System
Policy management (core insurance)
Knows
Contracts, guarantees, premiums, due dates, amendments — often on AS400/RPG or COBOL
Loading...
Loading...
For French and Maghreb life, non-life, bancassurance and health mutual insurers, two constraints converge: modernizing a core IT system often built on AS400, RPG or legacy COBOL, and deploying an AI orchestration layer compliant with ACPR-EIOPA, Solvency II and the AI Act. Access International combines both practices in nearshore mode from Tunis: industrialized insurance legacy migration under the ATLAS methodology, and AI workflows for underwriting, claims, fraud and advisory duty — without tracking the policyholder, by truly accompanying them.
The insurance sector — life, non-life, health mutuals and bancassurance — carries structural technical debt. A large share of mid-market French insurers and almost all Maghreb insurers run their core policy management system on AS400, RPG or legacy mainframe COBOL inherited from the eighties and nineties. The modules are stable, the processes hold, but the know-how that maintains them is fading every year. Insurance legacy modernization is no longer a roadmap option — it is a regulatory and operational continuity constraint.
Regulatory pressure accelerates: Solvency II demands granular prudential reporting that legacy systems struggle to produce without costly nightly batches; EIOPA pushes transparency and explainability of automated decisions; IFRS 17 transformed contract accounting; the AI Act classifies as high-risk insurance AI use cases touching pricing, eligibility or claim settlement. Without modernizing the core, the insurer compounds regulatory complexity and technical fragility.
On the policyholder side, expectations mirror online banking: a few-minutes underwriting, a few-hours claim handling, transparent indemnification, an advisor up to date with the file from the first call. Insurers stuck on 48-72 hour underwriting cycles or three-week settlement times on simple claims see attrition rise — often to neo-insurers built on modern stacks without legacy debt.
The French COBOL/AS400/RPG developer market shrinks every year. Tunisia nearshore for insurance — native French speaking, UTC+1 timezone aligned with Paris, GDPR and intra-EU transfers structured, an ENIT/INSAT/Esprit talent pool capable of rebuilding an RPG program's intent from its source code — has become the industrial response for companies willing to migrate their AS400 to Java, .NET or TypeScript under audited functional parity as part of our ATLAS methodology. See our technical journeys: AS400 modernization, COBOL → Java, COBOL → .NET Core, mainframe IBM → Azure.
The choice is no longer between modernizing legacy OR deploying AI. It is to run both programs coherently: a strangler fig on core insurance while an AI orchestration layer augments underwriting, claims and policyholder relationship on existing systems. Our banking sector carries the same doctrine on a closely related ground. The transition window narrows — every year without action adds debt and rupture risk.
Contracts, guarantees, premiums, due dates, amendments — often on AS400/RPG or COBOL
Declarations, expert reports, indemnifications, disputes, recoveries
Quotes, risk profiles, tariff segmentation, eligibility rules, actuarial scoring
Provision calculations, Solvency II projections, prudential ratios, EIOPA reporting
Suspicious patterns, alerts, files under investigation, identified fraud rings
Regulatory identity, supporting documents, AML profiles, sanction lists, advisory duty
Brokers, general agents, productions, commissions, framework agreements
Policyholder interactions, campaigns, segmentations, solicitations, NPS surveys
None of these systems sees the whole. The policyholder pays the price: they subscribe a health policy online and receive a sales call three weeks later for the same product; they declare a car claim Monday morning and get no news before next Thursday; they call to understand their premium calculation and the agent has to call actuarial for the answer; they change address, the info enters the CRM but does not flow to the policy management system, and their renewal notice ships to the old address. Each silo is right, the whole is wrong. And underneath, the AS400 core slows down every end-of-month batch, lengthening Solvency II reporting cycles — see our technical journeys AS400 modernization and mainframe IBM → Azure.
AI in insurance can reinforce policyholder trust or destroy it. The boundary is clear: serve vs surveil. Here are the two faces, to be confronted in every product or regulatory decision.
An AI orchestration layer integrating on top of existing systems to break silos in weeks — and in parallel, a legacy modernization program under the ATLAS methodology to migrate the AS400/RPG/COBOL core insurance to Java, .NET or TypeScript without service disruption. Here are six concrete use cases we know how to deploy for life, non-life, bancassurance and health mutual insurers.
