Blog
November 4, 2026
by
Andrej Evtimov
The arithmetic of a full desk, and the three things it misses first.
When caseloads climb, file quality does not degrade evenly. One specific kind of work stops first, and it is the kind that finds prior injuries, treatment gaps and duplicate bills. This is what that costs and how to measure it on your own book.
Start with the arithmetic, not the anecdote
Take an adjuster carrying 150 open bodily injury files. Assume a 37.5-hour week. Assume 60% of that time reaches the files at all. The rest goes to calls, diary, correspondence, internal meetings and system work.
That leaves about 22.5 hours a week across 150 files. Nine minutes per file per week. Around 39 minutes a month.
This is an illustration, not an industry benchmark. Run it with your own numbers and the shape holds. The total is fixed, so every file added subtracts minutes from every other file.
Nine minutes is enough to read new correspondence and update a diary note. It is not enough to read a 300-page medical record set against the prior claim history. So that reading does not happen. It also does not announce itself when it stops.
That is the problem worth your attention. Caseload does not degrade file quality evenly. It removes one specific kind of work first, and that work is where the largest dollars sit.
Where the hours actually go
Ask your own teams to log a normal week against four buckets:
- Reactive contact. Calls, emails, attorney correspondence, claimant contact.
- Administrative. Diary, notes, system entry, reserve worksheets, internal reporting.
- Single-document review. Reading one bill, one report, one demand as it arrives.
- Cross-document review. Reading several documents against each other to find what one alone cannot show.
Federal oversight of another caseload-driven profession suggests the administrative share can be very large. The Government Accountability Office found that child welfare caseworkers spent "between 50 and 80 percent of their time completing paperwork" (GAO-03-357, 31 March 2003). The report adds that this limited their time to assist children and families.
That is a different profession and a dated report. Treat it as a mechanism, not a number to apply to your desk. The limits of that comparison are set out further below.
What "review time per file" really means
Review time is not distributed evenly across a caseload. Most files need little. A minority consume most of the available hours.
That is healthy, up to a point. Past that point, the severe files absorb the hours the mid-tier files needed. Those mid-tier files then get processed rather than examined.
A mid-tier file that is never examined is how exposure moves by an order of magnitude. Say a claim reserved at $40,000 runs to $400,000 because a prior injury surfaced late. Those figures are illustrative, but the mechanism is routine.
Why caseloads climb even as headcount falls
The counterintuitive part is that this is happening while the occupation shrinks.
The BLS picture
The Bureau of Labor Statistics puts employment of claims adjusters, examiners and investigators at 376,100 in 2025. It projects a 6% decline from 2025 to 2035. The stated reason is automation: "Technology is expected to automate some of the tasks that these workers currently perform" (BLS Occupational Outlook Handbook).
Now read the second number. BLS still projects about 21,600 openings a year across the same decade. Those openings come from workers transferring out or leaving the labor force, including retirement.
Put the two together. The occupation is getting smaller, and it is still replacing roughly 21,600 people a year. The median annual wage was $78,000 in May 2025. That sets the cost of every seat you try to add.
The experience gap behind the headcount
Headcount is the wrong unit anyway. Two adjusters at 75 files each are not equivalent to one adjuster at 150. Two new hires are not equivalent to two ten-year veterans. Both comparisons are illustrations, but neither is controversial on a claims floor.
When an experienced reviewer leaves, three things happen at once. Their files redistribute to people already at capacity. The replacement takes months to reach useful speed. And the work the departing reviewer had done in their head, rather than in the file, leaves with them.
The GAO report names that third effect directly. Turnover "disrupts the continuity of services," particularly when new caseworkers must reevaluate assessments "due to poor or insufficient information in case files" (GAO-03-357). That is a 2003 child welfare finding, not a claims study, but the mechanism transfers.
Substitute "medical chronology" for "safety assessment" and you have the claims version. The successor either redoes the analysis, or proceeds without it. Under caseload pressure, the second is more likely.
What gets missed first
Findings do not fail at random. They fail in an order set by how many documents you must hold in view at once to see them.
The table below is a working model built from document complexity. It is not a measured research finding, and it is not drawn from a published study.
- A bill exceeds the fee schedule. Documents needed: 1. Fails at caseload: Late. Single-document checks survive.
- A demand asserts a new diagnosis. Documents needed: 1. Fails at caseload: Late.
- Treatment stops for 5 months, then resumes. Documents needed: Many, in date order. Fails at caseload: Early.
- A duplicate bill under a different provider name. Documents needed: Many, across payment history. Fails at caseload: Early.
- A prior injury to the same body part. Documents needed: Many, across claims and years. Fails at caseload: First.
Single-document checks survive caseload pressure because they fit in the nine minutes. Cross-document findings do not. They need an uninterrupted read of the whole file.
Prior injuries and prior claims
Prior injuries fail first because they are the most expensive to find. The record almost never labels them. There is no line that says "pre-existing."
Instead there is a passing reference in an intake form. There is an old imaging study listed in a history. There is a medication that implies a condition nobody has mentioned.
Finding it means reading the full history, then connecting it to a claim from years earlier. That earlier claim may sit with a different carrier. At nine minutes a week, that work is not deferred. It is skipped.
The consequence is largest where apportionment applies. Apportionment is the rule that splits an award between the current injury and a prior condition. It decides a large share of the exposure in workers' compensation claims handling.
Gaps and inconsistencies in treatment
A treatment gap is only visible on a timeline. Individual notes look normal. The absence between two of them is the finding, and an absence is invisible unless the records sit in date order.
The same applies to inconsistency. A functional capacity evaluation that contradicts a treating note three months earlier is a strong fact. It is also two documents that arrived four months apart, in different batches, possibly filed under different headings.
