Actuarial assumption management: why every pick was made, and whether it still holds
An auditor's question answered in minutes, with the original support attached.
See one case, screen by screen ↓Keeping every reserving and pricing assumption — development and tail factors, trends, expected loss ratios, discount rates — together with the reason it was chosen, so anyone can see why it was picked and whether it still holds. Actuarial Assumption Management traces each pick to its memo paragraph, workbook cell and data cut, and answers “why” questions with every sentence cited.
How to choose AI solutions for insurance →
Updated
Tail factors, development factors, severity trends, expected loss ratios and discount rates are chosen with care, and the reason is written down — in a memo paragraph, a workbook cell, the triangles the actuary looked at and the reserve committee minutes. ASOP No. 41 asks that the actuarial report set out its methods, assumptions and data clearly enough for another actuary to make an objective appraisal. The trouble is finding it again. When the external auditor's actuarial specialists send a PBC request about a pick made three year-ends ago, the reserving director has to find the right year's memo on the shared drive, the workbook version that matches it, the cell, the data cut and the minutes, then write it up.
Often the actuary who made the pick has left. Often the pick has been held since — “consistent with the prior year” — with nobody re-running the original method on the new triangles until the auditor asks. And some picks were agreed in committee and never written into the memo at all. Those are the picks an auditor finds first.
Estimated split, in hours, for one PBC request on a held pick — by hand and with the solution.
Six agents trace every pick to its cell, paragraph and data cut and re-run it when data moves; the appointed actuary signs.
Thornmere Mutual's year-end 2025 selections are in force. Morrow Abbott LLP asks about the 1.12 general liability tail, held since year-end 2023, whose author left in June 2025. Here is what happens next, screen by screen, in the working solution.
Inés opens the assumption lineage: 18 key picks of 1,284 across general liability, commercial auto, workers' comp and multi-peril — 13 supported, 3 stale, 2 without support. PBC-17 from Morrow Abbott is open, due Friday 16 October, four days away.
“Provide support for the 1.12 tail factor (120 months to ultimate) on commercial general liability, occurrence, held since year-end 2023 — including why the company's own indication was not used, and whether it remains appropriate given 2026 data.”
She clicks the Tail 120–ult leaf. Cell F42 of the Tail tab opens beside §4.3 of the year-end 2023 memo, page 23 of 41: an inverse power fit of 1.094 weighted one-third, the industry tail of 1.138 weighted two-thirds, 1.123 rounded to 1.12. The lineage runs on to the data cut RX-23Q3-CGL, the Reserve Committee of 21 February 2024 and peer review MV-24-07. The pick was made by Teodor Maksimović, ACAS, who left in June 2025.
Memo §4.3: “the last three diagonals show case reserve strengthening on 2014–2016 construction defect claims that the fit does not yet reflect.”
The leaf is marked stale. The freshness watcher re-ran the workbook's own formula and weights on RX-26Q2-CGL, loaded 14 July: the company fit is now 1.131 and the 2025 industry tail 1.152, so the same method indicates 1.145 — 2.0% above the 1.123 the pick was made on. Its own cells sit beside the original ones.
“Re-run on RX-26Q2-CGL: 1.145 indicated · +$4.7M IBNR at 1.145 · the pick is unchanged”
The history shows the pick against the same method on each year's data. 1.09 at year-end 2019 on the company fit only; 1.10 in 2021, when the industry tail first got one-third weight; 1.12 in 2023, with two-thirds. Then held at year-end 2024 and 2025 — “no change; see YE2023 memo §4.3” and “consistent with the prior year” — while the indication rose to 1.134 and 1.139.
She presses “Draft the answer”. The support researcher searches 2,316 documents across seven year-ends, reads 31 passages from 11 of them, follows the lineage, reads how the pick was held and checks freshness. It writes 9 sentences, one claim each, under the auditor's three questions. The citation checker reads every sentence back: 9 of 9 supported, none removed, figures quoted rather than recalculated.
“Disclosed in the answer. The tail was held at year-end 2024 and 2025 without a re-run, and the June 2026 data moves the indication by +2.0%. The answer says so rather than leave the auditor to find it.”
Saved as the support note for PBC-17: the selection, why the company's indication was not used alone, review and governance, and whether it remains appropriate. Seven attachments carry the original pages and cells — the year-end 2023 memo, the Tail tab, the committee minutes, MV-24-07, the year-end 2024 and 2025 memo extracts, the prior audit answer and the June re-run. It waits for the appointed actuary.
Marguerite reads each sentence with its passage, then opens the sign-off: 9 sentences, each cited to its source, 7 attachments, and the disclosure that the tail was held without a re-run.
“The note discloses that the tail was held without a re-run and that the June 2026 data indicates 1.145 (+2.0%), with the re-selection on the 5 November committee agenda.”
She signs and sends. The note goes to Gideon Farrah at Morrow Abbott LLP with its 7 attachments on Monday 12 October at 10:42 — four days before the due date. The pick itself is unchanged: re-selecting it is a decision for the reserve committee.
Two picks have no support, and the auditor has asked about both. The commercial auto AY2025 expected loss ratio of 74.5% was agreed at the Reserve Committee on 19 February 2026, after the memo was issued — the memo still shows 72.0%. The memo behind the workers' comp discount rate of 3.25% refers to a “discount workbook” whose link points to a folder moved in 2025. The gap finder drafts from the minutes for the line's owner to confirm; it never closes a gap itself.
Seven requests for the year-end 2026 audit planning and the Q3 2026 interim: PBC-17 sent, PBC-18 and PBC-19 flagged as gaps, PBC-15 in progress, two answered, and the state insurance department's triennial exam request with 41 documents already matched.
