Legacy App Modernization in Paris, France

Insurance / Mutual / Legacy Modernization / Paris

A Paris-based French mutual insurance group (EUR 8.4B gross written premium across auto, home, health, and professional liability, 12 million members, 18,000 employees, operating as a mutual — societe d'assurance mutuelle) needed to modernize their core insurance administration platform. The group operated a dual-mainframe legacy estate reflecting French technology history: an IBM z/OS mainframe running COBOL programmes for auto and home insurance (3.2 million lines of COBOL, deployed incrementally since 1978), and a Bull GCOS 8 system running PL/I and COBOL for health insurance and professional liability (1.8 million lines, deployed since 1982). Together, these systems administered 12 million member policies across all product lines, processed EUR 5.2B in annual claims payments, and generated regulatory reporting for ACPR. Core challenges: (1) Dual-mainframe complexity — the group operated two entirely separate mainframe environments with no shared architecture. IBM z/OS (auto and home): 3.2 million lines of COBOL processing 7.4 million auto and home policies. The system managing: French auto insurance (assurance automobile — with France's specific bonus-malus coefficient system, Convention IRSA for inter-company claims settlement, and Loi Badinter victim compensation rules), and multi-risk home insurance (MRH — multirisque habitation — with France's specific natural catastrophe regime under Cat Nat, the Fonds de Garantie, and French tenant/landlord insurance obligations). Bull GCOS 8 (health and professional liability): 1.8 million lines of PL/I and COBOL processing 4.6 million health and professional liability policies. The system managing: complementary health insurance (complementaire sante — mandatory since the ANI of 2013, with specific RAC 0 — reste a charge zero — processing requirements), and professional liability (responsabilite civile professionnelle — with France's specific decennial liability regime for construction professionals under the Loi Spinetta). Bull GCOS expertise was nearly extinct: only 3 developers in France could modify the system with confidence, all contractors charging EUR 1,200/day. Bull/Atos provided maintenance support but was actively encouraging customers to migrate off GCOS. (2) Regulatory modernization pressure — ACPR's supervisory assessment had noted: "The group's dependency on two ageing mainframe platforms, including a Bull GCOS system approaching end of vendor support, represents a material operational risk that the board should address with a credible modernization timeline." This examination finding created board-level urgency. Additionally, France's DDA (Directive sur la Distribution d'Assurance — implementing EU IDD) required insurers to demonstrate that they could provide appropriate product information to customers — requiring data analytics capabilities that the legacy systems could not support. Solvency II pillar 3 reporting required data extraction from both mainframes through a manual process taking 6 weeks per quarterly submission — the ACPR expecting faster, more accurate reporting. (3) French auto insurance complexity — France's auto insurance system was uniquely complex, with rules encoded in legacy COBOL over 45 years. Bonus-malus (CRM — Coefficient de Reduction-Majoration): each policyholder's premium modified by their claims history, calculated according to a precise regulatory formula (Code des assurances Article A121-1). The COBOL system tracked CRM coefficients back to 1978 — 45 years of claims history for long-standing members. Convention IRSA (Convention d'Indemnisation et de Recours entre Societes d'Assurance): the inter-company agreement governing how French insurers settled claims between themselves. IRSA processing involved: baremes (tariff tables for damage types), determination of responsibility percentages using IDA (International Declaration of Accident) diagrams, and inter-company financial settlement flows. This convention-specific processing was deeply embedded in COBOL business logic. Loi Badinter (1985): French victim compensation law creating automatic right to compensation for traffic accident victims regardless of fault — with specific rules for vehicle occupants, pedestrians, and cyclists. The COBOL system implementing Badinter-specific processing for bodily injury claims. Cat Nat (catastrophe naturelle): France's specific natural catastrophe insurance regime — mandatory coverage included in every home and auto policy, with claims triggered by government decree (arrete interministeriel). The COBOL system managing Cat Nat claims processing, Fonds de Garantie interface, and reinsurance cession to CCR (Caisse Centrale de Reassurance). (4) Health insurance complexity — French complementary health insurance (complementaire sante) was equally complex. RAC 0 (reste a charge zero): the "100 percent sante" reform requiring complementary health insurers to cover specific optical, dental, and hearing aid expenses with zero remaining charge to the insured. The GCOS system implementing RAC 0 processing with: tariff tables for each product category (optical: frame + lens combinations; dental: prosthetics, orthodontics; hearing: hearing aid categories), provider network management (network versus out-of-network reimbursement differences), and real-time interface with CPAM (Caisse Primaire d'Assurance Maladie) for coordination of benefits between mandatory social security and complementary coverage. Tiers payant (third-party payment): the French system where the insurer pays the healthcare provider directly (rather than reimbursing the member) — requiring real-time processing through the NOEMIE interface with CPAM and direct payment to 280,000 healthcare providers. The GCOS system processing 42 million health claims annually through these interfaces. (5) French labour law constraints — the group's CSE (comite social et economique — employee representative body) had to be consulted on the modernization programme because it represented: a significant technology change affecting working conditions, potential changes to employee roles (mainframe specialists transitioning to modern technology), and training obligations for affected staff. CSE consultation required: detailed information sharing 15 days before formal consultation, a formal opinion (avis) from the CSE, and documentation of how employee impacts would be managed (training plans, role evolution, potential GPEC). This process added 3-4 months to the programme timeline but was legally mandatory and culturally important in French organisational governance.

