AI Copilot Development in Munich

Insurance / Global Reinsurance

A Munich-based global reinsurer (EUR 52B gross premiums written, 40,000 employees across 50 countries, operating in all major reinsurance lines — property catastrophe, casualty, life and health, and specialty) faced productivity and knowledge management challenges across its underwriting and actuarial functions. (1) Treaty renewal volume: the reinsurer processed approximately 35,000 treaty renewals annually (concentrated in the January 1 and July 1 renewal seasons). Each renewal required: portfolio analysis (reviewing the ceding company's loss experience, premium volume, and risk profile — typically from a data submission of 50-200 pages), exposure modelling (for catastrophe-exposed treaties — running the cedant's portfolio through catastrophe models to assess risk), pricing analysis (applying the reinsurer's proprietary pricing models — with parameters adjusted for the specific risk, and benchmarking against market pricing), treaty wording review (reviewing the proposed contract terms — coverage provisions, exclusions, conditions, dispute resolution — comparing against the reinsurer's standard wordings and identifying deviations), and submission preparation (preparing the internal underwriting submission — a structured document presenting the analysis, risk assessment, and pricing recommendation for the underwriting committee). Thorough analysis of a complex treaty took 3-5 days for underwriters and 2-3 days for actuaries. During the January 1 renewal season (when approximately 60% of treaties renewed), the workload was impossible — underwriters and actuaries worked 70+ hour weeks and still could not analyse every renewal at the depth they wanted. Triage was necessary: complex, large-premium treaties received thorough analysis, while smaller or more straightforward renewals received abbreviated review. (2) Actuarial documentation burden: Solvency II regulatory requirements meant that actuarial models, assumptions, and results had to be extensively documented. The Solvency II standard formula and internal model calculations required: model documentation (comprehensive documentation of every actuarial model — methodology, assumptions, limitations, validation), ORSA narratives (Own Risk and Solvency Assessment — a forward-looking assessment of the reinsurer's risk and capital position, requiring actuarial analysis and narrative explanation), SCR calculation documentation (Solvency Capital Requirement — documenting how the SCR was calculated, which models were used, and what assumptions were made), and reserving reports (quarterly actuarial opinions on loss reserves — with detailed analysis of reserve development by line of business). Actuaries estimated that 45% of their time was spent on documentation rather than analytical work. (3) Knowledge concentration: the reinsurer's most valuable underwriting expertise resided with approximately 200 senior underwriters (average tenure: 18 years). These individuals carried critical knowledge: client relationship history (decades of relationship context with ceding companies — their claims behaviour, negotiation patterns, and risk management quality), market cycle experience (understanding how pricing, capacity, and terms evolved across multiple underwriting cycles — essential for appropriate cycle positioning), and specialist technical expertise (deep knowledge of specific risk categories — earthquake engineering, hurricane modelling, cyber risk quantification, aviation liability). With 30% of these senior underwriters eligible for retirement within 5 years, the reinsurer faced a significant knowledge loss risk. Junior underwriters required 8-12 years to develop comparable expertise — and in a competitive talent market, many departed before reaching senior level. (4) Cross-office inconsistency: the reinsurer operated from 12 major underwriting offices worldwide (Munich, London, New York, Singapore, Sydney, Toronto, and others). While corporate underwriting guidelines existed, interpretation and application varied by office — a treaty presented to Munich might receive different terms and pricing than the same treaty presented to New York. This inconsistency created adverse selection risk (brokers learning which office offered the most favourable terms) and complicated portfolio management. (5) Emerging risk assessment: new and evolving risk categories — cyber, climate change, pandemic, supply chain disruption — required underwriting approaches that could not rely heavily on historical loss data. Senior underwriters applied judgment and external research to assess these risks, but this analysis was ad hoc and not systematically captured or shared.