Duplicate and unbundled billing
Duplicate billing survives because it hides in volume. The same service appears twice under two provider entities, or with a slightly different code, or split across two dates.
Nothing about either bill looks wrong alone. You find it by comparing the bill against everything already paid. That is a cross-document task, so it follows the same curve as the others.
The pattern worth remembering
The order is not about difficulty. It is about how much of the file has to be in your head at the same time.
That is why caseload metrics based on file counts mislead. They measure how many files an adjuster holds. They do not measure whether anyone has read any of them end to end.
What the research actually supports
Be careful here, because this is where claims content usually overreaches.
What the adjacent evidence shows
There is no large published study measuring bodily injury adjuster caseload against claim outcomes. Anyone who tells you otherwise should be asked for the citation.
What does exist is evidence from other caseload-driven professions. The GAO analysis of federal child welfare reviews is one example. It found that "large caseloads and worker turnover delay the timeliness of investigations" (GAO-03-357). The same pressures limited how often workers visited children.
The same report is useful on caseload standards. Recommended standards were 12 to 15 children per caseworker, or a maximum of 18. Actual caseloads ranged from 10 to 110, averaging about 24 to 31. Some workers carried double the recommended number.
Two things transfer to claims. First, the degradation showed up as timeliness and contact frequency, not as visible errors. Second, the gap between the recommended caseload and the real one was a multiple, not a margin.
What it does not show
It does not give you a correct number of claims per adjuster. It does not measure dollars. It comes from a different profession, and the report was published in 2003.
Use it for the shape of the problem. Do not quote it as a claims statistic. Do not let a vendor quote it to you as one either.
What it costs, and how to measure it
Here is the uncomfortable part of the leakage conversation.
Why the leakage percentages you have seen are unreliable
Claims leakage figures circulate widely. You will see 2%, 5%, sometimes more.
Research for this article traced several of the commonly cited figures. Each led to a secondary article citing another secondary article. None carried a published methodology or a definition of what had been counted.
Some of those numbers may well be right. You cannot tell, which makes them useless for a business case that has to survive your CFO.
What is measurable is the direction of claim costs. WCRI reports a 6% average rise in total workers' compensation claims costs between 2022 and 2025. That figure is for the median study state (CompScope Benchmarks, 2026 edition). The study covers 18 states, and the growth spans medical payments, indemnity benefits and benefit delivery expenses.
Costs per claim are rising. Review minutes per claim are falling. That is the case, and it does not need an invented percentage.
A measurement you can run in one quarter
Stop looking for an industry number and produce your own. Pull a sample of closed files, stratified by severity, from a period when caseloads were high.
Re-review each one properly, with no time limit, and record four things:
- Findings present in the file but not acted on. The prior injury was in the records and nobody used it.
- Findings that required cross-document reading. Which of the above needed more than one document to see.
- The dollar value of each. What the finding would have changed: apportionment, a denied line item, a lower reserve, an earlier settlement.
- Caseload at the time. The handling adjuster's open file count in that period.
Now you have your own leakage figure and your own definition. You also have a correlation with caseload drawn from your own book. That survives scrutiny. A borrowed percentage does not.
Run the same exercise on a sample of auto liability claims involving injury. The prior-injury question and the treatment-gap question carry most of the disputed value there.
What a VP of Claims can actually change
Headcount is the obvious lever and usually the slowest one. Three others move faster.
Triage by complexity, not by count
Equalizing file counts across a team optimizes the wrong variable. A desk of 150 routine files and a desk of 150 litigated files are not comparable assignments.
Weight assignments by the review work each file needs. Record volume, treatment duration, prior claim history, attorney involvement and body parts claimed all predict required hours better than the file count does.
Protect the cross-document read
If cross-document findings fail first, protect that task rather than hoping it survives.
Three options work. Schedule a full-file read at defined triggers. Assign a second reviewer whose only job is the timeline and prior history. Or require a documented end-to-end read before any reserve above a threshold is set.
The point is to make that read a named step with a time budget. It should not be the thing that happens if the week allows.
Measure review depth, not just closure
Most claims dashboards measure throughput. Files closed, cycle time, open inventory, diary compliance. None of those show whether anyone read the file.
Add measures that do. Was a chronology built, and on what date relative to intake. Was prior claim history checked, and against which sources. Were bills reconciled against payment history. These are auditable, and they are the ones that fail silently as caseloads climb.
Where tooling helps, and where it does not
The cross-document read is the expensive part. It is also the most mechanical.
Putting hundreds of pages in date order is volume work. So is matching providers across bills, and surfacing every mention of a body part. Volume, not judgment, is the constraint.
That is the part worth automating. The amaise platform organizes the claim file and applies structured medical intelligence for injury claims. The adjuster then starts from an organized record rather than from page one. The judgment on what each finding means stays with the adjuster.
Be clear about what that does and does not solve. It shortens the read. It does not decide the claim, and it does not make a 150-file desk a good idea. What it changes is which work the nine minutes buys.
Reading a file end to end is still the job. The question for a VP of Claims is whether the desk you have designed leaves anyone the time to do it.
Key takeaways
- Caseload pressure removes cross-document review first. Single-document checks, such as a bill against a fee schedule, survive because they fit in the minutes available.
- Prior injuries fail first, treatment gaps and duplicate bills next. All three need several documents read against each other, which is exactly the work that stops.
- Headcount is the wrong unit. BLS projects a 6% decline in adjusters through 2035 alongside about 21,600 openings a year, so experience leaves faster than the seat count suggests.
- Do not quote industry leakage percentages. Trace them and most lead to a secondary article with no methodology. Measure your own on a stratified sample of closed files.
- Three fast levers: weight assignments by file complexity rather than count, make the end-to-end read a named step with a time budget, and measure review depth rather than closure alone.