We delivered a comprehensive dual-mainframe modernization over 56 weeks — replacing both IBM z/OS and Bull GCOS 8 with a unified, modern insurance platform while maintaining zero-downtime operations for 12 million members. (1) Dual-mainframe business rule extraction. IBM mainframe analysis: automated parsing of 3.2 million COBOL lines extracting 6,800 business rules covering: auto insurance (bonus-malus calculation, Convention IRSA processing, Loi Badinter bodily injury, Cat Nat claims, French auto regulatory reporting), home insurance (MRH coverage logic, natural catastrophe processing, landlord/tenant specific rules, burglar alarm discount calculations), and cross-product rules (multi-product discount logic, member loyalty calculations, direct debit and payment processing). Bull GCOS analysis: extracting 4,200 business rules from 1.8 million lines of PL/I and COBOL — a more complex extraction because: GCOS used different file management (UFAS — Universal File Access System) than IBM's VSAM, PL/I code structures differed significantly from COBOL, GCOS job control (GCL — GCOS Command Language) had no IBM JCL equivalent, and GCOS database structures (IDS/II — Integrated Data Store) differed from DB2. We built custom extraction tooling for the GCOS environment — one of fewer than 5 teams globally with active GCOS reverse-engineering capability. Actuarial and regulatory validation: every extracted rule validated against the Code des assurances (insurance code), ACPR regulations, convention IRSA specifications, and the group's actuarial documentation. This validation revealing 28 calculation discrepancies — rules that had diverged from regulatory specifications through decades of tactical modifications. All corrected in the modern system with actuarial and compliance sign-off. (2) Unified modern platform. Architecture: Kubernetes (OVHcloud — French sovereign cloud provider, meeting ANSSI expectations) with: Java/Kotlin microservices for insurance calculation engines, PostgreSQL for relational data (replacing both DB2 and IDS/II databases), event-driven architecture replacing mainframe batch processing, Vue.js frontend (French development community preference) for agent and member portals, and API gateway enabling integration with CPAM/NOEMIE, Convention IRSA, CCR, and regulatory reporting systems. Product configuration engine: all insurance products (auto, home, health, professional liability) defined through configuration rather than code. The bonus-malus coefficient system parameterised from regulatory tables (not hardcoded). RAC 0 tariffs configurable for annual updates. Convention IRSA baremes updatable through configuration. French regulatory integration: NOEMIE interface (real-time health claims coordination with CPAM), IRSA settlement interface, CCR reinsurance cession interface, ACPR reporting interface, and FFA (Federation Francaise de l'Assurance) statistical reporting. (3) Parallel running: validating 12 million member calculations. Both mainframes and the modern system processing all operations simultaneously: auto and home: 7.4 million policies — premium calculations, claims processing, bonus-malus coefficient updates compared daily. Health: 4.6 million policies — 42 million annual claims, CPAM coordination, provider payments compared daily. Any discrepancy exceeding EUR 0.01 investigated. After 10 weeks of parallel running: discrepancy rate at 0.002 percent (88 policies across 12 million — each attributable to legacy data quality issues). (4) Migration with CSE compliance. CSE consultation: conducted during weeks 4-16 — presenting the modernization programme, employee impact assessment, training plans (mainframe specialists offered 6-month reskilling programme to modern technologies — Java, Kubernetes, cloud), and role evolution documentation. CSE issued a positive opinion (avis favorable) based on the comprehensive training and transition plan. Migration sequence executed over weeks 16-52, product line by product line, with each phase validated through parallel running before cutover. (5) ACPR-compliant throughout. Regulatory communication: ACPR notified at programme initiation. Quarterly progress reports submitted. Parallel-running validation reports shared. Post-modernization ACPR examination: zero findings related to modernization — examiner noting "materially improved IT risk posture and regulatory reporting capability."

Results over 12 months post-modernization. System unification: two mainframe environments (IBM z/OS + Bull GCOS 8) consolidated into one modern platform. 5 million lines of legacy code (3.2M COBOL + 1.8M PL/I/COBOL) replaced with modern configurable platform. 12 million member policies administered on unified system. Bull GCOS dependency eliminated — removing the most critical technology risk (near-extinct expertise, approaching vendor end-of-support). Processing efficiency: health claims processing: 42 million annual claims processed in real-time (replacing batch). Tiers payant (direct provider payment): from 5-day cycle to same-day. Bonus-malus calculations: real-time (replacing annual batch). Cat Nat claims activation: from 48-hour mobilization to 4-hour (automated processing triggered by government decree publication). Regulatory reporting: ACPR Solvency II pillar 3 from 6 weeks manual to 8 hours automated. Regulatory change implementation: from 6-8 months to 3-5 weeks (configuration versus code). ACPR examination: zero modernization-related findings. Digital capabilities: member self-service portal: 2.8 million members registered (from zero). Mobile app: real-time policy, claims, and health reimbursement tracking. Agent portal: modern interface replacing mainframe green-screen terminals. Cost: dual mainframe elimination: EUR 7.2M annually (IBM z/OS: EUR 4.8M + Bull GCOS: EUR 2.4M). Modern platform: EUR 2.1M annually. Annual infrastructure saving: EUR 5.1M. Mainframe specialists (3 GCOS contractors at EUR 1,200/day + 6 IBM COBOL developers): EUR 2.8M annually — transitioning to modern team: EUR 1.6M. Net annual talent saving: EUR 1.2M. Total modernization investment: EUR 9.4M (ZTABS engagement + internal team + OVHcloud + training programme over 56 weeks). Annual savings: EUR 6.3M (infrastructure + talent + processing efficiency). Payback period: 18 months. 5-year TCO reduction: EUR 22.1M.