Executive Command Center
Real-time enterprise operating picture for the Board and C-Suite. Every metric answers one question: what decision does this require, and by when? The enterprise is not short of data — it is short of connective tissue between signal and action. This is that tissue.
Enterprise Decision Intelligence Engine ACTIVE
Executive Alert Level
Signal-to-Decision Health SPINE STATUS
Enterprises rarely fail for lack of data. They fail because insight sits in silos, priorities are unclear, and teams are not aligned on the next action. These four numbers say whether the spine is carrying signal through to outcome.
AI Control Tower FLEET SUPERVISION
44 models and 12 agents run inside the enterprise. The Control Tower is the one surface that says whether the estate as a whole is safe to keep flying.
AI Control Tower
Single supervisory surface for every pricing, reserving, fraud, catastrophe and claims model running inside the insurer. The Control Tower answers one question at all times: is the AI estate safe to keep underwriting?
Model & Agent Fleet LIVE SUPERVISION
Control Gates
No AI output reaches an executive without clearing five gates. A failed gate stops the decision, not the audit trail.
Open Incidents SLA CLOCK RUNNING
Autonomy Ladder CEILING: L3
Autonomy is granted per model, never per platform. Read the ladder from the bottom up.
Control Tower Operating Doctrine
Signal-to-Decision Spine
The insurer does not have a data problem. It has a fragmentation problem. Underwriting, claims, actuarial and distribution insight sit in separate systems, priorities across lines are unclear, and functions are not aligned on what to do next. The spine is the connective tissue that carries a signal through to a measured underwriting outcome — so decisions get made before the market, the regulator or the loss experience forces them.
Fragmentation Diagnostic WHERE THE SPINE BREAKS
Most enterprises are not short of data or tools. They are short of connective tissue. These are the four breaks that stop signal from becoming action — and what the operating system does about each.
Signal Sources → Decision Yield
| Source | Volume / wk | Insights | Decisions | Yield |
|---|---|---|---|---|
| Claims FNOL & severity telemetry | 4,180 | 62 | 4 | HIGH |
| Regulatory wires — MAS · BNM · APRA | 214 | 28 | 3 | HIGH |
| Competitor pricing & product moves | 860 | 34 | 2 | MEDIUM |
| Customer voice — NPS, complaints, churn | 1,940 | 41 | 2 | MEDIUM |
| Distribution & broker telemetry | 1,120 | 26 | 1 | MEDIUM |
| Macro, cat and reinsurance capacity | 380 | 23 | 2 | HIGH |
Foresight Posture DECIDE BEFORE THE MARKET FORCES IT
A decision made under market pressure costs more and yields less than the same decision made twelve weeks earlier. The spine is scored on how early it moves.
Prioritisation Board
Insurance executives are rarely short of options — a new line, a rate action, a treaty restructure, a market entry. They are short of an agreed method for choosing between them. Every open decision is scored on the same four criteria, sequenced into a commitment horizon, and the reasoning is captured at the moment of choice rather than reconstructed at the next Board offsite.
Commitment Horizon RANKED · NOT LISTED
Four columns, and only four slots in NOW. Scarcity in the first column is the whole point — a board where everything is urgent has made no decision at all.
Scoring Criteria SAME FOUR · EVERY OPTION
The criteria matter less than the fact that they are fixed, public and applied identically to a rate action and a market entry alike. Consistency is what makes a ranking defensible to a Board and to the regulator.
| Criterion | Weight | What it tests |
|---|---|---|
| Value at stake | 35% | GWP, underwriting margin or capital created or protected |
| Urgency & forcing event | 25% | Cost of missing the renewal, filing or treaty window |
| Confidence in evidence | 20% | Credibility of the experience data and actuarial basis |
| Reversibility | 20% | Cost of unwinding — a rate filing versus a licence |
Capture at the Moment of Choice
Rationale reconstructed six months later is a story. Rationale captured at the moment of choice is evidence. Every ranking action on this board writes four fields straight into Decision Memory.
AI Harness
The Harness is not the end product. It is the connective mechanism that allows every component of the decision environment to work as one integrated system rather than as isolated AI tools. It sits directly beneath the AI Control Tower and directly above the enterprise estate — and it is what makes model-agnostic operation possible.
The Eight Component Classes WHAT THE HARNESS BINDS
Position in the Stack READ BOTTOM UP
The Harness is layer 4 of 6. It is deliberately invisible to the executive — its job is to make the layers above it possible.
Model-Agnostic by Design NO LOCK-IN
We do not need to own every component of the stack. The future insurance enterprise will run multiple foundation models, specialist models, proprietary models, agents, data providers and platforms. The Harness is what lets them work together around a decision.
Harness Operating Principles
Adoption & Enablement
An operating system nobody uses is an expensive diagram. Adoption is not a rollout event at the end of the build — it is a system in its own right. These are the feedback loops, underwriting and claims playbooks, and the community of practice that keep the decision spine alive across every business line after the launch email.
Continuous Discovery Loop ADOPTION IS A SYSTEM, NOT A ROLLOUT
The loop closes back into capture. Decentralising discovery is what moves insight cycle time from months to weeks — line underwriters and claims managers run their own broker, agent and policyholder conversations rather than queuing behind a central actuarial or research function, while the centre maintains the method rather than doing the work.
Adoption by Function
| Function | Practitioners | Weekly use | Maturity |
|---|---|---|---|
| Underwriting | 34 | 41% | EMBEDDED |
| Claims | 41 | 38% | EMBEDDED |
| Actuarial & Finance | 22 | 36% | EMBEDDED |
| Distribution & Sales | 26 | 24% | SCALING |
| Risk & Compliance | 12 | 29% | SCALING |
| Marketing & Customer | 7 | 14% | EARLY |
Enablement Library
Nine assets, each written for a named insurance role and each moving a practitioner one step along the spine. Short beats comprehensive.
Adoption Outcomes MEASURED, NOT ASSERTED
Executive Decision Queue
The heart of the platform. Every decision card contains AI recommendation, evidence, scenarios and value at stake. The primary unit of executive work is a high-quality decision.
VALUE PROTECTED: USD 180M
Enterprise Decision Intelligence Engine
The AI engine continuously senses the enterprise, predicts outcomes, simulates options and recommends high-confidence decisions. Every stage is monitored and audited in real time.
Decision Pipeline 9 STAGES ACTIVE
Engine Performance
AI Control Tower CORE
ENGINE
Financial Performance Intelligence
Every financial metric is framed as a decision requirement. Why did profit change? What should we do? Where should capital be allocated?
Profit Bridge — Why Did Profit Change? AI ANALYSIS
| Driver | Impact | Direction | Decision Required |
|---|---|---|---|
| GWP Growth | +USD 28M | POSITIVE | Sustain growth momentum |
| Motor Claims Severity | -USD 22M | NEGATIVE | Pricing +7% — Decision #2 |
| Cat Events Q2 | -USD 14M | NEGATIVE | Reinsurance structure review |
| Reserve Strengthening | -USD 8M | CAUTION | Reserve increase — Decision #3 |
| Investment Income | +USD 18M | POSITIVE | Monitor rate environment |
| Expense Reduction | +USD 11M | POSITIVE | Accelerate automation |
| Tax & Other | -USD 1M | NEUTRAL | No action required |
Capital Allocation Recommendation AI
Underwriting Intelligence
Where should we write more business? Where should we shrink? Where should underwriting authority change? AI answers these executive questions continuously.
Portfolio Intelligence — Expand or Contract? AI
| Line | Loss Ratio | CR | AI Verdict | Action |
|---|---|---|---|---|
| Motor Private | 72.4% | 101% | CONTRACT | +7% price, de-risk HV |
| Motor Fleet | 78.2% | 107% | SHRINK | Exit unprofitable segments |
| Property SME | 58.1% | 88% | GROW | Increase capacity 20% |
| Liability EL | 64.8% | 94% | HOLD | Monitor inflation trend |
| Cyber SME | N/A | N/A | LAUNCH | Decision #4 pending |
| Marine Cargo | 61.2% | 91% | GROW | Target ASEAN corridors |
| Trade Credit | 69.4% | 99% | REVIEW | Tighten underwriting criteria |
Catastrophe Exposure MONITOR
Claims Intelligence
Claims is where insurance profitability is won or lost. AI continuously identifies leakage, fraud, severity trends and settlement inefficiency — and recommends executive actions.
Large Loss Monitor 3 ALERTS
| Claim Ref | Reserve | Type | Status | AI Flag |
|---|---|---|---|---|
| LL-2026-0441 | USD 4.2M | PI Liability | LITIGATION | Settle early |
| LL-2026-0387 | USD 3.8M | Property Fire | ESCALATED | Fraud indicators |
| LL-2026-0312 | USD 2.9M | Motor Fleet | AGREED | Reserve adequate |
| LL-2026-0298 | USD 2.1M | Cat Flood | ASSESSMENT | Reinsurance notify |
| LL-2026-0201 | USD 1.9M | EL Disease | LEGAL | Increase reserve |
AI Decision Recommendations AGENT
Sales & Distribution Intelligence
Which channels should we expand? Which producers should we terminate? Where does pipeline velocity need intervention? AI recommends distribution decisions continuously.
