Loss Run Forensics: 7 Red Flags Underwriters Miss on a First Pass

A loss run is a carrier's snapshot of an account's claims at a moment in time. The operative word is snapshot. It is not a history, not an audit, and not a prediction. It is what the carrier's claims system said on the day someone ran the report, filtered through whatever that carrier chooses to show.

Underwriters know this in the abstract. In practice, with a broker waiting, the first pass tends to be a scan for total incurred, a claim count, and a look for anything large. That scan misses most of what matters. Here are the seven things we train people to look for.

1. The valuation date is not the date on the cover

Every loss run has a valuation date, the "as of" date for every dollar figure on it. It is often printed in small type in the header or footer, and it is frequently not the date the report was generated. A run produced in January may be valued as of the prior September because that is when the carrier last updated reserves for that account.

Why it matters: incurred values change as claims develop. A run valued eighteen months after policy inception will look very different from the same policy valued at thirty months, and a run valued at six months is close to worthless for projecting the ultimate cost of that year. The rating bureaus know this, which is why they value claims on a fixed schedule. The WCIRB describes the first unit statistical valuation at eighteen months after the policy inception month, with subsequent valuations every twelve months thereafter for claims that remain open, and typically only the first three reports used in experience rating. NCCI's unit statistical plan works on a similar cadence.

The practical check: find the valuation date on every run in the submission, write it next to the policy period it covers, and compute the age of each policy year at valuation. A policy year valued at nine months is immature and will develop. One valued at forty months is close to settled. If two runs for the same policy year carry different valuation dates, the later one governs, and the difference between them is a development signal in its own right.

2. Open reserves are a carrier's opinion, not a cost

A loss run typically shows paid, reserved (or outstanding), and incurred, where incurred is paid plus reserve. Many first-pass reviews treat incurred as the cost of the claim. It is not. The reserve portion is the claims adjuster's estimate of what remains to be paid, and that estimate reflects the carrier's reserving philosophy as much as the facts of the injury.

Some carriers reserve conservatively and release as claims close, so open claims look worse than they will end up. Others reserve thin and step up as claims develop, so open claims look better than they will end up. A single large open reserve can be a serious injury or a placeholder an adjuster set and forgot.

The practical check: separate paid from reserved on every open claim, and look at the ratio. A claim two years old with a small paid amount and a large reserve deserves a question. A claim with paid approaching the reserve is likely near its ultimate. For the account as a whole, the share of incurred that is still reserved tells you how much of the loss history is opinion rather than fact, and a loss pick has to carry a margin for that opinion being wrong.

3. Lagged reporting hides the most recent year

Claims are not reported on the day of injury. There is a lag between injury and report, another between report and the carrier opening a file, and another before the claim appears with a meaningful reserve. For the most recent policy year on a run, this lag means the claim count is incomplete and the incurred total is understated.

A first-pass reviewer looks at the current year, sees two small claims, and concludes the account is improving. Sometimes it is. More often the current year simply has not been reported yet.

The practical check: compare the most recent year's claim count at its current age to what prior years looked like at the same age, if the history allows. If prior years each had eight claims at thirty months and the current year has two at six months, the current year is on pace for something like eight. Where you lack the history, assume the current year is incomplete and load the projection for incurred-but-not-reported claims.

4. Claims that vanish between runs

When a submission includes two runs from the same carrier at different valuation dates, line them up. Most claims will appear on both with incurred values that moved up or down. Some claims will appear on the earlier run and be missing from the later one.

A claim can legitimately disappear if it was closed with no payment, denied, transferred to another policy, or fully recovered through subrogation. It can also disappear because the report filter excludes closed claims, because someone requested an open-claims report instead of a full history, or because the claim was moved to a different entity in a restructuring.

The practical check: reconcile claim numbers across runs. Any claim present on one and absent from the other needs an explanation. A pattern of claims that vanish is sometimes benign and sometimes the first sign that the loss history you are looking at is not the complete loss history for the operations you are being asked to underwrite.

