Blog
September 27, 2026
by
Andrej Evtimov
What setting the reserve before anyone reads the file does to your development triangles.
On most bodily injury claims the first reserve is set days after first notice, on a file nobody has read. This looks at what that gap does to reserve development, how leakage and adverse development differ, and how a finance team can measure whether earlier file review actually moved the triangles.
The reserve is usually set before anyone has read the file
On most bodily injury claims, the first reserve goes up within days of first notice. At that point the file contains a loss report, some coverage detail, and whatever the claimant said on the phone.
Medical records arrive weeks later. The demand arrives months later. By then the reserve has been carried on the books through at least one quarter, and often through a year-end.
Everything finance later calls reserve development starts here. Not in the actuarial method, and not in the tail factor. In the gap between when the number is set and when the file is actually read.
So the question for a finance team is a narrow one. On your own triangles, how much of last year's development came from claims that were reserved before anyone had read them?
What reserve development measures, and what it does not
Reserve development is a backward-looking measure. It compares what a cohort of claims was expected to cost with what it has since proved to cost.
The reporting frame is Schedule P of the NAIC annual statement. It carries ten years of premiums earned, losses unpaid, and claims reported and outstanding, by line of business. The triangles, which array each accident year against successive valuation dates, are where a reserving problem becomes visible outside claims.
Development is a statement about the estimate, not about the claim
A claim that develops adversely did not necessarily get worse. It may have been understated from the start, and the record simply caught up with reality.
That distinction matters for where you look for a fix. If claims genuinely deteriorate, the answer is claim strategy. If estimates were thin because nobody had read the file, the answer is file review timing.
Either way the unit of analysis is the cohort: a group of claims from the same accident year, followed across valuations.
Most carriers have some of both. Few measure which is which.
The first number anchors the ones after it
Reserve changes tend to move in steps, at file review points, rather than continuously. An opening reserve that is too low is rarely corrected in one movement.
It is corrected in a sequence of increases, each triggered by a document somebody finally read. Every one of those steps lands in a different accident-year valuation, and the pattern of steps is what the triangle records.
Leakage and adverse development are different failures
These two terms get used interchangeably in management meetings. They are not the same thing, and they have different owners.
- What it is. Leakage: Paid more than the claim required. Adverse development: Reserved less than the claim proved to cost.
- Where it shows. Leakage: Paid loss, claim audit findings. Adverse development: Schedule P triangles, reserve movements.
- Typical cause. Leakage: Missed offsets, unchallenged specials, late liability defence. Adverse development: Late information, thin opening estimate, missed severity indicators.
- Who raises it. Leakage: Claims quality assurance. Adverse development: Actuarial and finance.
- Fixable by day-1 review. Leakage: Partly. Adverse development: Directly.
A file can leak without developing, and develop without leaking. A reserve set correctly on day 1 and then overpaid at settlement is a leakage problem with a clean triangle.
Early file review addresses the right-hand column directly. It affects the left-hand column only where the missed information would also have changed the payment.
What early file review actually changes
It does not make the estimate more sophisticated. It moves the facts earlier.
Three things a day-1 read can establish
A first-pass review of everything in the file, on day 1 rather than month 4, can usually settle three questions.
- Severity indicators. Prior treatment for the same body part, surgical recommendations, comorbidities, and any hospital admission. These move a reserve by an order of magnitude, and they sit in documents nobody opens early.
- Litigation signals. Represented status, a letter of protection, or a records request pattern that reads like a build-up.
- Coverage and exposure. Which policies respond, whether excess layers are in scope, and whether the primary limit is realistically in play.
None of these require an evaluation. They require someone, or something, to have read the documents.
It compresses the interval, not the judgement
The adjuster still sets the reserve. What changes is the information available when they set it.
This is the practical case for AI-assisted bodily injury claim file review. The sorting, extraction and timeline work happens in the first days rather than the first quarter. The evaluation, and the responsibility for it, do not move.
Cycle time is the metric finance can already see
Most finance teams cannot observe file review quality directly. They can observe time.
Two intervals predict development better than most quality scores.
The first is the interval from first notice to the first complete file read. The second is the interval from a document arriving to the reserve being revisited.