Channel Profitability Matrix AI RANKED
| Channel | GWP | CR | COA | AI Verdict |
|---|---|---|---|---|
| Digital Direct | USD 625M | 91% | 8% | EXPAND |
| Bancassurance | USD 511M | 93% | 12% | GROW |
| Agency — Tier 1 | USD 680M | 94% | 18% | RETAIN |
| Agency — Tier 3 | USD 342M | 104% | 22% | RATIONALISE |
| Broker Corporate | USD 341M | 96% | 15% | REVIEW |
| Embedded (Partners) | USD 340M | 89% | 6% | ACCELERATE |
Sales Pipeline
Customer Intelligence
Understand every customer segment deeply. AI identifies churn risk, cross-sell opportunity and journey bottlenecks — translating customer data into executive decisions.
Churn Risk by Segment ACTION REQ.
| Segment | Churn Risk | Customers | LTV at Risk | Action |
|---|---|---|---|---|
| Motor — Renewal Due 90d | HIGH 24% | 48,200 | USD 89M | Retention campaign |
| Single Product Holders | MED 16% | 124,000 | USD 228M | Cross-sell AI nudge |
| Claims-Dissatisfied | HIGH 31% | 18,400 | USD 34M | CEO apology + resolve |
| Digital-Only Customers | LOW 8% | 420,000 | USD 774M | Loyalty programme |
| HNW Multi-Line | LOW 5% | 12,800 | USD 235M | Priority service |
Voice of Customer AI SENTIMENT
Enterprise Risk Intelligence
AI continuously monitors, predicts and connects risks across the enterprise — surfacing decision implications and weak signals before they become crises.
Risk Heat Map LIVE · AI UPDATED
Risk Register — Decision Implications
| Risk | RAG | Trend | Decision Required |
|---|---|---|---|
| Cyber accumulation | CRITICAL | ↑ | Product cap + reinsurance |
| Climate — flood PML | HIGH | ↑ | Treaty renewal priority |
| Motor claims inflation | HIGH | ↑ | Pricing +7% (Decision #2) |
| Reserve adequacy | HIGH | → | USD 42M increase (Decision #3) |
| Vietnam country risk | MEDIUM | ↓ | Entry approved — mitigation built in |
| AI model risk | MEDIUM | ↑ | Governance framework Q3 |
| Regulatory — ICS 2.0 | MEDIUM | → | Capital modelling review |
| Talent — actuarial | LOW | → | Succession plan in place |
Scenario Simulation Center
Change any variable — claims inflation, interest rates, cat events, growth — and watch all enterprise KPIs recalculate in real time. AI recommends the best strategy for each scenario.
Executive Variable Controls LIVE SIMULATION
Scenario Library
Executive Copilot
Ask anything. The AI synthesises evidence across all enterprise data, models and decisions — answering in plain English with supporting evidence and confidence scores.
Key alerts: Motor claims severity is up 14% MoM — pricing action needed within 8 days. Vietnam Board paper due in 4 days. Reserve adequacy gap of USD 42M identified. I am ready to help you think through any of these.
What would you like to explore first?
AI Executive Agents
12 specialised AI agents continuously monitoring, analysing and recommending across every dimension of the enterprise. Each agent contributes to active decisions.
Enterprise Knowledge Graph
The full enterprise as an interconnected intelligence map. Policies, claims, customers, reinsurance, regulations, financials and AI models — all relationships visualised and explorable.
Node Legend
Knowledge Connections
Decision Memory
Every decision — who approved it, why, on what evidence, what they expected, and what actually happened. Decision memory is what stops the enterprise relitigating settled questions and repeating available mistakes. The gap between expected and actual is what trains the next recommendation.
Decision Log — Rationale & Outcome Trail
AI Governance
Every AI model used in decision-making is inventoried, monitored and audited. Human oversight is enforced. Explainability, bias and compliance are tracked in real time.
Model Inventory 34 MODELS
| Model | Type | Health | Last Audit | Decisions |
|---|---|---|---|---|
| Motor Pricing AI | Pricing | HEALTHY | Jun 2026 | Decision #2 |
| Claims Severity Model | Predictive | HEALTHY | May 2026 | Reserve #3 |
| Fraud Detection V3 | Detection | HEALTHY | Jun 2026 | Fraud ring |
| Churn Prediction | Retention | HEALTHY | Apr 2026 | Retention |
| Cat Accumulation | Risk | REVIEW | Mar 2026 | Reinsurance |
| Vietnam Market AI | Strategy | HEALTHY | Jun 2026 | Decision #1 |
| LTV Segmentation | Customer | HEALTHY | May 2026 | Cross-sell |
| Reserve Chain Ladder | Actuarial | REVIEW | Jun 2026 | Decision #3 |
AI Governance Maturity
Strategy Execution
Corporate strategy as a live, AI-monitored execution system. Every strategic objective has decision dependencies, KPIs and risk signals — updated continuously.
Strategic Pillars — AI-Monitored Execution
| Strategic Pillar | Status | KPI | Target | Actual | Key Decision | Owner |
|---|---|---|---|---|---|---|
| Geographic Expansion | AT RISK | New markets | 2 by 2027 | 1 complete | Vietnam — Decision #1 | CEO |
| Digital Transformation | ON TRACK | Digital GWP % | 30% | 22% ▲ | Digital invest approved | CDO |
| Underwriting Excellence | AT RISK | Combined Ratio | 95% | 97.4% | Motor pricing Decision #2 | CUO |
| Customer Centricity | AT RISK | NPS | +50 | +42 | Claims reform initiative | CCO |
| Capital Optimisation | ON TRACK | ROE | 14% | 12.4% ▲ | Capital allocation plan | CFO |
| Product Innovation | ON TRACK | New products | 3 launches | 2 complete | Cyber SME — Decision #4 | CUO |
| AI & Data Strategy | ON TRACK | AI adoption | 80% | 74% | AI governance approved | CDO |
| Distribution Transformation | ON TRACK | Channel efficiency | CR 92% | 91% avg | Agency rationalisation #5 | COO |
Decision Fabric
Executive decisions are never isolated. The Decision Fabric visualises how one decision cascades through the entire enterprise — showing every downstream consequence before you approve anything.
Decision Dependency Graph LIVE · AI MAPPED
Motor Pricing Decision — Full Cascade Chain
Decision Economics — ROI Ranking AI
| Decision | EV Created | Capital | Time | ROI | Rank |
|---|---|---|---|---|---|
| Cyber SME Launch | +USD 190M | USD 45M | 18mo | 422% | #1 |
| Vietnam Entry | +USD 180M | USD 120M | 36mo | 150% | #2 |
| Motor Pricing +7% | +USD 14.2M | USD 0 | 3mo | ∞ | #3 |
| Agency Rationalise | +USD 22M | USD 8M | 12mo | 275% | #4 |
| Reserve Increase | Risk mitigation | USD 42M | 0mo | Mandatory | MUST |
Enterprise Value Flow
Not KPIs. Value. Every executive decision is framed by its enterprise value consequence — created, protected, at risk or destroyed. This is the level at which Boards and investors think.
Confidence: 88% avg
5 decisions active
3 decisions at deadline
Cost of delay: USD 1.4M/day
Decision Value Ledger EVERY DECISION PRICED
Value Realisation Timeline AI PROJECTED
Cost of Delay Calculator
Enterprise Digital Twin
Not variable sliders. A complete virtual replica of the enterprise. Ask "what if" — and watch a simulated future company unfold in real time before you commit capital or strategy.
Twin Intelligence Narrative AI GENERATED
Enterprise Cognitive Infrastructure
InsurEDIC³ is not a dashboard. It is the world's first Enterprise Cognitive Infrastructure — a living digital layer that gives the enterprise the ability to sense, reason, simulate, decide, execute and learn as a unified cognitive system.
Enterprise Nervous System Architecture 8 COGNITIVE LAYERS
SENSING
MEMORY
REASONING
COGNITION
IMAGINATION
CORTEX
ACTION
LEARNING
ECI Maturity Roadmap — From Dashboard to Cognitive Enterprise
Why Enterprise Cognitive Infrastructure? POSITIONING
Not just a data store — the enterprise knows what it knows, how things connect and what has happened before.
The ability to reason across interconnected decisions — not execute them in isolation — is what separates cognitive from intelligent.
The capacity to simulate a future state before committing to it. This is the definition of strategic intelligence.
Every decision outcome makes the system smarter. The enterprise accumulates experience — not just data.
Market Category Positioning
Regulatory Command Centre
The supervisory and industry-development picture in one operating view. Two questions run through every panel: what is the industry being asked to comply with, and what is the industry being asked to build. Everything here sits at sector aggregate level — InsurEDIC³ holds no firm-level book, and this view cannot produce one.