5. Multi-carrier gaps and overlaps

Most accounts worth pursuing have been with more than one carrier over the experience period. A submission that includes three years from the current carrier and nothing from the prior carrier is missing a year or two of history, and the missing years are not randomly missing. An employer that changed carriers after a bad year often has a bad year sitting with the carrier that was not asked for a run.

The reverse problem is overlap. A policy that was cancelled mid-term and rewritten with a new carrier produces two runs covering parts of the same twelve months, and a reviewer who adds them together double-counts the overlap period's exposure while possibly missing claims that fell in the gap between cancellation and rewrite.

The practical check: build a timeline. List every carrier and policy period from the ACORD 130's prior coverage section and the dec pages, then map every loss run onto it. Every month in the experience period should be covered by exactly one run. Gaps are a request for more documents. Overlaps are a reconciliation task.

6. Policy-year aggregation errors

Loss runs are organized by policy, but underwriting is organized by policy year, and the two do not always line up. A carrier run may list claims by date of injury across a multi-year relationship without breaking them into policy periods. A run from a carrier with a non-standard policy term may cover fourteen months. A run for a short-rate cancelled policy may cover five.

When a reviewer sums claims into "years" by calendar year of injury rather than by policy period, the totals are wrong in ways that are hard to see. Claims from a single policy year get split across two calendar years; development patterns get smeared; the mod projection, which depends on assigning each claim to the correct policy year, drifts.

The practical check: assign every claim to a policy period by its date of injury and the policy's effective dates, not by calendar year, and total by policy period. If a run does not identify the policy period, get it from the dec page or the ACORD. This is tedious, which is why it is the most common error in spreadsheet loss summaries and why ModIQ does it as the first step after parsing.

7. The medical-only versus indemnity mix is shifting

A claim with no lost time is a medical-only claim. A claim with wage replacement is an indemnity claim. The ratio between them on a loss run says something about both the hazard of the operation and how the employer manages injuries. An account whose claims are mostly medical-only has injuries that are being treated and returned to work quickly, or has a return-to-work program that keeps people on the job.

When that mix shifts across the experience period, pay attention. A rising share of indemnity claims can mean the work got more hazardous, the employer stopped managing return-to-work, a new supervisor arrived, or the claims are being handled by a different adjuster with a different philosophy. A falling share can mean improvement or can mean the employer has started treating minor injuries off the books.

The mix also matters for the mod in a way that varies by state. In NCCI states that have adopted the Experience Rating Adjustment, medical-only claims are reduced by seventy percent before entering the experience rating calculation, which makes the indemnity share of the claim history disproportionately important to the mod. California has not adopted that adjustment; under the WCIRB plan, a claim's actual losses are the sum of indemnity and medical incurred with no medical-only reduction. The same claim mix produces different mod consequences depending on where the employer is.

The practical check: tag every claim as medical-only or indemnity, compute the share by policy year, and look at the trend. A step change in either direction is a question for the broker or the employer, and the answer usually tells you more about the account than the incurred total does.

If a reviewer does the seven checks above, the output is not a number. It is a set of facts about the data: the valuation age of every policy year, the share of incurred still in reserve, the completeness of the most recent year, any claims that disappeared, any gaps or overlaps in carrier coverage, a clean policy-year aggregation, and the trend in claim type. The loss pick comes after, and it is only as defensible as those facts.

This is the discipline ModIQ in HALO IQ is built around. Loss runs come in through InboxIQ with carrier, policy period, and valuation date preserved on every claim. ModIQ aggregates by policy year, lines up multiple runs and carriers against the coverage timeline, separates paid from reserved, and carries the result into the mod projection and loss pick with the upward loads that open reserves and immature years require. The red flags above are not an afterthought in that process; they are the process. A loss run is a snapshot, and the job is to figure out what the camera was pointed away from.