P&C Loss Reserving AI — Chain-Ladder, BF & Development Factors

Why P&C Reserving Is the Right Problem for AI

Loss reserving is structured, repetitive, and numerically intensive — exactly where AI compresses the most time. A quarterly reserve review for a mid-size carrier involves building development triangles for 10–20 lines of business, selecting age-to-age factors, applying at least two methods per line, reconciling method differences, and drafting management commentary explaining the movements. The judgment calls are few. The mechanical work is enormous.

ClaudeFinanceLab's Insurance & Actuarial templates handle the structural work: triangle formatting, factor selection logic, BF a priori selection rationale, and reserve movement commentary. The actuary signs off on the assumptions; Claude builds the framework and writes the documentation.

Chain-Ladder Loss Development Triangle Analysis

The chain-ladder method requires selecting volume-weighted or simple average age-to-age factors from historical triangles, applying tail factors, and projecting ultimates. The judgment is in factor selection — identifying anomalous accident years, justifying departures from the volume-weighted average, and selecting the tail. Claude builds the triangle, calculates all standard factor selections, and flags years where the selected factor deviates significantly from the average.

• "Build a chain-ladder reserve analysis for commercial auto liability. Here is my paid loss development triangle (accident year rows, development age columns in months: 12, 24, 36, 48, 60, 72):
  AY2019: 1,240 / 2,890 / 3,820 / 4,210 / 4,390 / 4,450
  AY2020: 1,180 / 2,740 / 3,650 / 4,080 / 4,290
  AY2021: 1,320 / 3,100 / 4,020 / 4,480
  AY2022: 1,450 / 3,380 / 4,410
  AY2023: 1,510 / 3,520
  AY2024: 1,680
  Calculate: (1) age-to-age development factors for each period and accident year, (2) volume-weighted average factors, (3) simple average factors, (4) selected factors with brief rationale for any departure from volume-weighted average, (5) cumulative development factors (CDFs) to ultimate, (6) tail factor selection rationale at 72+ months, (7) projected ultimate losses by accident year, (8) IBNR by accident year and total. Flag any accident year where the observed factor deviates more than 10% from the volume-weighted average."
• "Analyze my workers' compensation loss development triangle for signs of a reserve development pattern shift. Prior to AY2020, development factors 12-to-24 averaged 2.35. Starting AY2020, the 12-to-24 factors are: AY2020: 2.58, AY2021: 2.71, AY2022: 2.63, AY2023: 2.69. (1) Is this shift statistically meaningful or within normal variation? Run a simple test comparing pre-2020 vs post-2020 factor distributions. (2) What are the most common causes of a sustained upward shift in early development in workers' comp? (3) If I apply the post-2020 average factor instead of the all-year average, what is the dollar impact on my AY2023 and AY2024 IBNR estimates, given AY2023 paid losses of $3.2M at 24 months and AY2024 paid losses of $1.9M at 12 months?"

Bornhuetter-Ferguson Method and A Priori Selection

The BF method blends the chain-ladder development approach with an a priori expected loss ratio, giving more weight to the a priori for immature accident years where actual experience is sparse. The defensibility of a BF reserve depends almost entirely on the a priori selection rationale. Claude helps structure the selection: comparing pricing loss ratios, plan loss ratios, and industry benchmarks, and documenting why each source receives what weight.

• "Apply the Bornhuetter-Ferguson method to my general liability book. Inputs: earned premium by accident year: AY2022 $8.4M, AY2023 $9.1M, AY2024 $10.2M. Pricing loss ratio: 62%. Plan loss ratio: 65%. Industry benchmark (ISO): 68%. Chain-ladder CDFs: AY2022: 1.08, AY2023: 1.22, AY2024: 1.65. Paid losses to date: AY2022 $5.12M, AY2023 $4.87M, AY2024 $3.31M. (1) Calculate expected unreported losses using BF for each accident year. (2) Calculate BF ultimate = paid + expected unreported. (3) Compare BF ultimates to chain-ladder ultimates. (4) For AY2024 where development is most immature, explain why BF is generally preferred over pure chain-ladder and how much weight I should give to the a priori. (5) Draft a reserve memorandum paragraph explaining the BF selection rationale and a priori basis."
• "I need to select an a priori loss ratio for a new line of business (cyber liability) entering its third policy year. I have: (1) our own 2 years of experience showing loss ratios of 58% and 71% on small premium volumes ($1.2M and $2.8M earned); (2) industry data from Verisk showing average cyber loss ratios of 65-72% over 2022-2024; (3) our pricing actuarial team's filed loss ratio of 63%. Help me: (a) weight these three sources for BF a priori selection, explaining the logic, (b) calculate a selected a priori and document the selection in a format suitable for reserve committee review, (c) perform a sensitivity test showing what the BF IBNR would be at a priori LRs of 60%, 65%, 70%, and 75%, given earned premium of $4.5M and a CDF of 1.85."