Four Supervisory Pillars — Current Posture
Priority Supervisory Signals
What the Regulator Is Asking the Industry to Build
| Development objective | Instrument | What it asks of industry | Status |
|---|---|---|---|
| Alternative risk transfer capacity | Proposed PCC framework | Stand up captive, rent-a-captive, ILS and sovereign risk pool structures without incorporating a separate legal entity per programme — the cost of the vehicle was the deterrent, not the appetite | CONSULTED |
| ILS market depth | ILS Grant Scheme, Jan 2026 – Dec 2028 | Issue catastrophe bonds, collateralised reinsurance and sidecars from Singapore. Property cat bonds with any APAC risk attract 70% of upfront issuance costs capped at S$1m; without APAC risk, 50% capped at S$1m; non-property cat bonds 70% capped at S$1m; collateralised reinsurance and sidecars 70% capped at S$500,000; renewals 30% capped at S$500,000 MAS | OPEN |
| Closing the Asian protection gap | Policy priority | Build products and capacity for underinsured Asian risk — natural catastrophe, longevity, mortality, operational and cyber | STANDING |
| Centre-of-excellence capability | Global-Asia Insurance Partnership | A tripartite partnership between the global insurance industry, regulators and policymakers, and academia, established in Singapore with an initial focus on pandemic and climate risk | OPERATING |
| Regional disaster risk financing | SEADRIF | Support the first regional catastrophe risk facility established in Asia by ASEAN member states, incorporated and licensed as a general insurer in Singapore in October 2019 | OPERATING |
| AI-ready workforce | GenAI Jobs Transformation Map | MAS and IBF, with Workforce Singapore, partnered eleven financial institutions — including Income Insurance, Manulife and Prudential — to pilot workforce transformation and prepare the sector workforce for an AI-enabled future IBF | RUNNING |
Regulatory Signal Scanner
Every regulatory instrument moves through the same life cycle, and every stage of that life cycle imposes a different decision on a licensed entity. The scanner tracks where each instrument sits, what it triggers, and how much runway remains before the obligation bites.
Instrument Life Cycle Board
| Instrument | Stage | Key date | Decision it forces on a licensed insurer |
|---|---|---|---|
| Guidelines on AI Risk Management Consultation P017-2025 | PENDING FINAL | Consulted 13 Nov 2025 – 31 Jan 2026 MAS | Build the AI inventory now or wait. Board-level AI oversight, three lines of defence, life-cycle controls from development through retirement, and materiality-based proportionality. A twelve-month transition is proposed from issuance |
| Protected Cell Companies framework Consultation P013-2026 | PENDING FINAL | 7 Jul – 7 Aug 2026 MAS | Whether to restructure existing SPV-based risk programmes into Cells once the Act is in force. Draft PCC Act and subsidiary legislation to be consulted on later |
| Operational Risk Management guidelines | PENDING FINAL | Consultation issued SOURCED | Re-map the operational risk taxonomy and control library against the revised expectations |
| Third-Party Risk Management guidelines | TRANSITION | Consultation issued SOURCED | Re-tier the vendor estate, including AI vendors, and rebuild the exit and concentration analysis |
| Guidelines on Transition Planning | TRANSITION | Final, March 2026 SOURCED | Stand up a transition plan covering physical and transition climate risk for both the underwriting book and the investment portfolio |
| Technology Risk Management notices | TRANSITION | Consultations planned SOURCED | Prepare for IT asset management, continuous system monitoring and enhanced oversight of critical systems |
| Notice 133 — AT1 / Tier 2 criteria | TRANSITION | From 1 Jan 2026 MAS | Capital instruments recognised as AT1 or Tier 2 under RBC 2 must be sold only to persons who are not retail investors in Singapore |
| Notice 133 — Valuation and Capital Framework (RBC 2) | IN FORCE | Amended 19 Dec 2022 MAS | Supervisory intervention levels, valuation of life and general policy liabilities, and total risk requirement calculation. Applies to all licensed insurers; sections 2–5 exclude captives, marine mutuals and SPRVs |
| IFRS 17 Insurance Contracts | IN FORCE | Effective 1 Jan 2023 SOURCED | Portfolio definition and aggregation level, LIC and LRC calculation tooling, risk adjustment methodology, and revenue recognition pattern aligned to the underwriting plan |
| Guidelines on Environmental Risk Management (Insurers) | IN FORCE | Plus transition addendum MAS | Board-level sustainability goals and risk tolerance; climate scenario analysis across portfolios; environmental risk disclosure |
| Guidelines on Individual Accountability and Conduct | IN FORCE | Five high-level outcomes MAS | Name senior managers, map material risk personnel, and evidence conduct standards across all employees |
| Guidelines on Outsourcing | IN FORCE | Risk management of outsourcing MAS | Register, assess and monitor outsourcing arrangements including cloud and analytics |
| Management of Outward Reinsurance Arrangements | IN FORCE | Requirements and principles MAS | Document reinsurance strategy, counterparty selection and credit exposure control |
| AML / CFT for general and A&H business | IN FORCE | Guidelines MAS | Processes and controls to prevent money laundering and counter terrorism financing across general, reinsurance and accident and health business |
| Policy Owners' Protection Scheme | IN FORCE | DIPOP-N02 MAS | Direct insurers comply with scheme membership and levy obligations |
| ILS Grant Scheme | INCENTIVE OPEN | Jan 2026 – Dec 2028 MAS | Whether to bring an issuance to Singapore. Now extended to non-APAC risks and renewals, while prioritising issuances addressing APAC protection needs |
| FinTech Regulatory Sandbox | INCENTIVE OPEN | Standing MAS | Whether to test a new insurance innovation in a controlled environment before wider adoption |
| GenAI Jobs Transformation Map | INCENTIVE OPEN | MAS · IBF · WSG IBF | Whether to map AI adoption to specific job roles and enter the upskilling and reskilling pipeline |
AI Risk Management Readiness
Singapore's AI supervision arrived in pieces — the FEAT Principles in 2018, the Veritas Initiative, the AI Model Risk Management information paper in 2024, Project MindForge on generative AI. The proposed Guidelines pull those pieces into one instrument and shift the register from ethical principles to supervisory expectations. This view maps what that means operationally.
Lineage of the Instrument
FEAT Principles
Fairness, Ethics, Accountability and Transparency in the use of AI and data analytics in the financial sector.
Veritas Initiative
Industry consortium work turning the FEAT principles into assessment methodology and open-source toolkits.
AI Model Risk Management
Information paper on observed practices in AI model risk management across financial institutions.
Project MindForge
Generative AI risk framework, with attention to hallucination, prompt injection and data leakage.
Proposed AIRM Guidelines
Consolidated supervisory expectations across the full AI life cycle, including generative AI and autonomous agents.
Five Expectation Domains — What Each Requires
1 · Oversight and Governance BOARD LEVEL
The draft places the board and senior management at the centre of AI risk governance — the point being that AI risk is a leadership responsibility, not something delegated to the technology function.
2 · Risk Management Systems, Policies and Procedures
An enterprise-wide AI risk framework with systematic internal control procedures, applied proportionately to the size and nature of activities, use of AI, and risk profile.
3 · Life Cycle Controls
Controls across development, deployment, monitoring and retirement — not a single gate at go-live.
4 · Capabilities and Capacity SCARCEST
The Guidelines address the capabilities and capacity needed for the use of AI. This is the domain a platform cannot solve with software alone — it is why the Academy sits inside InsurEDIC³ rather than beside it.
Third-Party and Agentic AI — the Two Hard Edges
Vendor AI does not transfer the obligation
Third-party AI tools fall within scope. An insurer cannot delegate governance to the vendor. In practice this means every externally supplied model — pricing, fraud, triage, document extraction, chat — needs the same inventory entry, materiality classification, validation evidence and monitoring record as an internally built one.
Agentic AI is explicitly in scope
The proposed scope covers generative AI, AI agents and newer technologies. MAS has confirmed publicly that the Guidelines apply to all AI use cases including agentic AI. An agent that can take an action — bind, decline, settle, refer, price — is a different supervisory object from a model that produces a score a human then acts on.
Sector Readiness — Anonymised Aggregate
| Control domain | Established | In progress | Not started | Where the gap concentrates |
|---|---|---|---|---|
| Board-level AI risk mandate | 58% | 31% | 11% | Branches and smaller general insurers |
| Comprehensive AI inventory | 34% | 47% | 19% | Shadow AI in distribution and operations teams |
| Documented materiality method | 29% | 44% | 27% | No industry-standard scale to anchor against |
| Independent model validation | 41% | 38% | 21% | Validation capacity, not validation policy |
| Post-deployment drift monitoring | 26% | 41% | 33% | Vendor-hosted models with no telemetry access |
| Third-party AI governance | 22% | 39% | 39% | Contracts predating the AI question entirely |
| GenAI-specific controls | 19% | 42% | 39% | Hallucination, prompt injection, data leakage |
| Agentic authority envelopes | 9% | 27% | 64% | The newest expectation and the thinnest practice |
Obligations Register
A single consolidated register of the regulatory instruments a MAS-licensed insurer or reinsurer operates under, grouped by supervisory theme, with the decision each one forces and the InsurEDIC³ component that carries the evidence for it.