A life insurer operates its policy management on AS400/RPG for twenty-five years. RPG know-how erodes, evolutions take six months when they should take a few weeks, and Solvency II reporting needs three days of batches. Under the ATLAS methodology, the Access nearshore team in Tunis maps the estate program by program, extracts business rules into a discrepancy registry signed by program directors, and migrates by strangler fig — one subsystem at a time — to Java Spring Boot. Production runs in parallel during each wave until audited functional parity. No functionality exits scope without written business validation.
ATLAS methodology, RPG/COBOL estate mapping, discrepancy registry, strangler fig, migration to Java Spring Boot or .NET Core, audited functional parity, Tunis nearshore team + France co-delivery via Vivantro.
The policyholder sees no interruption during migration. Request processing times decrease progressively as modern bricks replace nightly batches. Advisory duty is better served because the advisor accesses policy data in real time, not after a batch.
Legacy maintenance cost decreases wave after wave. Solvency II reporting moves from three days to a few hours. The capacity to launch new products multiplies because evolutions ship in weeks, not quarters. Technical debt ceases to be a strategic blocker.
Human risk (senior RPG developer retirement) ceases to be blocking because knowledge is documented and transferred to Java teams available at scale. The program continuously produces auditable documentation usable for Solvency II and the AI Act.
A prospect requests a life or health insurance quote online. They fill the form, upload supporting documents (ID, bank details, latest tax notice for tariff scoring). The orchestration layer validates in parallel: identity (KYC), documentary compliance, risk profile, engagement capacity, tariff segmentation, advisory duty. The prospect receives in minutes: a reasoned personalized quote, or a request for additional documents with a precise list, or motivated refusal with possible appeal. No more opaque 48-72 hour queues.
Multimodal document analysis, KYC and sanction list integration, actuarial tariff scoring, underwriting journey orchestration, HITL framework for borderline cases, native advisory duty compliance.
The prospect perceives a responsive insurer that says yes or no clearly. Sense of respect, not opacity. Adhesion rate up, loss to neo-insurers down.
Digital path abandonment rate drops. The underwriting cycle compresses, generated cash flow arrives faster. Advisory duty compliance is automatically documented, reducing post-underwriting dispute risk.
The underwriting team processes five to ten times more cases at constant headcount. Human underwriters focus on real risk cases or high-value commercial arbitrations. Workload more evenly distributed, team less stressed.
A policyholder declares a car or home claim via the mobile app. They upload photos, repair quotes, joint accident statement. The orchestration layer immediately cross-references: contract guarantee conditions (from core insurance), claim history, risk profile, anti-fraud scoring, expert assessment estimated by visual AI. Simple cases below threshold are settled directly after HITL human-expert validation. Complex or risky cases route to a specialized expert with full context and a recommended settlement scenario. Target: settlement in days, not weeks.
Visual analysis of claim photos, RAG on guarantee conditions, multi-signal anti-fraud scoring, multi-system context aggregation, HITL framework for expert validation, multi-channel policyholder notification orchestration.
The policyholder leaves the declaration with a credible estimate and an announced deadline. Settlement arrives in days for simple cases. Sense of effective protection, no opaque queue. Post-claim NPS — the moment of truth in insurance — moves from detractor to promoter.
Claim-management costs drop through automated processing of simple cases. Finer fraud detection reduces undue indemnifications. Post-claim retention — a key profitability factor — improves significantly.
Human claim experts handle cases that truly justify their expertise. Upstream triage is automated. Processing capacity rises without additional hiring, which matters in a market where expert recruitment is tight.
A policyholder declares a home claim. The analysis engine cross-references: claim history, geolocation, contract seniority, geographic-zone patterns, consistency between provided photos and quotes, behavioral signals (declaration at month-end, amount near guarantee ceiling). Rather than blindly blocking or sending an investigator, orchestration produces a suspicion score with reasoned context. Weak cases are processed in flow. Investigation-worthy cases receive a human investigator with a pre-instructed file. Organized rings are detected by cross-referencing several apparently independent files.
Real-time anomaly detection, multi-signal analysis, criticality scoring, RAG on historical fraud cases, network pattern detection, HITL framework, AI Act limited-risk compliance.
The honest policyholder does not suffer abusive investigation on a legitimate claim. They are settled quickly. Only cases with suspicious signals get justified investigation. Sense of protection without generalized suspicion.
Real fraud losses drop thanks to better detection of borderline cases. Customer-friction losses (honest policyholders treated as suspects) drop even more. Investigation costs concentrate on real high-stakes files.
Investigators handle fewer false positives and more real high-impact investigations. The HITL framework learns from each correction, the model improves continuously without drift on discriminating profiles forbidden by the AI Act.
