded employer health is not in the table. The techniques still apply, but the patterns are different enough to matter.
About two thirds of US workers with employer coverage are now in self-funded plans according to KFF’s 2025 Employer Health Benefits Survey, and the share rises to roughly eighty percent at firms with 200 or more workers. The audience that needs reserving explained in plain language is no longer a niche one.
Four practical differences from a workers compensation self-insurance program shape everything that follows.
Tail length. Most health claims close within twelve to eighteen months of the date the service occurred. A claim incurred in January is usually paid by the following March and is almost always paid by the end of that calendar year. Workers compensation can pay out over a decade or more.
Claim frequency. Health plans process thousands of small claims. A mid-sized self-funded employer sees hundreds of thousands of claim lines per year. Workers compensation programs of the same dollar size see orders of magnitude fewer claims at much higher average severity.
Severity. Average severity on a health plan is modest except in the large-claim tail, which is where stop-loss insurance comes in. One catastrophic claim can reshape a year, and your stop-loss attachment point is the line that determines how much of that claim your plan retains.
Who sees the claim first. Your third-party administrator sees a claim when the provider submits it, not when the member walks into the clinic. The gap between the service date and the adjudication date is called the reporting lag, and it is the single most important input to a health reserve estimate.
The same math that Friedland uses for workers compensation reserving applies to health. The triangles are just finer-grained and shorter, and the things that break them are different.
What “IBNR” actually includes for a health plan
The label IBNR is doing a lot of work. In Friedland’s terminology, the broad definition of IBNR has two components (Friedland, p. 14):
- Pure IBNR, which is claims for services that have already happened but have not yet been reported to your TPA at all. A member saw a specialist on December 28 and the provider has not filed the claim yet.
- IBNER, short for incurred but not enough reported. These are claims that are in your TPA’s system but have not been fully adjudicated and paid. The claim is known, the final dollars are not.
When a reserving report quotes you an IBNR number, it almost always means the sum of both components. Friedland is explicit: “Throughout this book, unless specifically noted otherwise, we use the broad definition of IBNR” (Friedland, p. 14). Chapter 17 of the same study note returns to the broad and narrow definitions in the context of allocating unallocated loss adjustment expenses (Friedland, pp. 388 to 390), which is worth knowing if you ever need to reconcile an ULAE discussion back to this same distinction.
In health-plan operating language the same concept has a friendlier name: run-out reserves. It is the money your plan will pay out as claims incurred on or before the plan year close get submitted and adjudicated. Run-out reserves and broad-definition IBNR are the same number. Different words, different audiences.
A few things run-out reserves are not.
- Claim payable is the line on your balance sheet for claims your TPA has already adjudicated and queued for payment. It is not an estimate. It is a known number sitting in a check run.
- A cut-off accrual is a rough year-end adjustment that approximates unpaid claims without fitting a model to the historical lag pattern. Some small employers still close the year this way. Your auditor will ask whether it is defensible, and if your plan is material to the company’s financials the answer is usually no.
- Your ASO fee and stop-loss premium accruals are separate operating expenses. They live in different lines and follow different rules.
If your finance team confuses IBNR with any of the above three, the balance sheet conversation gets messy fast. The number you need to book against an audit is the actuary’s estimate of claims incurred on or before the reporting date that your plan has not yet paid. Nothing more.
How an actuary estimates it (the 90-second version)
Every common method starts with a lag triangle, which is a spreadsheet of paid claims organized by two dimensions. One axis is the incurred month (or service month) in which the care happened. The other axis is the number of months of run-out observed since that incurred month. Each cell is a cumulative paid-to-date number.
The actuary builds this triangle from data you already have. Most health plans use thirty-six to forty-eight months of history because that is typically enough for the run-out pattern to stabilize. Once the triangle is built, the age-to-age ratios of one column to the next describe how a typical incurred month completes.