Reserve Adequacy Testing and Reconciliation

Reserve adequacy testing compares the selected reserve to method indications, prior year development, and external benchmarks. When multiple methods diverge, the actuary must explain the divergence and justify the selection. Claude structures the reconciliation table, identifies the source of divergence, and drafts the explanation — the three tasks that consume the most time in a reserve review.

• "Perform a reserve adequacy test and method reconciliation for my commercial property book. Method indications: (1) Chain-ladder paid: $14.2M IBNR, (2) Chain-ladder incurred: $11.8M IBNR, (3) BF paid: $13.1M IBNR, (4) BF incurred: $12.4M IBNR. Selected reserve: $12.9M IBNR. Prior year-end selected reserve: $11.4M. Paid loss development since prior year-end: $2.1M on prior accident years. (1) Calculate the prior year reserve development (favorable or adverse). (2) Build a method reconciliation table showing all four indications, selected, and deviation from selected. (3) The paid chain-ladder is highest because AY2021 had atypical large-loss activity — draft two sentences explaining why this year is excluded from the selected paid LDF. (4) Is the $12.9M selected reserve within a reasonable range of the indications? What would constitute a material departure?"

Tail Factor Selection and Long-Tail Lines

Long-tail lines — medical malpractice, workers' compensation, general liability — require tail factors extending development beyond the observable triangle. Tail selection is the single assumption with the most leverage on ultimate loss estimates for mature accident years. It's also the hardest to document defensibly. Claude helps by benchmarking against industry tail factors, fitting curves to observed development, and producing comparison tables across methods.

• "Select and document a tail factor for medical malpractice paid losses beyond 120 months. Observed development: AY2014 (120 months): $42.1M. AY2013 (120 months): $39.8M. AY2012 (120 months): $38.2M. The 108-to-120 month factors for these years are 1.009, 1.011, 1.008 respectively. Industry data: A.M. Best medical malpractice development tables suggest 120-to-ultimate factors of 1.015-1.025 for hospital professional liability. (1) Fit an exponential decay curve to the observed late-development factors to extrapolate beyond 120 months. (2) Compare the curve-fitted tail to the industry benchmark range. (3) Select a tail factor and justify it. (4) Calculate the dollar impact on total IBNR if the tail is 1.010 vs 1.020 vs 1.030, given $280M of case reserves at 120+ months."

Reserve Movement Commentary

Reserve committees and external auditors require written explanations of quarter-over-quarter and year-over-year IBNR movements. These narratives are mechanical — the same structure, different numbers — but writing them under time pressure produces vague, incomplete explanations. Claude generates the full reserve movement narrative from the numbers, covering prior year development, current accident year additions, and method or assumption changes.

• "Write a reserve movement commentary for Q3 2026. Total IBNR changed from $48.2M (Q2 2026) to $51.6M (Q3 2026), an increase of $3.4M. Components: (1) Prior accident year development: +$1.1M adverse (AY2022 commercial auto showed higher-than-expected severity); (2) Current accident year (AY2026) additions: +$2.8M for Q3 earned premium; (3) Method assumption change: -$0.5M favorable (updated industry tail factors reduced long-tail IBNR). Draft: (a) a one-paragraph reserve committee narrative, (b) a three-bullet executive summary for the CFO, (c) one sentence suitable for the MD&A section of the quarterly filing. For each, maintain appropriate technical precision while being accessible to non-actuaries at the executive level."

Where to Start

The Insurance & Actuarial category includes tools for the full actuarial reserving workflow: loss triangle construction, IBNR calculation, BF method, tail factor selection, and reserve adequacy documentation. Related reading: IBNR Reserving — Chain-Ladder & Bornhuetter-Ferguson with AI, Life Insurance Reserving AI — VM-20, PBR & Actuarial, Insurance Claims Analytics AI — Loss Development & IBNR.