We built an actuarial and underwriting copilot over 18 weeks, deployed across the reinsurer's global underwriting platform. (1) Treaty Renewal Pre-Analysis: automated submission analysis. Data submission ingestion: ceding company submissions (arriving as PDF, Excel, and structured data files) automatically ingested and parsed. The copilot extracted: portfolio characteristics (lines of business, premium volumes, geographic distribution, loss experience — 5-10 year history), exposure data (for catastrophe-exposed treaties — aggregate exposure by geography, occupancy, construction — formatted for catastrophe model ingestion), claims data (individual large claims and aggregate loss development — with trend analysis and benchmarking against industry loss ratios), and ceding company profile (financial strength — using AM Best, S&P ratings and financial data, regulatory information, market reputation). Pre-analysis report: for each renewal, the copilot generated a structured pre-analysis — portfolio overview, loss experience assessment (trend analysis, frequency/severity decomposition, comparison to prior year and industry benchmarks), exposure assessment (catastrophe model results summary, aggregate exposure analysis), pricing indication (applying the reinsurer's pricing models with the parsed data and presenting preliminary rate indications), wording comparison (comparing the proposed treaty wording against the reinsurer's standard terms — flagging deviations, identifying provisions missing from the proposal, and noting terms that were more or less favourable than standard). Time savings: underwriter analysis reduced from 3-5 days to 1-2 days per treaty, and actuarial pricing analysis reduced from 2-3 days to 1 day — because the copilot performed the data extraction, initial analysis, and documentation that previously consumed most of the professional's time. (2) Solvency II Documentation: automated regulatory reporting. Model documentation copilot: the copilot generated model documentation per Solvency II requirements — describing the methodology, assumptions, and limitations of each actuarial model in the standardised format expected by BaFin and the company's internal model governance framework. When models were updated, the copilot drafted the corresponding documentation updates. ORSA narrative generation: the copilot drafted ORSA narrative sections — synthesising actuarial analysis results into the forward-looking risk and capital assessment that ORSA requires. Actuaries reviewed and refined the copilot's draft rather than writing from scratch. Reserving reports: for quarterly reserve reviews, the copilot generated the analytical sections of the reserving report — loss development triangle analysis, reserve movement commentary, and actuarial opinion drafts — with the actuary focusing their review on judgment-intensive areas (assumption changes, emerging trends, management estimates). Documentation time: reduced from 45% of actuary time to 20% — freeing actuarial capacity for analytical work. (3) Underwriting Knowledge Base: institutional expertise preservation. Knowledge capture: the copilot ingested the reinsurer's accumulated underwriting knowledge — 25 years of underwriting submissions and decisions (structured data from the underwriting system), client relationship notes (from the CRM system — enriched with context from senior underwriters during the deployment phase), claims experience data (individual large claims with detailed cause-of-loss information and outcome data), and market intelligence archives (historical market reports, renewal season assessments, and cycle positioning analyses). Expert interview integration: during deployment, we conducted structured interviews with 40 senior underwriters — capturing their expertise in specific risk categories, client relationships, and market cycle management. This knowledge was indexed and made queryable through the copilot. Contextual retrieval: when an underwriter reviewed a renewal, the copilot surfaced relevant historical context — "This cedant has presented this programme to Munich for the past 12 years. Historical loss ratio: 62% (vs. market average 71%). Three large claims in 2021-2023 totalling EUR 45M — all resolved within original estimates. Client considered low-maintenance, strong risk management." Junior underwriter augmentation: junior underwriters queried the copilot for context they would previously have needed to obtain from senior colleagues or not had at all — "What is our typical approach to pricing excess-of-loss treaties for German industrial risks?" The copilot provided structured responses drawing from the institutional knowledge base. (4) Cross-Office Consistency: global underwriting harmonisation. Pricing benchmarking: for each treaty, the copilot compared the proposed pricing against the pricing applied to comparable treaties across all offices — identifying cases where one office's pricing significantly deviated from others. Terms and conditions comparison: the copilot identified where treaty terms offered by different offices for similar risks diverged — flagging inconsistencies for senior underwriting management review. Best practice sharing: the copilot surfaced underwriting approaches and innovations from across the global network — "The Singapore office developed an enhanced pricing approach for Asian natural catastrophe treaties that improved loss ratio by 3 points — approach documentation available here." (5) Emerging Risk Intelligence: structured analysis of evolving risks. Research aggregation: the copilot continuously ingested research on emerging risks — academic publications, industry reports, regulatory guidance, and news — from internal and external sources. Emerging risk briefings: for underwriters reviewing treaties with emerging risk exposure, the copilot generated contextual emerging risk briefings — "Cyber exposure in this portfolio: estimated EUR 1.2B in cyber limits. Recent developments: ransomware frequency increased 28% YoY, average ransom payment decreased 15% as law enforcement action increased. Munich Re emerging risk view: cyber pricing adequate at current levels for primary insurance, but accumulation risk in reinsurance portfolios requires careful monitoring."

Treaty renewal throughput: underwriters analysed 35% more treaties at full analytical depth during the January 1 renewal season — reducing the triage that had previously been necessary. Revenue impact: the additional analytical capacity enabled the reinsurer to identify and capture EUR 280M in additional premium volume — treaties that would previously have been renewed on broker terms without detailed analysis were now analysed and priced appropriately, and new business opportunities that would have been declined due to insufficient review capacity were evaluated and accepted. Pricing accuracy: pricing deviation between offices reduced by 60% — cross-office consistency eliminated the adverse selection that had cost an estimated EUR 45M annually (brokers placing treaties at whichever office offered the most favourable terms). Documentation efficiency: actuarial documentation time reduced from 45% to 20% of total actuary time — freeing approximately 25 actuarial FTEs worth of capacity for analytical work (without headcount changes). Solvency II regulatory submissions: preparation time reduced by 40% — with improved consistency and completeness (BaFin review comments reduced from 12 to 3 in the first annual submission after deployment). Knowledge preservation: 40 senior underwriter interviews captured and indexed — creating a queryable institutional memory. Junior underwriter development: estimated 3-year acceleration in junior underwriter development (from 8-12 years to 5-9 years to senior-level competence) through copilot-augmented learning. Emerging risk management: systematic emerging risk intelligence improved portfolio management — the copilot identified a cyber accumulation concentration that was 18% higher than previously estimated, leading to portfolio management action that averted an estimated EUR 120M in potential unmonitored exposure. Annual value: EUR 280M (additional premium captured through enhanced analytical capacity — at the reinsurer's average combined ratio), EUR 45M (adverse selection reduction through cross-office consistency), EUR 18M (actuarial productivity — 25 FTE equivalents of documentation time freed), EUR 12M (Solvency II efficiency and reduced regulatory risk), EUR 8M (emerging risk management — portfolio optimisation) = EUR 363M. Development cost: EUR 520,000. Payback: under 13 hours.