An agency or remote advisor receives a policyholder for a life-insurance arbitration or a health-coverage review. The AI copilot prepares a brief: global asset situation, detected life events (child birth, divorce, near retirement), identified tax optimization opportunities, advisory-duty vigilance points (MIFID risk profile change). During the meeting, the advisor has natural-language access to all product documentation, internal case law, ACPR-compliant response scripts. The junior advisor delivers senior-grade quality, and every piece of advice is traced for advisory-duty audit.
RAG on insurance product documentation and internal case law, multi-system policyholder context aggregation, real-time copilot interface, native advisory-duty and MIFID compliance, auditable logging.
The policyholder perceives an insurer that knows them, not a counter that changes faces. Advisory quality is constant, advisory duty is homogeneously respected. Two years later, their story is not forgotten.
Products per policyholder increases because arbitration and complementary coverage opportunities are seized at the right time. Post-sale dispute risk on advisory-duty defects drops because written trace exists and is auditable.
New advisor onboarding moves from months to weeks for full productivity. A senior's departure is no longer a disaster because knowledge stays in the system. ACPR advisory-duty compliance is documented automatically.
A mutual member sends their health bill via the app (consultation, optical, dental, hospitalization). The orchestration layer analyzes in parallel: billing document, CCAM/NABM coding, member rights in the guarantee, remaining annual ceilings, conformity to the care basket, anti-fraud (flagged doctor or facility, recurring suspicious reason). Conformant cases are automatically reimbursed in minutes. Borderline cases (ceiling overflow, off-nomenclature act) escalate to a manager with full context. For online third-party payment, validation to partner health professionals happens in real time.
Advanced OCR on health bills, RAG on CCAM/NABM nomenclature and guarantee conditions, health anti-fraud, core insurance and member-rights integration, HITL framework for borderline cases, medical-secret and GDPR health-data compliance.
The member receives reimbursement in minutes for simple cases, not weeks. For third-party payment, no more surprise at the pharmacist or health professional counter. Sense of a mutual that really serves, not a slow operator holding cash.
Member NPS rises. Post-annual-renewal retention improves. Cost per reimbursement drops drastically. Cash tied up in pending reimbursements decreases. Capacity to propose new guarantees rises without operational overload.
Managers focus on high-value arbitrations and disputed cases. Upstream triage is automated. Fraud is better detected without burdening the journey of honest members.
Six industrial, proven steps to migrate a legacy insurance core to Java, .NET or TypeScript without service disruption. Each step produces a deliverable signed by program directors, not a promise.
Program-by-program mapping of the AS400/RPG or COBOL estate. Inventory of modules, dependencies, batches, interfaces. Identification of subsystems eligible for strangler fig as priority.
Extraction of business rules into specifications readable by insurance program directors. Glossary of cryptic variables. Validation by business experts before any target-code writing.
Definition of the target architecture (Java Spring Boot, .NET Core 8 or TypeScript/Node). Technology mapping table: every legacy pattern has its documented modern equivalent.
Migration of one subsystem at a time. Legacy and target production run in parallel during each wave. No service disruption. Continuous validation by business users.
For every migrated subsystem, an independent audit proving functional parity. Discrepancy registry signed by program directors — not promised parity, audited parity.
Definitive traffic cutover to the modern target. Controlled legacy decommission with documented rollback plan. Knowledge transfer to your teams or maintenance operated by Access under TMA.
All these workflows share one goal: increase the service felt by the policyholder when they need it. A policyholder who feels suspected when declaring a legitimate claim, who waits three weeks for a simple settlement, who changes address without their insurer noticing, becomes dissatisfied and leaves at renewal. A policyholder who feels accompanied — clear underwriting, fast claim handling, advisor aware, mutual that reimburses in minutes — recommends their insurer. The difference is measured in NPS, renewal rate, products per policyholder. And underneath, core insurance modernization serves only one thing: enabling these workflows to run in real time rather than after the nightly batch.
Architecture designed to produce granular prudential reports in real time or near real time. Core insurance modernization enables exiting nightly batch cycles and meeting EIOPA requirements without operational overload.
Advisory duty is automatically traced on every advisor-policyholder interaction. Auditable documentation of each product recommendation. HITL framework applied on at-risk arbitrations (MIFID, suitability test).
Architecture compartmentalized for health data (mutuals, provident). Legal bases documented per use case. Automated right to erasure. Strict role-based compartmentalization for medical-secret respect.
AI use cases touching pricing, eligibility and claim settlement are classified high risk. Our architecture natively integrates traceability, the HITL framework, model documentation, per-use-case compliance assessment.