Consolidated Register
| Theme | Instrument | Applies to | Obligation in operating terms | InsurEDIC³ evidence component |
|---|---|---|---|---|
| CAPITAL | MAS Notice 133 — Valuation and Capital Framework for Insurers (RBC 2) | All licensed insurers; s.2–5 exclude captives, marine mutuals, SPRVs | Supervisory intervention levels, valuation of life and general policy liabilities, and calculation of the total risk requirement across C1, C2 and other components MAS | Financial Intelligence · capital adequacy decision object |
| CAPITAL | Notice 133 — AT1 / Tier 2 recognition criteria | Insurers issuing capital instruments | Additional criteria for recognition as AT1 or Tier 2 capital, conditional on sale only to non-retail investors in Singapore from 1 January 2026 MAS | Financial Intelligence · instrument eligibility check |
| CAPITAL | IFRS 17 Insurance Contracts | Reporting insurers | Portfolio definition and level of aggregation, LIC and LRC calculation, risk adjustment methodology, revenue recognition aligned to the underwriting plan, and sensitivity analysis SOURCED | Financial Intelligence · reserving decision chain |
| CAPITAL | Management of Outward Reinsurance Arrangements | Direct insurers | Requirements and guiding principles for reinsurance strategy, counterparty selection, credit exposure and documentation MAS | Enterprise Risk · reinsurance counterparty board |
| RESILIENCE | Technology Risk Management | All FIs | Technology risk governance and controls; planned notice updates on IT asset management, continuous system monitoring and enhanced oversight of critical systems SOURCED | AI Control Tower · critical system register |
| RESILIENCE | Cyber Hygiene requirements | All FIs | Baseline cyber controls, reinforced by MAS reminders to strengthen defences in view of AI advancement SOURCED | AI Control Tower · control attestation ledger |
| RESILIENCE | Operational Risk Management guidelines | All FIs | Updated operational risk taxonomy, control library and loss event capture — consultation issued SOURCED | Enterprise Risk · operational risk decision object |
| RESILIENCE | Third-Party Risk Management guidelines | All FIs | Vendor tiering, concentration analysis, exit planning — expanded to cover AI vendors — consultation issued SOURCED | AI Control Tower · third-party AI register |
| RESILIENCE | Guidelines on Outsourcing | All FIs | Risk management of outsourcing arrangements including cloud and analytics services MAS | AI Control Tower · outsourcing dependency map |
| RESILIENCE | Business Continuity Management | All insurers | Notification and submission requirements relating to business continuity management, internal audit and compliance functions, board oversight and data breaches MAS | Scenario Simulation · continuity stress library |
| AI & DATA | Proposed Guidelines on AI Risk Management | All FIs, proportionate | Oversight of AI risk management, risk systems and procedures, life-cycle controls, and capabilities and capacity — covering generative AI and AI agents MAS | AI Control Tower · full evidence stack |
| AI & DATA | FEAT Principles | All FIs | Fairness, ethics, accountability and transparency in the use of AI and data analytics SOURCED | AI Governance · fairness assessment kit |
| AI & DATA | Personal Data Protection Act | All entities | Consent, purpose limitation, notification, access and correction, and data breach notification obligations SOURCED | Data Sovereignty Charter · boundary controls |
| AI & DATA | Internal controls and business process controls guidance | All FIs | Sound practices for the internal control environment and business process controls MAS | Decision Ledger · control evidence trail |
| CONDUCT | Guidelines on Individual Accountability and Conduct | All FIs | Five high-level outcomes promoting senior manager accountability, oversight of material risk personnel and conduct standards across all employees MAS | AI Governance · accountability mapping |
| CONDUCT | Fair Dealing Guidelines | All FIs | Fair dealing outcomes for customers across product design, sales, advice and post-sale service SOURCED | Customer Intelligence · fair dealing lens |
| CONDUCT | Product development and pricing (life and ILP sub-funds) | Direct life insurers | Requirements for the development and pricing of life insurance products and investment-linked policy sub-funds MAS | Underwriting Intelligence · product gate |
| CONDUCT | AML / CFT guidelines | General, reinsurance, A&H | Processes and controls to prevent money laundering and counter terrorism financing MAS | Enterprise Risk · financial crime signals |
| CLIMATE | Guidelines on Environmental Risk Management (Insurers) | Life, general and composite insurers | Board-level sustainability goals, risk tolerance and accountability; environmental risk embedded in enterprise risk management; climate scenario analysis; transparent disclosure MAS | Enterprise Risk · environmental risk lens |
| CLIMATE | Guidelines on Transition Planning | Banks, insurers, asset managers | Final guidelines published March 2026 setting supervisory expectations for managing transition and physical climate risk through a sound transition planning process SOURCED | Scenario Simulation · climate pathway sets |
| STRUCTURAL | Insurance Act 1966 — licensing, control and takeover | All licensed insurers | Prior MAS approval for effective control (20% or more of shares or voting power), substantial shareholding, change of key executive person, chairman or director, or reduction in paid-up capital SOURCED | Strategy Execution · corporate action gate |
| STRUCTURAL | Insurance business transfer | Licensed insurers | MAS approval required for transfer of the whole or part of the insurance business, with confirmation by the High Court of Singapore SOURCED | Strategy Execution · portfolio transfer chain |
| STRUCTURAL | Insurance funds maintenance | Licensed insurers | Requirements relating to establishment and maintenance of insurance funds MAS | Financial Intelligence · fund segregation |
| STRUCTURAL | Policy Owners' Protection Scheme | Direct insurers | Scheme membership, levy and disclosure obligations under the Deposit Insurance and Policy Owners' Protection Schemes framework MAS | Financial Intelligence · scheme obligations |
Sector Readiness Heatmap
Readiness by obligation theme and entity archetype. The value of a shared platform to a supervisor is not that it reveals more about individual firms — it is that it reveals structural weakness at sector level earlier, without anyone having to hand over a book.
Readiness by Theme and Archetype
Where Structural Weakness Concentrates
Agentic authority envelopes
The thinnest column across every archetype. Agentic AI is explicitly in scope of the proposed Guidelines, but the practice of granting, bounding and revoking machine authority at runtime barely exists in the sector. This is the gap a shared control tower closes fastest, because the design is identical for every carrier.
Third-party AI governance
Concentrated in branches, MGAs and smaller general insurers — the entities most dependent on vendor models and least able to negotiate telemetry access or validation artefacts into a contract. Collective bargaining position is a genuine platform benefit here, distinct from the software.
Model validation capacity
Not a policy gap — a people gap. Firms have written the validation standard and cannot staff it. Software does not fix this. It is the reason the Academy is a first-class component of InsurEDIC³ rather than a marketing add-on.
Archetype Profiles
| Archetype | Typical constraint | What shared infrastructure changes for them |
|---|---|---|
| Global composite insurer | LOW | Least dependent on shared build — but gains from the common evidence format that lets one control set satisfy several ASEAN supervisors at once |
| Domestic life insurer | MEDIUM | Gains most on the AI life-cycle control set and on validation capacity through pooled Academy pathways |
| Mid-sized general insurer | MEDIUM-HIGH | The core case: full obligation set, fraction of the build budget. Pays for the control architecture once, through subscription, rather than designing it alone |
| Reinsurer / branch | MEDIUM | Gains on local evidence generation where the group model sits offshore and group tooling does not produce Singapore-shaped artefacts |
| Broker / MGA | HIGH | Least able to build alone, most exposed to vendor AI. Shared platform is the difference between having a governance function and having a policy document |
| Captive / PCC cell | HIGH | Thin operating teams by design. A shared decision layer is the only realistic route to the analytical capability a cell owner actually wants |
Capital, Valuation and IFRS 17
Capital is where regulatory obligation and enterprise decision-making are least separable. Every underwriting, reserving, investment and reinsurance decision resolves into a capital number, and the supervisor reads that number as the summary of everything upstream of it. This view holds the instruments and the decision chain that produces them.
RBC 2 Under Notice 133 — Structure
IFRS 17 — The Decision Chain Behind the Disclosure
Where Decision Intelligence Adds Value in the Capital Chain
Assumption traceability
Every capital and reserving assumption carries a provenance record: who set it, on what evidence, against which prior, and what changed since the last close. A supervisor asking why a number moved gets the chain, not a reconstruction.
Scenario coherence
The same shock runs through capital, reserving, reinsurance and the investment portfolio simultaneously rather than in four disconnected models producing four incompatible answers.
Decision-to-number linkage
Pricing, appetite and reinsurance decisions are recorded as decision objects linked to the capital outcome they moved — so the capital position becomes explainable as a sequence of choices rather than an emergent surprise.
Operational and Third-Party Resilience
MAS has been working to improve operational resilience across four pillars — operational risk management, technology and cyber risk management, third-party risk management, and business continuity management. AI adoption touches all four at once, which is why resilience and AI supervision cannot be run as separate programmes.