From the triangle the actuary derives a completion factor for each incurred month. A completion factor is the fraction of ultimate claims you would expect to have paid by a given age. An incurred month that is three months old might be 94 percent complete; one that is twelve months old is very close to 100 percent. Ultimate claims for an incomplete month equal the paid-to-date amount divided by the completion factor.
Completion factors are the health-industry inverse of the cumulative development factors you see in property and casualty reserving. Same math, different vocabulary. A CDF of 1.33 is the same object as a completion factor of 0.75. If you want the CDF version first, the loss development triangle explainer walks through it without the health-plan framing.
Friedland devotes most of Chapter 7 to the underlying assumptions of this technique (Friedland, p. 84). The core one is that claims recorded to date will continue to develop in a similar manner in the future. The supporting assumptions are consistent claim processing, a stable mix of types of claims, and a stable retention structure. For a self-funded health plan, every one of those assumptions can fail in ways that a workers compensation program’s would not. We get to the specific failure modes next.
The actuary’s second step is a sanity check that does not depend on the triangle at all. The cleanest cross-check for a health plan is PMPM trend multiplied by member months. If your historical paid claims per member per month are trending at roughly inflation plus a known margin, and your member count for the period is known, expected claims for the period are the product of those two numbers. If the triangle answer and the PMPM answer are meaningfully different, there is something worth investigating before the number gets booked.
The five things that make a health-plan IBNR estimate go wrong
Friedland builds most of the discussion around a single idea: the development technique works when conditions are stable and it fails in predictable ways when they are not (Friedland, p. 84). A changing environment breaks the triangle’s assumptions and the projected ultimate drifts from the truth. The same framework applies to your health plan, just with different triggers.
1. TPA system or vendor change. The most common reason a health reserve estimate is wrong is that the TPA changed something about how it processes claims. A platform migration, a new claims system, a change in the medical review queue, a shift from in-house adjusters to outsourced ones. Any of these can speed up or slow down the payment lag without changing what the plan actually owes. A triangle built from the prior lag pattern will misestimate completion factors until several months of post-change data fill in. If your plan changed TPAs or had a system upgrade in the last twelve months, the actuary needs to know and the estimate should explicitly handle it.
2. Large-claim leakage through the stop-loss attachment. Health reserving is dominated by the small-claim majority, but the answer is sensitive to a handful of large claims that sit near your stop-loss attachment point. If your attachment moved at the last renewal, the portion of a given large claim that your plan retains moved with it. A triangle built on the prior retention level over-states or under-states depending on direction. Ask the actuary how large claims were capped in the triangle and whether the capping reflects your current retention.
3. Plan design change mid-year. A mid-year change to the deductible, the network, the prior authorization rules, or the specialty drug formulary will change both the frequency and the severity pattern of claims. The triangle will not see the change for several months because the lag means recent service dates are not yet fully paid. During that window, completion factors from history are not exactly describing the new plan. This is the kind of error that gets found at the next review and labeled as prior-period development. You want your actuary to flag known plan changes in writing before the estimate is finalized.
4. Enrollment growth or shrinkage. The raw triangle is a dollar tool. If your covered population grew materially in the incurred months near the edge of the triangle, the dollar figures understate ultimate claims for those months unless the actuary normalizes by member months. Normalizing on a per-member per-month basis is how a good health reserving workflow handles enrollment changes. Ask whether the estimate is PMPM-based or dollar-based, and if dollar-based, whether there is an exposure adjustment.
5. A frequency or severity shock. The clearest recent example is COVID. Claim volumes dropped for several months and then rebounded with both deferred-care and new-risk components. Any model trained on pre-shock data will misestimate the post-shock period, and any model trained only on post-shock data will misestimate normal conditions. The broader lesson is that a single external event can make the last three to six months of your triangle temporarily unreliable. A good actuary will tell you which months were adjusted or excluded and why. Todd Sherman’s CAS 2020 CLRS healthcare reserving presentation walked through the COVID-era adjustments in detail and is a good reference for exactly this situation.
Underneath all five is a single principle: a lag-based reserve estimate is a statement about stability. When the environment changes, the estimate needs to be adjusted, not just re-run.