Core insurance modernization facilitates IFRS 17 compliance by exposing contract bricks in modern data structures, queryable without application gymnastics over AS400.
Provided contractually. Documentation, journals, logs and architecture audited by an independent third party at defined frequency. Mandatory criterion for insurers transitioning to Solvency II and preparing for the AI Act.
Legacy track: mapping of one AS400/RPG core insurance subsystem and signed discrepancy registry. AI track: one signature workflow deployed (accelerated underwriting OR mutual medical-act validation), with NPS impact measurement on a policyholder subsegment. Joint model validation and multi-wave program sizing.
Three to four months
Legacy track: first strangler-fig migration wave to Java Spring Boot or .NET Core on a targeted subsystem, audited functional parity. AI track: extension to three or four complementary workflows (claims, advisor copilot, anti-fraud). HITL framework rollout. Solvency II and AI Act compliance validated.
Six to nine months
Legacy track: strangler-fig deployment across the whole core insurance estate, wave after wave, with audited functional parity per delivery. AI track: company-wide deployment of all workflows. Executive program and impact dashboard. Independent audit plan established.
Twelve to twenty-four months
Three reasons converge in 2026: compliance (Solvency II and IFRS 17 demand granular prudential reporting hard to produce from a legacy AS400 without costly nightly batch cycles), know-how availability (COBOL and RPG developers in France retire massively, the onshore recruitment market shrinks), and operational urgency (product evolutions take six months on AS400 when they should take weeks, slowing competitiveness against neo-insurers on modern stacks). Postponing further means compounding technical debt and rupture risk.
Strangler fig under the ATLAS methodology. Migration subsystem by subsystem, legacy and target production running in parallel during each wave, functional parity audited by an independent third party before each cutover, discrepancy registry signed by program directors. Target: Java Spring Boot, .NET Core 8 or TypeScript/Node depending on existing IT constraints. No big bang. No functionality exits scope without written business validation. Production runs uninterrupted across the whole program.
For a mid-market French insurer with a legacy AS400/RPG estate, the program deploys over eighteen to thirty-six months depending on scope. Phase 1 (pilot on one subsystem, signed discrepancy registry): three to four months. Phase 2 (first migration wave and fidelity audit): six to nine months. Phase 3 (strangler-fig industrialization on the rest of the estate): twelve to twenty-four months. At each wave, production keeps running without disruption. Duration depends mainly on the number of subsystems and the coupling level.
Both practices run in parallel, not in sequence. While the legacy modernization program progresses wave after wave (typically eighteen to thirty-six months), the AI orchestration layer deploys workflows with immediate value on existing systems: accelerated underwriting, mutual medical-act validation, advisor copilot, anti-fraud. The policyholder benefits from felt service improvements in three to six months, without waiting for the legacy modernization end. The bricks modernized along the way allow the AI layer to run in real time rather than after nightly batches.
Our architecture natively integrates multi-regulatory insurance compliance. Solvency II and IFRS 17: granular prudential reports in real time or near real time, produced from the modernized core. ACPR-EIOPA: advisory duty automatically traced, auditable documentation of every product recommendation, HITL framework on MIFID arbitrations. AI Act: AI workflows touching pricing, eligibility and claim settlement are treated as high risk with traceability, mandatory HITL, model documentation and policyholder right of appeal. Independent audits provided contractually. Compliance is not an add-on module, it is an architectural constraint integrated from the mapping phase.
Yes. The technical fundamentals are common: policy management, underwriting, claims or benefits management, anti-fraud, compliance, policyholder relationship. Sector specifics are addressed at scoping: for health mutuals, real-time medical-act validation and third-party payment; for life insurance, actuarial projection and MIFID advisory duty; for bancassurance, core banking integration and customer journey cross-referencing; for non-life, visual claim handling and anti-fraud by geographic patterns. Our ATLAS methodology applies uniformly, the business content adapts to the domain.
No publicly shareable French insurance customer case at this stage. Our ATLAS methodology and our technical journeys (cobol-to-java, as400-modernisation, cobol-to-dotnet-core) are proven on other sectors: North American public sector in co-delivery with a prime contractor, French telecom on Power BI and data engineering, NLP HITL platform in production. Operational capacity ready to start a French or Maghreb insurance program within three to four weeks. Cross-sector references shareable under NDA during proposal review. We regularly propose short POCs of four to six weeks to validate the trajectory before a multi-year engagement.
3 products are available for deployment in this sector.
Free initial scoping. We assess your context and identify the most relevant solutions.