The Concentration Risk a Shared Platform Creates — Stated Plainly
An honest account of the risk this platform introduces
A common platform used by many insurers is, by construction, a concentration. If the platform fails, more than one participant is affected simultaneously. That is a real supervisory concern and it deserves a direct answer rather than a marketing one. Four design commitments follow from it:
| No operational dependency for licensed activity | InsurEDIC³ supports decisions; it does not execute regulated activity. It does not bind risk, issue policies, hold funds or settle claims. A platform outage degrades analytical capability — it does not stop a participant writing business or paying a claim |
| Participant-side instance and local fallback | Each participant's decision records, model inventory and evidence pack are held in and exportable from its own instance. A participant can operate its control framework from its own export with the platform offline |
| No shared data store to lose | Because no proprietary data is pooled, a platform compromise does not expose any participant's book. There is no aggregated policyholder dataset on this platform to breach — that is a security property of the architecture, not a policy promise |
| Exit portability by design | Frameworks, methods, control sets and a participant's own decision history are exportable in open formats at any time. Lock-in through data gravity is the failure mode this platform is specifically built to avoid |
Third-Party AI Estate — What Each Participant Must Be Able to Show
| Artefact | Who produces it | Why a supervisor asks for it |
|---|---|---|
| Model card and intended use | Platform ships it | Establishes what the model is for, and by exclusion what it is not for — the boundary a misuse finding turns on |
| Training and evaluation provenance | Platform ships it | Whether the model was built on data appropriate to the participant's book and market |
| Independent validation record | Platform ships · participant reviews | Validation cannot be self-certified by the builder alone; the participant must show it reviewed and accepted |
| Materiality classification | Participant owns | Proportionality is assessed against the participant's own use, scale and risk profile — the platform cannot classify this on the participant's behalf |
| Monitoring configuration and results | Platform ships · participant configures | Drift and performance monitoring is a continuing obligation, not a go-live gate |
| Decision trail | Participant owns | Which decisions the model actually influenced, with what inputs and what human intervention |
| Authority envelope and revocation log | Platform ships · participant sets | For agentic use: what the agent may do alone, and every instance where that authority was narrowed or withdrawn |
| Exit and retirement plan | Joint | How the participant continues to operate if the arrangement ends, and how records are retained after retirement |
Environmental Risk and Transition Planning
Insurers carry climate risk twice — once in the underwriting book as physical loss, and once in the investment portfolio as transition and repricing risk. The supervisory expectations address both, and the final Guidelines on Transition Planning published in March 2026 add the forward-looking dimension to what was previously a risk-management instrument.
Supervisory Expectations — Operational Reading
Guidelines on Environmental Risk Management
Applies to life, general and composite insurers, shaped to the nature, scale and complexity of operations.
Guidelines on Transition Planning FINAL MAR 2026
Published for banks, insurers and asset management. Sets more detailed supervisory expectations for managing transition and physical climate risk as part of a sound transition planning process.
Why Climate Analytics Is the Strongest Case for Shared Infrastructure
The economics are unambiguous here
Climate scenario analysis has the highest fixed cost and the lowest competitive value of any analytical capability an insurer is required to hold. Hazard models, downscaled climate pathways, exposure geocoding and vulnerability functions cost the same to build whether the book behind them is S$50 million or S$5 billion — and no insurer wins business because its physical hazard model is proprietary. Firms compete on appetite, pricing, distribution and claims handling, not on whether their flood depth-damage curve is secret.
Hazard model libraries, climate pathway sets, downscaling methods, vulnerability functions, scenario narratives, disclosure templates
The participant's exposure file, its geocoded portfolio, its accumulation position, its pricing loadings, its reinsurance structure
Sector-level accumulation by peril and geography, contributed voluntarily, minimum cell size five, no attribution
Conduct and Fair Dealing
AI in insurance meets conduct regulation at three points: how a risk is priced, how a claim is decided, and how a product is sold. Each is a place where a model can produce a technically defensible result and an unfair outcome at the same time — which is precisely why fairness is a named FEAT principle rather than an implied one.
Accountability Architecture
Individual Accountability and Conduct
Guidance on five high-level outcomes that financial institutions should achieve to promote the accountability of senior managers, strengthen oversight over material risk personnel, and reinforce conduct standards among all employees. MAS
Fair Dealing Outcomes
Fair dealing expectations run across product design, sales, advice and post-sale service. Where AI is used in any of those stages, the fairness of the outcome is the participant's obligation regardless of who built the model.
Fairness Testing — the Shared Method Case
| Test | Applies to | What it checks | Shared or private |
|---|---|---|---|
| Protected-attribute proxy scan | Pricing, underwriting | Whether permitted rating variables are jointly reconstructing an attribute that could not lawfully be used directly | METHOD SHARED |
| Outcome parity across cohorts | Claims, underwriting | Whether decline, referral and settlement rates diverge across cohorts beyond what risk explains | METHOD SHARED |
| Friction asymmetry | Claims | Whether some claimants face systematically more evidence requests, longer cycles or more referrals for the same claim type | METHOD SHARED |
| Explanation adequacy | All customer-facing AI | Whether an adverse decision can be explained to the affected customer in terms they can act on | METHOD SHARED |
| Elasticity separation | Pricing | Whether price movement tracks risk or tracks inferred willingness to pay | METHOD SHARED |
| Test results on the participant's own book | — | The actual numbers produced when the shared method is run against a participant's portfolio | NEVER SHARED |
Industry Growth Agenda
MAS operates two mandates simultaneously — supervising the industry and developing it. Most compliance tooling only sees the first. This view holds the second: what the industry is being invited to build, the instruments available to support it, and where a shared decision layer materially lowers the cost of participation.
MAS Development Instruments — What Is Actually Available
| Instrument | Window | Terms and intent | Where InsurEDIC³ lowers the barrier |
|---|---|---|---|
| ILS Grant Scheme | Jan 2026 – Dec 2028 | Property cat bonds covering any proportion of APAC risk: 70% of upfront issuance costs capped at S$1m. Not covering APAC risk: 50% capped at S$1m. Non-property cat bonds: 70% capped at S$1m. Collateralised reinsurance and sidecars: 70% capped at S$500,000. Renewals: 30% capped at S$500,000. Applies to natural catastrophe, longevity, mortality, operational and cyber risks. Open to onshore and offshore companies, financial institutions and multilateral organisations MAS | Structuring analytics, trigger design and basis risk modelling as shared method rather than a per-issuance consulting build |
| Proposed PCC framework | Consulted Jul–Aug 2026 | A single corporate vehicle with assets and liabilities statutorily segregated within the entity, available to MAS-licensed entities carrying out captive insurance, insurance-linked securities and sovereign risk pools MAS | Cell-level decision infrastructure that a thin cell operating team could not otherwise justify building |
| Global-Asia Insurance Partnership | Operating | A tripartite partnership between the global insurance industry, regulators and policymakers, and academia, established in Singapore as a centre of excellence in insurance and risk management with a focus on Asia, initially on pandemic and climate risk MAS | A natural counterpart for the Academy's research-to-practice pathway and for methodology validation |
| SEADRIF | Operating since 2019 | The first regional catastrophe risk facility established in Asia by ASEAN member states, incorporated and domiciled in Singapore, licensed as a general insurer in October 2019, supported by the World Bank in partnership with Japan MAS | Parametric trigger design, impact-data pipelines and payout decision logic as shared components |
| Natural Catastrophe Data Analytics Exchange | Operating | An MAS–industry–academia public-private partnership led by NTU's Insurance Risk and Financial Research Centre, focused on Singapore and Asia with global collaborations MAS | Direct architectural precedent — a shared analytical layer with sovereign data boundaries already exists in this market |
| FinTech Regulatory Sandbox | Standing | Helps financial institutions and FinTech players test new innovations in a controlled environment before wider adoption in Singapore and abroad MAS | A defined route for testing agentic decision components under supervision before scaling |
| GenAI Jobs Transformation Map | Running | MAS and IBF, supported by Workforce Singapore, partnered eleven financial institutions — Citibank, DBS, Franklin Templeton, HSBC, Income Insurance, Manulife, OCBC, Prudential, SCB, UBS and UOB — to pilot workforce transformation and unpack what AI adoption means for specific job roles IBF | The Academy's role-based pathways map directly onto this structure |
Why the Growth Agenda Needs a Decision Layer, Not Just Capital
Vehicles do not create demand
A PCC lowers the cost of building a structure. It does not create the buyers or the investor base to fill one. What converts a cheaper vehicle into a used vehicle is the analytical case for the risk owner — quantified retained risk, credible loss modelling, and a defensible view of what the captive or cell is actually worth against commercial market pricing. That is a decision problem.
The constraint on ILS in Asia is comfort, not incentive
Market commentary is direct on this: the grant scheme is generous, but the harder problem is investor familiarity and the stickiness that brings sponsors back. Capacity remains concentrated, with slightly over half of direct investor capacity in North America and most of the remainder in Europe. Transparent, reproducible risk analytics travel further with an investor base than a subsidy does.
Protection gap closure is an underwriting problem
Underinsured risk stays underinsured because the loss distribution is poorly characterised, the data is thin, and no single carrier can justify the modelling cost against an uncertain market. Shared method with private books is the only structure that makes that modelling economic — each carrier still competes on price and appetite.
PCC and Alternative Risk Transfer
FutureInsurance.ai LLP filed a response to MAS Consultation P013-2026 on the proposed Protected Cell Company framework, and has since responded to MAS follow-up queries on the cell-level re-domiciliation recommendation. This view holds the framework as proposed and the decision infrastructure a cell-based market would need.
Structure as Proposed
The corporate architecture
Eligible uses
Open to MAS-licensed entities carrying out three activity classes:
A self-insurance programme for a firm's own or its affiliates' risks, allowing management of risk and premium rate volatility arising from reliance on the commercial market. Companies could run multiple self-insurance programmes through separate cells, or participate in rent-a-captive arrangements typically sponsored by the insurance management arm of a licensed broker.
Financial instruments that securitise insurance contracts, allowing insurers and reinsurers to transfer specific risks to capital markets investors. The PCC route supports collateralised reinsurance arrangements including sidecars, and more efficient ILS issuance.
