Reserving data validation that stops a bad extract at the gate, before it becomes a loss triangle
Breaks and late bookings found before the actuary selects a single factor.
See one case, screen by screen ↓Checking the claim and payment data before it is used to build loss triangles and estimate unpaid claims: that counts roll forward, paid ties to the ledger, prior diagonals have not changed and claims have not moved between lines. Reserving Data Validation runs those checks the day the extract lands and holds any segment that fails, before it becomes a triangle.
How to choose AI solutions for insurance →
Updated
Every quarter the reserving team rolls its paid and incurred triangles forward with a new diagonal from a fresh claim extract. That extract moves under them: claims platforms are migrated in waves, coding tables change, payments are back-dated after a close, and a new module posts recoveries in the wrong field. The extract still loads, company totals still look close, and the triangle still builds.
The error shows up later, as a development factor that makes no sense or an indication that moved for no reason — after factors have been reviewed and exhibits built. Then someone has to find which claims caused it, ask claims data why, rebuild the triangles and redo the work. A coding change that moves claims between lines leaves totals unchanged; only a claim-by-claim comparison with last quarter shows it, and nobody has time to run one by hand.
Estimated split for one quarter’s extract, by hand and at the gate — modelled on the working solution’s sample quarter that took 21 days by hand.
Six specialist agents check every extract against the ledger and last quarter; the reserving actuary decides and releases.
Corrigan Bay Mutual’s Q3 2026 claim extract, valued 30 September, lands at 06:40. Here is what happens next, screen by screen, in the working solution.
486,215 claims and 3,118,440 transactions, compared with the Q2 extract locked on 14 July. Nine reserving segments, each a lane through eight checks. Six went through, three are held, and three breaks need Maren.
“Held · 1,240 claims moved line of business since last quarter”
The extract profiler checks the open-claim roll-forward, negative case reserves, reopened claims and dates. The ledger reconciler ties paid in the quarter to the ledger. The quarter-on-quarter comparer matches every claim to Q2, and the change-log reader looks for the change behind each failure. The gate log shows every step with its time.
Gate log: “471,902 matched · 14,313 new · 1,240 moved Commercial Auto → General Liability.”
3,714 matched claims changed something. Most are expected: 2,104 catastrophe codes added for the July hail, test claims purged, bodily injury codes merged — each tied to its change-log entry. One move holds the gate: 1,240 claims, $38.6M incurred, from Commercial Auto to General Liability.
“A change is expected only when a change-log entry says so; anything else is a break or within tolerance.”
B-301, the line-of-business move, waits for Maren. B-302 is Homeowners: 312 claims with a negative case reserve, −$1.9M, waiting for claims data. B-303 is 418 July payments dated 30 June that change the locked Q2 diagonal. The reopened workers’ compensation claims and 23 date errors are noted in the memo.
“Nothing reaches the triangles while a segment is held.”
Every moved claim is listed with its accident year, insured, coverage HNO-47, its Q2 line and its Q3 line, and the transaction that re-stamped it. Accident year 2024 carries the most: $12.4M. The side panel prices it: Commercial Auto IBNR falls from $130.2M to $99.8M, General Liability rises from $138.6M to $171.6M.
“Agents suggest; only the reserving actuary changes what reaches the triangles.”
The Commercial Auto incurred triangle, accident years 2017–2026, with the Q3 diagonal shaded. As extracted, the AY 2024 21→33 month factor reads 0.974 against 1.175 restated — negative development nobody would select. Both rows of age-to-age factors sit under the triangle.
“Factors are volume-weighted, all years — the selection stays with the actuary.”
The change-log reader quotes the ticket: coding table LOB_MAP moved from v12 to v13 on 14 August, HNO-47 now maps to Annual Statement line 17.1, and existing claims take the new code at their next transaction. Its impact assessment did not list actuarial. Maren chooses “Restate to Q2 coding for reserving” — adjustment A-01, listed for the appointed actuary.
“Existing claims take the new code at their next transaction; no back-conversion is planned.”
Commercial Auto and General Liability pass under A-01, and the gate now reads 8 through, 1 held. Homeowners stays held: Maren sends Dana Whitlock the drafted note with the 312 claims, due 13 October.
“Passes under A-01 · restated to Q2 coding”
Q3 paid by segment against ledger account 5100 for the September close, case reserves against the balance sheet, and year-end 2025 incurred against Schedule P, Part 1. The $2.3M Personal Auto difference is B-303; Maren re-dates those 418 payments to their booking date as A-02.
Total: extract $512.4M, ledger $514.7M — “Unexplained: $0.0M”.
The memo drafter writes it from the accepted adjustments only: data used, review performed, A-01 and A-02 with their claims, amounts, causes and approver, items noted, the open Homeowners item and the data limitations. Maren signs with the meaning “Prepared and reviewed”, and it goes to Robert Achterberg, the appointed actuary.
“In line with ASOP No. 23, I reviewed the data for reasonableness and consistency. I did not audit it.”
The same extract, seen by the five people who carry it — what their quarter looked like, and what it looks like now.
Six specialist agents check every extract against the ledger and last quarter; the reserving actuary decides and releases.
Reads each claim extract as it lands: counts, totals, field profiles, open-claim roll-forward, negative values and dates, per segment — and reports each check as passed, flagged or failed against the tolerances in Settings.
Matches every claim to the same claim in last quarter’s locked extract and lists silent changes — line, dates, state, cause, catastrophe code and claims that vanished.