Where a reserve threshold triggers escalation, the same principle applies to notification. Automated large loss notice drafting closes the second interval. It pulls the underlying facts together when the threshold is crossed, and the adjuster reviews the draft before it goes out.
The intervals matter most in long-tail lines. In workers' compensation claim handling, a thin opening estimate can sit on the books across several valuations before anything contradicts it.
Track both intervals by accident year. If they shorten and one-year development improves in the same cohorts, you have an argument. If they shorten and development does not move, the problem was never timing.
What the actuary needs from claims, in their own terms
Reserving actuaries are explicit about this dependency, and the standards are worth reading if you have not.
Actuarial Standard of Practice No. 43, adopted by the Actuarial Standards Board in June 2007, governs property and casualty unpaid claim estimates. It defines an unpaid claim estimate as "the actuary's estimate of the obligation for future payment resulting from claims due to past events."
The standard tells the actuary to ask about claim handling
Section 3.6.7 directs actuaries to consider whether there have been "significant changes in conditions, particularly with regard to claims, losses, or exposures." Where the actuary relies on representations from others, the standard permits reliance "unless, in the actuary's professional judgment, they appear to be unreasonable."
Read that from the claims side. The actuary is asking a question, and the claims organisation is the one answering it.
ASOP No. 36, adopted in March 2024 and effective from 1 October 2024, governs the statement of actuarial opinion itself. Under that standard, the opinion turns on whether the booked figure falls "within a range of estimates that the actuary considers reasonable."
The range is built from history. History is built from files.
Changing claim handling distorts the triangle before it improves it
This is the part that gets skipped in business cases, and it is the objection a reserving actuary will raise first.
Development factors are derived from historical patterns. If you change when information enters the file, you change the pattern. Reserves that previously stepped up in months 9 and 18 may now step up in month 1.
For a period, the historical factors overstate future development, because the development they were built on has already happened earlier. An actuary who is not told about the change will read the result as favourable development and may not know why.
Tell them before you deploy, not after. Date the change, identify the cohorts it affects, and expect the first two valuations to be harder to interpret rather than easier.
How to tell whether it worked
Four measures, in the order they become available.
- Process, available immediately. The share of claims whose opening reserve followed a complete file read, plus median days from first notice to that read.
- Movement, available within two quarters. The number and size of reserve steps per claim, compared with the prior cohort. Fewer, earlier steps is the expected signature.
- Development, available at the first annual valuation. One-year development for the affected accident years, isolated from cohorts the change did not touch.
- Leakage, available from audit. Whether claim audit findings on missed offsets and unchallenged specials fall in the same cohorts.
Compare cohorts, not calendar periods. A calendar-year comparison mixes the change with everything else that happened that year. It will not survive scrutiny from anyone who reserves for a living.
What early review does not fix
It does not fix a reserving philosophy that sets opening reserves at a formula value regardless of file content. That is a policy decision, and reading the file earlier will not override it.
It does not fix inflation in medical costs or in jury awards. Those move severity for everyone, and no amount of early reading changes the underlying claim.
It does not fix a claims organisation with no capacity to act on what the review finds. A severity indicator surfaced on day 1 and actioned in month 6 has produced a better-documented file with the same reserve pattern.
What it removes is one specific, measurable cause of adverse development. That cause is the months a claim spends reserved by someone who has not read it.
That is a narrower claim than most business cases make. It is also one you can test on your own triangles.
Key takeaways
- Adverse development often starts on day 1, in the gap between when the reserve is set and when the file is first read. A claim that develops adversely did not necessarily get worse; the estimate may have been thin from the start.
- Leakage and adverse development are different failures with different owners. Leakage is paying more than the claim required and shows in paid loss and audit findings. Adverse development is reserving less than the claim proved to cost and shows in the Schedule P triangles.
- Early file review does not make the estimate more sophisticated. It moves severity indicators, litigation signals and coverage facts forward, so the adjuster sets the opening reserve with them rather than without them.
- Two intervals predict development better than most quality scores: first notice to first complete file read, and document arrival to reserve revisit. Both are visible to finance without auditing file quality.
- Warn the actuary before deploying. Changing when information enters the file breaks the historical development pattern, and for a period the old factors will overstate future development.
- Compare cohorts, not calendar periods. A calendar-year comparison mixes the change with everything else that happened that year.