Regional and national risk pooling arrangements — the structural class SEADRIF occupies.
Why a Cell Market Needs Shared Decision Infrastructure
The economics of a cell make a private build impossible
The whole point of the PCC is that a cell costs far less to establish than a standalone entity. But a cell inherits the same analytical questions a standalone captive faces — how much risk to retain, at what attachment, against what loss distribution, with what collateral, and whether the economics beat commercial market pricing this renewal. If answering those questions requires a bespoke actuarial and modelling build, the vehicle is cheap and the capability is not, and the barrier has simply moved.
Retention optimisation
What to keep in the cell against what to cede, at what attachment and limit
Loss distribution
Characterising the cell's own loss experience where the data history is short and thin
Commercial comparison
Whether the cell beats the commercial market at this point in the cycle, and by how much
Collateral and capital
Funding adequacy for the cell against its own segregated liability, and stress behaviour
Open Questions a Cell Market Will Have to Answer
| Question | Why it matters operationally |
|---|---|
| Cell-level re-domiciliation | Whether an individual Cell can migrate between PCCs or jurisdictions without unwinding and reconstituting the programme. This is the point on which MAS raised follow-up queries with FutureInsurance.ai LLP after its consultation response |
| Cross-cell recourse in stress | Statutory segregation is clear on paper; the operational question is how a Core-level failure or a Core-level obligation interacts with Cell assets in practice |
| Supervisory reporting granularity | Whether reporting sits at Core level, Cell level or both — and what that means for the reporting burden on a thin cell team |
| Cell governance minimums | What board, actuarial and risk function a Cell must have in its own right versus what it may draw from the Core or the sponsor |
| Cell exit and run-off | How a Cell is closed, run off or transferred, and how policyholder or counterparty protection is preserved through that process |
| Interaction with RBC 2 | How the Notice 133 framework applies to Cells, given that sections 2 to 5 currently exclude captives, marine mutuals and SPRVs |
Protection Gap Observatory
The protection gap is the clearest expression of the industry's unfinished work, and the clearest case for capability that no single carrier will build alone. Every figure on this page is sourced and attributed; where a number is a demonstration figure it is marked as one.
Recorded Events and the Gap They Exposed
| Event | Period | Reported impact | Note |
|---|---|---|---|
| Tropical Storm Penha, Philippines | H1 2026 | 12 fatalities · ~US$30m economic loss | Landed in one of the region's most underinsured markets AON |
| June earthquake, Philippines and Indonesia | H1 2026 | 93 fatalities · ~US$250m economic loss | Cross-border event affecting two low-penetration markets simultaneously AON |
| Indonesia flooding | Jan–Feb 2026 | At least 87 fatalities | Multiple events in a single quarter AON |
| Lao PDR rainfall and flooding | Sep 2026 | US$1.14m paid in five business days | SEADRIF payouts to the Government of Lao PDR and the World Food Programme, triggered by government-reported disaster impact data under the people-affected trigger SEADRIF |
| Asia aggregate | FY2025 | ~US$65bn economic loss | More than 90% uninsured, cited by DPM and MAS Chairman Gan Kim Yong MAS |
Why the Gap Persists — the OECD Reading
Income levels
Across much of ASEAN, insurance penetration remains relatively low, and the OECD attributes this in part to income levels. Premium affordability is a binding constraint before it is a product design constraint.
Financial literacy
Limited financial literacy is cited as a contributing factor. This is a capability problem in the population, and it sits alongside the capability problem in the industry that the Academy addresses.
Distribution
Underdeveloped insurance distribution limits reach even where willingness and affordability exist. Embedded and digital channels are the structural response the region is now testing.
What a Shared Decision Layer Contributes to Gap Closure
| Characterising thin-data perils | Parametric and index-based products need a defensible loss distribution where claims history barely exists. The modelling method is shareable; the resulting book is not |
| Basis risk quantification | Parametric cover fails commercially when basis risk is unquantified and a payout misses a real loss. A shared basis-risk framework is a public good with no competitive value to any single carrier |
| Trigger design libraries | SEADRIF's people-affected impact trigger demonstrates that non-parametric-index triggers can pay in days. Trigger design patterns should be common property |
| Affordability-constrained product design | Designing to a price point rather than pricing a design — a method most carriers have not systematised because the target segment has never been profitable enough to justify the build |
| Distribution economics modelling | Whether an embedded or microinsurance channel can carry its own acquisition cost at the premium level the segment can bear |
Supervisory Data Exchange
A shared industry platform raises an obvious question from both directions: does the regulator get a window into every participant, and does the platform operator get one. The answer to both is no, and the architecture has to make that answer verifiable rather than merely asserted.
- Policyholder and claimant records
- Policy, claims and exposure files
- Pricing models, rating factors and loadings
- Underwriting appetite and declinature rules
- Reserving assumptions and reserve balances
- Reinsurance structures, terms and counterparties
- Distribution economics and commission structures
- Financial results ahead of publication
- Model inventories and validation evidence
- Individual decision records and their outcomes
BOUNDARY
- Frameworks, taxonomies and decision templates
- Control sets mapped to regulatory instruments
- Model architectures and reference implementations
- Validation and fairness testing methodologies
- Hazard model libraries and climate pathway sets
- Regulatory intelligence and obligation tracking
- Agent designs and authority envelope patterns
- Training curricula and certification pathways
- Anonymised aggregate readiness, minimum cell size five
- Published market statistics and regulatory text
What a Supervisor Gains — Without Gaining Anyone's Book
Earlier sight of structural weakness
Sector-level readiness against a common control set, refreshed continuously rather than assembled through a thematic review cycle. A supervisor sees that agentic authority envelopes are thin across the market months before an inspection programme would surface it — and still cannot name a single firm.
Comparability of evidence
When participants build against a common control taxonomy and a common evidence format, submissions become genuinely comparable. Supervisory time currently spent normalising heterogeneous submissions is returned to actual supervision.
Faster sector-wide remediation
When a control gap is identified sector-wide, the remediation pattern is published once to the shared library and every participant can adopt it. The current alternative is fifty parallel remediation projects reaching fifty different standards.
A measurable capability baseline
Capabilities and capacity is a named domain in the proposed AI Guidelines and the hardest for any supervisor to assess. Certification completion by role, aggregated across the sector, is a real signal — and it is generated without any firm disclosing anything about its business.
What This Platform Will Not Do
| No supervisory back door | There is no regulator-facing view of an individual participant's instance, and none can be built without re-architecting the platform. Aggregate views are computed from voluntary contributions with a minimum cell size, and cells below it are suppressed rather than estimated |
| No training on participant data | No participant's data is used to train, fine-tune or evaluate any shared model. A model improved by one participant's book would carry that book's signal to competitors — the platform is architected so this is impossible, not discouraged |
| No cross-participant inference | The platform does not compute anything that requires reading more than one participant's data together. There is no cross-book analytics surface, so there is nothing to leak through one |
| No operator access to participant instances | Platform operations do not require read access to participant decision data. Support is delivered against configuration and telemetry, not content |
| No silent scope expansion | Any change to what leaves a participant instance is a contractual change requiring affirmative consent, not a terms-of-service update |
| No lock-in through data gravity | Frameworks, control sets and a participant's own decision history export in open formats at any time. The platform earns renewal on value, not on the cost of leaving |
ASEAN Executive View
Ten member states, ten insurance supervisors, ten regulatory regimes at very different stages of development, and one shared exposure to climate, demographic and digital risk. The regional case for a common decision layer is not efficiency — it is that several of these markets cannot build a modern decision capability alone at any price, and the ones that can are building ten incompatible versions of the same thing.
Three Forces Named by the Region's Own Regulators
Climate risk FRONT LINE
Climate impacts are described as already becoming a major source of financial instability across ASEAN, with the region remaining highly vulnerable to floods, storms, droughts and other extreme weather. The ASEAN Taxonomy for Sustainable Finance is cited as the instrument that should guide future investment and underwriting to improve long-term resilience.
Hazard modelling, scenario pathways and accumulation methods are the highest-fixed-cost, lowest-competitive-value capability in the region. They should be built once regionally, not eleven times nationally.
Digital transformation ACCELERATING
Digital transformation emerged as a critical priority at the 2025 regional meeting, with insurers across the region rapidly adopting advanced analytics and AI for underwriting, claims and customer engagement. Adoption is running ahead of supervisory frameworks in several markets.
Where AI adoption outpaces AI governance, the gap becomes a regional risk rather than a national one. Shared governance architecture closes it faster than ten separate regimes can.
Ageing populations STRUCTURAL
Ageing trends were highlighted as reinforcing the need to expand insurance penetration, particularly in markets where coverage is still developing. Demographic transition is arriving in several ASEAN markets before insurance penetration has matured.
Longevity, morbidity and health-cost modelling in markets with short data histories — the archetypal case for shared method with private books.
The Regional Asymmetry
Two ASEANs, one insurance market
Singapore's insurance industry reached roughly S$78 billion in total premiums at end-2024 with a workforce of around 19,000, and its reinsurance premiums accounted for about 21% of Asia's reinsurance market in 2023. Cambodia's penetration stands at 1.13% with density of US$20.65 per person. These two markets sit in the same regional framework, attend the same annual meeting, and face the same climate exposure. They cannot plausibly build the same decision capability by the same route.