Ties paid in the quarter and case reserves to the general ledger by segment, and year-end incurred to Schedule P, Part 1; explains each difference by a break or marks it unexplained.
Reads the system change and migration log and the coding tables, and ties each break to the ticket and table row that caused it, quoting the passage.
Builds paid and incurred triangles as extracted and as adjusted, and prices each break’s effect on age-to-age factors and an all-year chain-ladder indication, for comparison only.
Drafts the data memo for the reserve file from accepted adjustments and noted items, in ASOP No. 23 terms: data used, review performed, adjustments, items noted, open items and limitations.
Releases segments; appointed actuary reviews restatements. The agents propose; a named person decides.
The reserving team can ask about the quarter’s extract in plain words — or tell it what to change. Answers point to the screen they came from.
Why is Commercial Auto held at the gate?
Because 1,240 claims moved from Commercial Auto to General Liability since the Q2 extract. Change CHG-4471 put coding table v13 live on 14 Aug and maps hired and non-owned auto, HNO-47, to line 17.1. Only claims with a transaction after 14 Aug took the new code, so both lines would mix two codings. They carry $38.6M incurred.
What happens to the indication if we do not restate?
On the all-year factors, Commercial Auto’s indicated IBNR falls from $130.2M to $99.8M and General Liability’s rises from $138.6M to $171.6M — a $63M swing between two lines from coding alone. The AY 2024 21→33 month factor reads 0.974 instead of 1.175.
Hold any segment whose open count misses its roll-forward by more than 2%
Added. Open-claim counts that miss the roll-forward by more than 2% now hold the segment. Today that holds Commercial Auto (639 short, 9.3%) and General Liability (639 over, 10.6%) — both already held by B-301 — and nothing else. The rule is in Settings and on the audit trail.
Draft the note to Dana about the negative case reserves
Drafted. It lists the 312 Homeowners claims with negative case reserves, −$1.9M, points to the recoveries module change CHG-4502 of 22 Sep, and asks her to re-post them as recoveries by 13 Oct. Homeowners stays held until then. Review it and press Send.
13 screens from the working solution, on its sample data. Pick one to see it large.
The checks at the gate, their tolerances, who releases each segment and where the data comes from are settings — and a change shows what it would hold today.
| Setting | Default | Choose from |
|---|---|---|
| Paid ties to the ledger | $0.25M | Raise or lower the tolerance; on or off |
| Reopened in range | 1.5× the four-quarter average | Raise or lower the multiple; on or off |
| Line of business unchanged since Q2 | Hold above 25 claims | Raise or lower the claim count; on or off |
| Prior diagonals unchanged | Flag above 0.25% | Every earlier cell against the locked Q2 triangle |
| No negative case reserves | Any | Any claim below zero holds the segment |
| Dates valid | Any | No future dates; accident on or before report; closed after opened |
| Who releases each segment | Maren Holt, FCAS · all nine segments | Each person named with what they may do: release, review restatements, prepare only, fix the source, sign the ledger tie-out |
| New checks | Add a check | e.g. average case reserve per open claim against Q2 |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate. Context: Actuarial Standards Board, ASOP No. 23 Data Quality; NAIC Annual Statement Instructions, Statement of Actuarial Opinion (data reconciled to Schedule P, Part 1). People, companies and claims named on this page are fictional — characters and sample data in the working solution.
Checking the claim and payment data before it is used to build loss triangles and estimate unpaid claims: that counts roll forward, paid ties to the ledger, prior diagonals have not changed and claims have not moved between lines. Reserving Data Validation runs those checks the day the extract lands and holds any segment that fails, before it becomes a triangle.
The quarter-on-quarter comparer matches every claim in this quarter’s extract to the same claim in last quarter’s locked extract and compares line, coverage, accident date, state, cause and catastrophe code. In the sample quarter it matched 471,902 claims and found 1,240 that had moved from Commercial Auto to General Liability after a coding-table change.
No. The agents never edit source data, post journal entries or select factors. Locked prior-quarter extracts are read-only. A restatement for reserving is an adjustment the reserving actuary accepts, and it is listed in the data memo with its claims.
Yes. The ledger reconciler ties paid in the quarter and case reserves to the general ledger by segment, and year-end incurred by accident year to Schedule P, Part 1. Each difference is explained by a named break or marked unexplained.
Data used, the review performed, the adjustments made — only those a person accepted, each with its claims, amount, cause and approver — items noted but not adjusted, open items and data limitations. It is written in ASOP No. 23 terms and signed by the reserving actuary before it goes to the appointed actuary.
Yes. Each check can be switched on or off and its tolerance changed — the paid tie, the reopened ratio, how many claims may move segment — and the screen shows what the new tolerance would hold today. You can add checks of your own and choose who releases each segment.
Yes. A failed check holds the segment, and only the reserving actuary releases it or accepts an adjustment. Restatements above $10M are listed for the appointed actuary, and analysts can prepare but not release.
The Agentic Solution Engine builds and deploys it from your requirements — your reserving segments, coding tables, ledger accounts and two quarters of extracts — and it goes live once every quality gate has passed. We will walk you through it on your own extracts first.
Adverse development seen in weeks, not at the next quarterly review.
1-min demoActuarial Assumption ManagementAn auditor's question answered in minutes, with the original support attached.
1-min demoPremium & Claims ReconciliationBreaks found and explained on day one of close, each tied to its source records.
1-min demoWe’ll run Reserving Data Validation on two quarters of your own extracts.