The same tail factor, seen by the four people who carry it — the reserving director, the appointed actuary, the auditor and model validation.
Six agents trace every pick to its cell, paragraph and data cut and re-run it when data moves; the appointed actuary signs.
Reads every memo, workbook, set of minutes, audit answer and peer review note, and keeps the page, paragraph, tab and cell of each statement.
Links each pick to its workbook cell, memo paragraph, data cut, committee minutes and peer review, and matches the same pick across year-ends.
When a data cut is refreshed, re-runs each pick's own recorded method and weights on the new data, and marks the pick stale when the indication moves past the threshold.
Answers a “why” question from an auditor, regulator or colleague with one claim per sentence, each carrying the passage behind it.
Reads each drafted sentence back against its passage and removes anything the source does not say.
Sweeps every pick for a missing paragraph, a broken workbook link or an author who has left, and drafts a request to the person who can close it.
Reads and signs. The agents propose; a named person decides.
Any reserving or pricing actuary can ask why a pick was made, or what has moved since it was. Every answer is cited to the memo, workbook or minutes behind it.
Why is the general liability tail 1.12?
It was chosen at year-end 2023 by Teodor Maksimović: one-third the company fit of 1.094 and two-thirds the industry tail of 1.138, giving 1.123, rounded. The company fit alone was judged too low because of case strengthening on 2014–2016 construction defect claims.
Which picks are stale since the June data?
Three, at your 1.0% and 0.5-point thresholds: the general liability tail 1.12 (+2.0%, about $4.7M of IBNR), general liability severity trend 7.5% (+0.8 pts) and commercial auto 24–36 1.31 (+1.9%, about $2.1M). Each was re-run with its own method and weights; no pick was changed.
Draft the answer to PBC-17
Nine sentences under the auditor's three questions, each cited to the memo, the Tail tab, the minutes, the peer review, the prior audit answer and the June re-run. It discloses the +2.0% move rather than leave it to be found.
How has the auto 24–36 factor moved since 2019?
It rose a step at a time from 1.27 at year-end 2019 to 1.31 at year-end 2025, each step following the five-year average, the last one for attorney representation rising from 41% to 46%. The June 2026 triangles put the same average at 1.327.
12 screens from the working solution, on its sample data. Pick one to see it large.
When a pick is called stale, who signs what leaves, and where the support comes from — all settings.
| Setting | Default | Choose from |
|---|---|---|
| Stale when a development or tail factor's re-run moves by more than | 1.0%, relative | 0.5% to 3.0%, in steps of 0.5% |
| Stale when a trend or loss ratio's re-run moves by more than | 0.5 pts, absolute | 0.25 to 1.5 pts, in steps of 0.25 |
| Show the reserve effect on the leaf when it is at least | $1.0M | Your floor |
| Re-run when a data cut is refreshed | On | Triangles quarterly · industry benchmarks annually |
| Notes to auditors and regulators | Signed by the appointed actuary | Required |
| Internal answers | Signed by the line's reserving director | Required |
| Agents may change a pick | Never | Off · locked |
| Retention | 7 years | Report and workpapers seven years · questions and answers kept permanently |
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 · “typical” = published figures (Assured Research via Carrier Management, 20 Mar 2026: US other liability (occurrence) reserves $12.5B deficient at year-end 2025). ASOP No. 41 §3.2 (Actuarial Standards Board) is named as context. People, companies and firms named on this page are fictional — characters and sample data in the working solution.
Keeping every reserving and pricing assumption — development and tail factors, trends, expected loss ratios, discount rates — together with the reason it was chosen, so anyone can see why it was picked and whether it still holds. Actuarial Assumption Management traces each pick to its memo paragraph, workbook cell and data cut, and answers “why” questions with every sentence cited.
The source reader reads memos, workbooks, minutes, audit answers and peer review notes by paragraph and by cell. The lineage builder links each pick to its workbook cell, memo paragraph, data cut, committee minutes and peer review, and matches the same pick across year-ends. One click opens the cell and the paragraph together.
When a data cut is refreshed, the freshness watcher re-runs the pick's own method and weights from the original workbook on the new data. If the indication moves past your threshold — 1.0% for factors and 0.5 points for trends and loss ratios by default — the pick is marked stale with how far it moved and the reserve effect.
No. Agents never change a pick or propose a new one, and that setting is locked off. A re-selection is a person's decision, taken to the reserve committee.
It drafts the answer. Requests from the audit portal are matched to the picks they name; the support researcher drafts one claim per sentence with its passage, the citation checker removes anything the source does not say, and the note goes out with the original pages and cells attached once the appointed actuary signs it.
The gap finder flags it — a missing paragraph, a broken workbook link or an author who has left — and drafts a request from existing records, such as committee minutes, for the owner of the line to confirm. It never closes a gap itself.
The actuarial shared drive, the workbook store with every saved version, committee minutes, the audit request portal, model validation notes, data warehouse extracts of triangles and severity, and the actuarial team's email archive. Agents have read-only access to every source.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your memos, selection workbooks and a sample of past audit requests — and it goes live once every quality gate has passed. We will walk you through it on your own picks first.
Adverse development seen in weeks, not at the next quarterly review.
1-min demoPremium & Claims ReconciliationBreaks found and explained on day one of close, each tied to its source records.
1-min demoStatutory, GAAP & IFRS 17 ReconciliationOne explained walk from statutory to GAAP or LDTI to IFRS 17 results.
1-min demoWe'll trace a sample of your own year-end selections to their memos and workbooks.