What Sits Above the National Level Already
Market Data Explorer
Insurance data across ASEAN is published unevenly, on different bases, at different frequencies, in different currencies. This explorer holds what is verifiable and declares what is not. A cell marked as a data gap is a data gap — it is not estimated, interpolated or filled from a regional average.
Penetration, Density and Premium Base
| Market | Penetration | Density | Premium base | Source and note |
|---|---|---|---|---|
| Singapore | Data gap GAP | Data gap GAP | S$78bn total, end-2024 MAS | Life, general and reinsurance combined; grew >8% p.a. 2019–2024. Life new business S$6.53bn in 2025, up 11.3% LIA. Reinsurance premiums S$27.6bn in 2023, ~21% of Asia's reinsurance market MAS |
| Philippines | 1.79% of GDP IC | PHP 4,384.56 IC | PHP 502.64bn, 2025 IC | Total premiums up 14.1% from PHP 440.53bn in 2024. Life 80.77%, non-life 16.41%, mutual benefit associations 3.37%. Total benefits PHP 121.88bn. Penetration remains below the regulator's 2% target |
| Indonesia | 2.61%, 2024 SOURCED | Data gap GAP | Data gap GAP | Penetration down from 2023 and the lowest since 2019. Takaful growth is a noted structural driver given the world's largest Muslim population |
| Cambodia | 1.13% SOURCED | US$20.65 per capita SOURCED | Data gap GAP | Stated by Cambodia's Ministry of Economy and Finance at the 28th AIRM, November 2025. National priorities: strengthening disaster-risk insurance, widening digital distribution, improving public understanding |
| Malaysia | Data gap GAP | Data gap GAP | Data gap GAP | Reported as experiencing significant growth driven by rising awareness and demand for comprehensive financial protection. Confirmed advanced consideration of joining SEADRIF by May 2026 WORLD BANK |
| Vietnam | Data gap GAP | Data gap GAP | Data gap GAP | Reported double-digit life premium growth buoyed by bancassurance partnerships, expected to continue into 2026 SOURCED |
| Thailand | Data gap GAP | Data gap GAP | Data gap GAP | Strong bancassurance growth reported, with banks integrating insurance into digital banking platforms. Signalled intent to explore SEADRIF membership WORLD BANK |
| Brunei Darussalam | Data gap GAP | Data gap GAP | Data gap GAP | Signalled intent to explore SEADRIF membership WORLD BANK |
| Lao PDR | Data gap GAP | Data gap GAP | Data gap GAP | Among the first countries to adopt a sovereign disaster risk insurance policy with a people-affected impact trigger, May 2025. Purchased its first disaster risk insurance in 2021, a three-year hybrid flood product WORLD BANK |
| Myanmar | Data gap GAP | Data gap GAP | Data gap GAP | Myanmar Insurance Association recognised as the 15th member of the ASEAN Insurance Council in April 2018 SOURCED |
Regional Aggregates and Trajectory
Life and health, Southeast Asia
Motor, Southeast Asia
Emerging Asia baseline
Country Benchmark Board
Ten market cards. Where a verifiable figure exists it is shown with its source; where it does not, the card says so. The purpose is not to rank the markets — it is to make visible how uneven the regional evidence base actually is, because that unevenness is itself the constraint on regional decision-making.
Protection Gap and Natural Catastrophe
ASEAN countries, home to more than 650 million people, are highly prone to disaster and climate shocks — floods, tropical storms, droughts, earthquakes and tsunamis have left severe physical, economic and human impacts across the region. The insurance response to that exposure is the smallest it is anywhere in the developed or emerging world.
The Asymmetry, Stated Precisely
Asia carries the losses and not the cover
In full-year 2025, Asia accounted for approximately 30% of global economic catastrophe losses while representing only around 5% of insured losses. That single ratio is the region's insurance problem in one line: the risk is here, the capital that absorbs it is not.
Swiss Re Institute attributes the region's large protection gap to rapid urbanisation and economic development outpacing insurance penetration. Exposure is being created faster than cover is being written — so growth in premiums can coexist with a widening gap, and does.
Exposure that has never been insured has never been modelled to underwriting standard. There is no claims history because there were no claims — only losses. Characterising these perils requires methods that do not depend on a loss history, which is exactly the capability no single ASEAN carrier can justify building alone.
Recorded Regional Events
| Event | Period | Reported impact | Insurance response |
|---|---|---|---|
| Tropical Storm Penha, Philippines | H1 2026 | 12 fatalities · ~US$30m | Landed in a market with ~98% catastrophe protection gap AON |
| Earthquake, Philippines and Indonesia | June 2026 | 93 fatalities · ~US$250m | Cross-border event, two low-penetration markets simultaneously AON |
| Flooding, Indonesia | Jan–Feb 2026 | At least 87 fatalities | Multiple events in a single quarter AON |
| Heavy rainfall and flooding, Lao PDR | Sept 2026 | US$1.14m paid | SEADRIF paid the Government of Lao PDR and the World Food Programme within five business days, triggered by government-reported disaster impact data under the people-affected trigger SEADRIF |
Structural Causes and the Instruments That Address Them
| Cause | Instrument in play | What is still missing |
|---|---|---|
| Income levels and affordability | Microinsurance, embedded distribution, premium subsidy through grants such as the Global Shield Financing Facility | A systematic method for designing to a price point rather than pricing a design |
| Limited financial literacy | National awareness programmes; public understanding named as a Cambodian priority | Product structures simple enough that literacy is not a precondition for value |
| Underdeveloped distribution | Digital and embedded channels through e-commerce and ride-hailing platforms across Southeast Asia | Distribution economics that work at the premium sizes the segment can bear |
| Sovereign fiscal exposure | SEADRIF sovereign policies; ADRFI; SEADRIF-SAFE for public infrastructure | Scale — regional uptake occurred more slowly than anticipated during the initial project phase WORLD BANK |
| Agricultural exposure | SEADRIF-RAISE with FAO, consulted across six countries in February 2026 | Yield and index modelling for smallholder agriculture at affordable basis risk |
| Thin loss data | ANDREWS natural disaster research work sharing; AHA Centre disaster impact reporting | A common analytical layer that turns shared impact data into underwriting-grade loss characterisation |
Correlated Exposure
Diversification is the assumption underneath every regional portfolio and every regional risk pool. It only holds where the underlying exposures are genuinely independent. Across ASEAN, several are not — and the correlations that matter most are the ones that are structural rather than statistical.
Structural Correlation Channels
Climate and hydrology HIGH
Regional-scale climate drivers do not respect national boundaries. Monsoon behaviour, El Niño–Southern Oscillation phases and Indian Ocean Dipole conditions modulate flood, drought and cyclone risk across multiple member states in the same season.
Agricultural yield cover, property catastrophe, business interruption, sovereign disaster policies, parametric drought triggers
Seismic and volcanic MEDIUM-HIGH
The June 2026 earthquake affecting both the Philippines and Indonesia is the illustration: a single tectonic event producing simultaneous loss in two member states, both with low penetration and both plausible members of the same pool.
Property, marine, engineering, sovereign infrastructure cover, contingent business interruption
Supply chain and trade MEDIUM
Deep intra-ASEAN supply chain integration means a physical loss in one market propagates as contingent business interruption in several others. The insured event and the insured loss occur in different jurisdictions.
Contingent BI, marine cargo, trade credit, political risk, cyber where a shared logistics platform is involved
Technology and vendor concentration MEDIUM · RISING
The least discussed and the fastest growing. Insurers across the region are rapidly adopting analytics and AI for underwriting, claims and customer engagement — and drawing on a small number of the same vendors. A single model failure or vendor outage becomes a regional event.
Operational risk, cyber, mispricing across multiple carriers using the same model, claims handling capacity
Correlation Matrix — Peril Channel by Market Cluster
What the Region Cannot Currently See
| Regional accumulation by peril | No participant, regulator or pool can currently see how much of a given peril the region's insurers collectively hold, because every carrier models it differently and no comparable aggregate exists |
| Cross-border contingent exposure | Where an insured loss in one market is triggered by a physical event in another, neither supervisor sees the full chain |
| Reinsurance concentration | Whether the region's cedants are concentrated on the same reinsurance counterparties, which converts a counterparty failure into a regional solvency event |
| Shared model concentration | How many carriers across how many markets are pricing the same peril off the same vendor model, and what a common error in that model would cost the region |
Digital and Insurtech Landscape
Digital transformation was named a critical priority at the 2025 regional regulators' meeting, with insurers across ASEAN rapidly adopting advanced analytics and AI for underwriting, claims and customer engagement. In several markets the adoption curve is ahead of the supervisory framework — which makes shared governance architecture a regional risk-reduction measure rather than a convenience.