What you should expect to receive from your actuary
A health reserve study that is worth the engagement fee will give you, at a minimum, five things in writing.
A point estimate. This is the single number you will book. The actuarial standard that governs it is ASOP 43, which per Friedland (p. 10) defines the actuarial central estimate as “an estimate that represents an expected value over the range of reasonably possible outcomes”. In practice it is the actuary’s best single answer given the methods and data.
A reasonable range. A range is not the same thing as a confidence interval. It is the window of estimates that a different qualified actuary could also defend as reasonable on the same data. For most health plans the range is fairly narrow because the short tail keeps method-to-method disagreement contained, but it should still be disclosed. The point vs range explainer walks through the distinction for a non-actuarial reader.
Documented assumptions. Specifically: which months of data were excluded or adjusted, how large claims were treated, which completion factor averaging window was used, what trend rate was assumed, and what PMPM or exposure base was used for the sanity check. A report that does not name its assumptions is asking you to trust defaults you cannot audit.
The lag triangle itself. Paid claims by incurred month and months of run-out, for the full study period. The triangle is the evidentiary basis for the estimate. If the actuary will not share it, something is off.
A PMPM reasonableness check. The completion-factor answer should be cross-checked against PMPM trend times member months, and the delta should be explained. If the two numbers diverge materially and the report does not acknowledge it, the estimate is not ready to book.
One note on formal actuarial opinions. If your plan is required to file a Statement of Actuarial Opinion for regulatory purposes, that is governed by ASOP 36 rather than ASOP 43. The Friedland reference does not cover ASOP 36 in detail, so ask your actuary whether a formal opinion is in scope and under which standard it is being issued.
What it should cost and how often
Fees for a health IBNR study vary more than you would expect. A limited quarterly update for an established client runs in the low four figures. A full annual study for a mid-sized self-funded plan with multiple layers, stop-loss interactions, and a formal opinion sits in the five figures. If you are being quoted numbers materially above or below that range for standard scope, ask what is driving it.
On cadence, most self-funded plans get a fresh estimate at least quarterly for accrual purposes, a fuller study once a year for the audit, and an ad-hoc study whenever the plan changes structure. A TPA change, a stop-loss retention change, a merger, or a material benefit redesign all qualify. If you are only getting one estimate a year and your plan is material to the company’s financials, you are effectively booking a stale number for eleven months out of twelve.
The point of a quarterly refresh is not to rebuild the triangle from scratch. It is to reapply the existing completion factors to a fresh paid-claims diagonal, flag any drift in the lag pattern, and produce a revised point estimate. A week of the actuary’s time, not a month.
What to do next
If you have read this far and want to act on it, three concrete moves will get you most of the value.
Ask your TPA for a paid-and-incurred lag triangle by service month for the last thirty-six months. This is the core data artifact for any health reserve estimate. Your TPA already produces it, or produces something that can be transformed into it. You do not need the actuary to get the triangle. You need it first so the actuary has something real to analyze. If the TPA says they cannot produce it, that is itself diagnostic information about how well instrumented your plan is.
Ask your stop-loss carrier what IBNR assumption is baked into your renewal. The carrier is estimating your plan’s incurred-but-unpaid claims for their own pricing, and their number is one useful external benchmark. It will not match the actuary’s estimate exactly because the two are built for different purposes, but a large disagreement is worth understanding before you sign the renewal.
Get an independent actuarial estimate at least once a year if you have material reserves on the balance sheet. “Material” is a judgment call, but a useful rule of thumb: if the reserve number is large enough that a ten percent change in it would be a visible event for the finance committee, it is material. Independent here means an actuary whose primary client is you, not your TPA or your carrier. The independent reviews note has more on why the distinction matters for self-insured programs generally.
None of these steps require you to become an actuary. They do require you to ask questions your actuary should be happy to answer. When the answers come back clear and specific, the number you are being asked to book is probably trustworthy. When they come back vague or defensive, that is itself the result you went looking for.