Open-source Claude skills for property & casualty actuaries.
Skills • Quick Start • Example • Roadmap • Contributing
AI is reshaping how actuaries work — but the profession benefits most when practical tools are shared openly. This project provides ready-to-use Claude skills that encode standard actuarial methods, so any actuary can leverage them immediately. Think of it as a starting point: use these skills as they are, adapt them to your workflows, or contribute new ones. The goal is to build a shared foundation that helps the insurance industry move forward together.
Skills are packaged workflows that extend Claude's capabilities for domain-specific tasks. When you install an actuarial skill into a Claude Project, Claude automatically recognizes when to use it — upload a loss triangle and ask "check my reserves," and Claude runs a full analysis using standard actuarial methods without you writing a single line of code.
Skills work in Claude.ai Projects and Claude Code.
Upload a loss development triangle or raw claim-level transaction data (Excel or CSV) and get a multi-method reserve analysis with formatted exhibits — in seconds.
Methods included:
- Chain Ladder — volume-weighted and simple average age-to-age factors
- Bornhuetter-Ferguson — blends a priori ELR with development patterns
- Cape Cod (Stanard-Bühlmann) — derives ELR directly from the data
- Tail factor estimation via exponential decay extrapolation
Methods reference standard actuarial literature including Friedland's Estimating Unpaid Claims Using Basic Techniques and relevant ASOPs (No. 43, 23, 25, 36) for educational context.
Output: A 6-exhibit Excel workbook containing:
| Exhibit | Contents |
|---|---|
| 1 — Triangle | Your input data, formatted |
| 2 — ATA Factors | Individual and selected age-to-age factors (volume-weighted, simple, medial) |
| 3 — CL Ultimates | Chain ladder projected ultimates, CDFs, IBNR by accident period |
| 4 — BF & Cape Cod | Bornhuetter-Ferguson and Cape Cod results (when premium is provided) |
| 5 — Diagnostics | Calendar year test, outlier factor detection, tail factor sensitivity |
| 6 — Summary | Side-by-side method comparison with range analysis |
Diagnostics automatically flag:
- Outlier development factors (>2σ from the column mean)
- Calendar year diagonal inconsistencies
- Negative development (factors < 1.0)
- Sensitivity of total IBNR to ±10% and ±25% tail factor changes
We're building additional skills for common actuarial workflows. See the Roadmap below.
- Download
loss-reserve-analysis.skillfrom the Releases page - Open a Claude.ai Project → Project Settings → Skills
- Upload the
.skillfile - Upload any loss triangle and ask Claude to analyze it
- Clone this repo
- In your Claude.ai Project, add the contents of
loss-reserve-analysis/to your Project Knowledge - Claude will automatically reference the skill when you upload triangles
Upload a file like this to your Claude Project conversation:
| Accident Year | 12 | 24 | 36 | 48 | 60 |
|---|---|---|---|---|---|
| 2019 | 1,610 | 2,450 | 2,810 | 2,990 | 3,070 |
| 2020 | 1,390 | 2,080 | 2,390 | 2,540 | — |
| 2021 | 1,550 | 2,340 | 2,680 | — | — |
| 2022 | 1,820 | 2,750 | — | — | — |
| 2023 | 2,100 | — | — | — | — |
Then ask:
"Run a loss reserve analysis on this triangle. It's incurred losses in thousands, development in months."
Or, if you also have premium data:
"Check reserves on the attached triangle. Earned premiums are in the second sheet. Use BF and Cape Cod too."
Claude parses the triangle, runs all applicable methods, and returns a formatted Excel report with the exhibits described above plus a narrative summary of findings.
The skill handles common triangle layouts automatically:
- Standard triangle — rows are accident periods, columns are development periods (most common)
- Columnar / long format — three columns: accident period, development period, loss amount
- Transaction-level data — raw claim-level records with
claim_id,accident_date,evaluation_date, and loss columns — automatically aggregated into a development triangle - Excel or CSV —
.xlsx,.xls,.xlsm,.csv - Multiple sheets — specify which sheet contains the triangle; premium data can be on a separate sheet
Development periods can be in months or years. Accident periods can be annual or quarterly. If you have raw claims data rather than a pre-built triangle, the skill handles the aggregation for you.
This is a quick check, not a full reserve study. Specifically:
- Methods are standard textbook implementations — they don't incorporate claim-level information, operational context, or judgment that a credentialed actuary would apply
- Tail factor selection is mechanical (exponential decay). Production reserve analyses require actuarial judgment on tail selection
- Results should be cross-referenced with knowledge of changes in claims handling, coverage, legal environment, or reinsurance
- This does not constitute an actuarial opinion under ASOP No. 43 or a Statement of Actuarial Opinion per ASOP No. 36
- ASOPs are referenced throughout for educational context — they do not constitute compliance guidance
Use this as a starting point, a sanity check, or a way to quickly explore your data — not as a substitute for a signed actuarial analysis.
We plan to add skills for other common P&C actuarial workflows. Feedback welcome via Issues.
| Skill | Status | Description |
|---|---|---|
ratemaking-analysis |
Planned | Full P&C ratemaking workflow: loss trending, on-level premium, development to ultimate, expense loading, credibility weighting, and rate indications. References Werner & Modlin Basic Ratemaking, CAS Exam 5 syllabus, ASOPs 12/25/30. |
reinsurance-analysis |
Planned | Treaty and facultative reinsurance analysis: ceded loss development, net vs. gross comparisons, layer analysis, sliding scale commission evaluation, and reinsurance pricing. |
financial-statement-review |
Planned | Schedule P analysis, SAP/GAAP exhibit cross-checks, IRIS ratio calculations, reserve adequacy diagnostics, and annual statement reconciliation. Complements the Loss Reserve Analysis skill. |
If you have ideas for skills that would save you time, open an issue or reach out.
We welcome contributions from actuaries and developers. Some ways to help:
- Report bugs — if a triangle format doesn't parse correctly, open an issue with a sample (anonymized) file
- Suggest methods — want to see Mack's model, bootstrapping, or GLM-based reserving? Let us know
- Improve diagnostics — the more red flags we can automatically surface, the more useful the quick check becomes
- Add skills — if you've built a workflow that other actuaries would benefit from, submit a PR
Please see CONTRIBUTING.md for guidelines.
MIT License. See LICENSE for details.
Originally created by Kohei Kudo and Kalta.