Where the Market Is Actually Moving
| Shift | What is reported | Decision-layer consequence |
|---|---|---|
| Embedded distribution | Embedded insurance distribution through e-commerce and ride-hailing apps is reshaping traditional agent networks across Southeast Asia, forcing carriers to rebuild underwriting infrastructure around API-first policy issuance and real-time data ingestion SOURCED | Underwriting decisions move from days to milliseconds — the governance model has to move with them or it is decorative |
| Continuous underwriting | A reported 2026 shift from static annual underwriting to continuous underwriting, with risk assessed in real time from telematics, IoT and streaming data SOURCED | A model that is re-scoring continuously needs continuous monitoring, not annual validation |
| Agentic workflows | Cytora launched Autopilot in March 2026, an agentic capability running end-to-end risk workflows autonomously; Zurich Insurance completed a 90-day rollout across five countries with plans to expand to more than 20 markets SOURCED | Agentic authority envelopes stop being theoretical — this is production autonomy at multi-market scale |
| Underwriting cycle compression | Underwriting timelines reported collapsing from three days to three minutes on standard SME risks SOURCED | The human review step that carried the governance burden has been removed; the control must be designed into the machine path |
| Parametric agriculture | Satellite-indexed crop insurance reported as underpenetrated, with automated payout triggers being piloted SOURCED | Trigger design and basis risk become the core underwriting decision rather than a technical annexe |
| Cross-border licensing | ASEAN member states reported to be negotiating mutual recognition frameworks for insurance licences, potentially enabling single-platform distribution across multiple jurisdictions SOURCED | One product, ten supervisors — a common control and evidence format becomes an operating necessity |
| Fraud on direct channels | Synthetic identity schemes and staged accidents reported proliferating on direct-to-consumer platforms SOURCED | Fraud detection is a shared-method problem — every carrier faces the same rings, and none benefits from a private taxonomy |
The Integration Constraint Nobody Prices In
The binding constraint is absorption, not model quality
Published analysis puts the problem directly: for most traditional insurers the real constraint is not model quality but what IT can absorb, with BCG cited for the finding that 70% of insurers fail to execute on innovation because of IT limitations. The policy core works, the cost of replacing it is prohibitive, and any proposal that starts with a migration dies in committee.
Regulatory Divergence Matrix
The AIRM continues to exchange updates on improving observance of the International Association of Insurance Supervisors' Insurance Core Principles. That common reference point is what makes convergence conceivable. Divergence in how it is implemented is what makes a common compliance layer valuable.
Common Reference Architecture
| Layer | Status | What it establishes |
|---|---|---|
| IAIS Insurance Core Principles | COMMON REFERENCE | The AIRM continues to exchange updates on improving observance of the ICPs — this is the shared standard against which ten supervisors measure themselves ASEAN |
| ASEAN Insurance Regulators' Meeting | EST. 1998 | Platform to strengthen insurance cooperation in the development of regulatory and supervisory frameworks, and research and capacity building through AITRI ASEAN |
| ASEAN Taxonomy for Sustainable Finance | IN OPERATION | Cited at the 28th AIRM as the instrument that must guide future investment and underwriting to improve long-term resilience SOURCED |
| ASEAN Council of Bureaux | OPERATING | The regional mechanism addressing motor and personal injury matters arising from vehicles crossing borders between ASEAN countries SOURCED |
| Mutual recognition of licences | REPORTED IN NEGOTIATION | ASEAN member states reported to be negotiating mutual recognition frameworks for insurance licences, potentially enabling single-platform distribution across jurisdictions SOURCED |
| Regional AI supervision | NO COMMON INSTRUMENT | No ASEAN-level AI supervisory instrument identified in this build. Singapore has consulted on Guidelines on AI Risk Management; Bank Negara Malaysia issued a discussion paper on AI in the Malaysian financial sector in 2025 SOURCED |
Supervisory Bodies by Market
| Market | Supervisory authority | AI instrument identified | Note |
|---|---|---|---|
| Singapore | Monetary Authority of Singapore | CONSULTED | Guidelines on AI Risk Management consulted Nov 2025 – Jan 2026, applying to all FIs including agentic AI, pending finalisation MAS |
| Malaysia | Bank Negara Malaysia | DISCUSSION PAPER | Discussion Paper on Artificial Intelligence in the Malaysian Financial Sector, 2025 SOURCED |
| Philippines | Insurance Commission | NOT VERIFIED | Publishes penetration, density and premium statistics; operates a stated 2% penetration target IC |
| Indonesia | Otoritas Jasa Keuangan (OJK) | NOT VERIFIED | Requires jurisdiction verification |
| Thailand | Office of Insurance Commission | NOT VERIFIED | Requires jurisdiction verification |
| Vietnam | Ministry of Finance | NOT VERIFIED | Deputy Minister of Finance addressed the 26th AIRM in 2023 SOURCED |
| Cambodia | Insurance Regulator of Cambodia · Non-Bank Financial Services Authority | NOT VERIFIED | Hosted the 28th AIRM and 51st AIC, Siem Reap, November 2025 SOURCED |
| Lao PDR | Ministry of Finance | NOT VERIFIED | Sovereign disaster risk insurance policyholder with SEADRIF WORLD BANK |
| Brunei Darussalam | Brunei Darussalam Central Bank | NOT VERIFIED | Requires jurisdiction verification |
| Myanmar | Financial Regulatory Department · Insurance Business Regulatory Board | NOT VERIFIED | Hosted the 22nd AIRM and 45th AIC, Nay Pyi Taw, 2019 SOURCED |
The Multi-Jurisdiction Compliance Case
One control set, ten evidence formats
A regional insurer operating in six ASEAN markets currently maintains six compliance programmes that overlap heavily in substance and not at all in format. The IAIS Insurance Core Principles mean the underlying expectations are largely convergent; the implementation is where the cost sits. A common control taxonomy mapped once to each supervisor's instrument set lets one piece of evidence satisfy several regimes — and where a market has no instrument yet, the common set is a defensible default rather than an absence.
Control taxonomy, evidence schema, test methodology, model documentation standard
Instrument crosswalk, local reporting format, filing calendar, language and disclosure requirements
The actual evidence, the actual model inventory, the actual decisions — never shared, never pooled
Regional Institutional Map
ASEAN's insurance cooperation architecture is nearly fifty years old and considerably more developed than is generally understood. Any regional decision layer has to complement this architecture rather than propose an alternative to it — which means knowing precisely what each body already does.
Regional Bodies
| Body | Established | Mandate | Where a decision layer complements it |
|---|---|---|---|
| ASEAN Insurance Council (AIC) | 4 April 1978, Jakarta | The main regional platform for insurance leaders, professionals and practitioners to network and share knowledge and expertise across areas of insurance business. An accredited entity of ASEAN comprising 15 members from insurance associations across the ten member countries. Works closely with ASEAN insurance regulators and other stakeholders for the development of the industry SOURCED | Turning shared knowledge into shared operating tools — the AIC convenes the expertise, a decision layer would carry the method |
| ASEAN Insurance Regulators' Meeting (AIRM) | 1998 | Platform to strengthen insurance cooperation in the development of regulatory and supervisory frameworks and in research and capacity building through AITRI. Continues to exchange updates on improving observance of the IAIS Insurance Core Principles ASEAN | Aggregate, anonymised readiness against a common control set — visibility without any firm-level disclosure |
| ASEAN Insurance Training and Research Institute (AITRI) | Under AIRM | Research and capacity building for the region's insurance sector ASEAN | The natural regional counterpart to the AI for Insurance Academy — curriculum and certification, delivered at regional scale |
| ASEAN Council of Bureaux (COB) | Working group from 2015 | Addresses automotive and personal injury matters relating to border crossing among ASEAN countries — the cross-border compulsory motor mechanism SOURCED | Cross-border claims decisioning and fraud detection across jurisdictions is a shared-method problem by construction |
| ASEAN Insurance Education Committee (AIEC) | Under AIRM/AIC | Regional insurance education coordination; a working group whose outcomes are reviewed at the annual AIC meeting SOURCED | Certification portability across the region — a qualification recognised in one market recognised in ten |
| ASEAN Natural Disasters Research Works Sharing (ANDREWS) | Under AIRM/AIC | Regional sharing of natural disaster research work ASEAN | Research-to-underwriting translation: turning shared hazard research into usable loss characterisation for pricing and pools |
| SEADRIF Insurance Company | Licensed Oct 2019 | The first regional catastrophe risk facility established in Asia by ASEAN member states. Incorporated and domiciled in Singapore, licensed as a general insurer. Established and owned by ASEAN+3 countries with financial support from the Governments of Japan and Singapore and technical support from the World Bank MAS · SEADRIF | Trigger design, impact data pipelines and payout decision logic as reusable shared components |
| ADRFI Programme | Under AIRM agenda | The ASEAN Disaster Risk Financing and Insurance Programme, considered at AIRM alongside the ASEAN Taxonomy for Sustainable Finance and the Framework for Circular Economy SOURCED | Risk financing strategy analytics at sovereign level — layered retention, pool and market transfer decisions |
| AHA Centre | ASEAN body | The ASEAN Coordinating Centre for Humanitarian Assistance on disaster management. Signed a Memorandum of Intent with SEADRIF Insurance Company in Bali on 9 March 2026 on disaster impact reporting and scalable risk financing solutions, considering a regional insurance mechanism for pre-arranged financing of rapid humanitarian response SEADRIF | The disaster impact data spine that parametric and impact-triggered products depend on |
| Global-Asia Insurance Partnership | Singapore | A tripartite partnership between the global insurance industry, regulators and policymakers, and academia, established as a centre of excellence in insurance and risk management with a focus on Asia, initially on pandemic and climate risk MAS | Methodology validation and the research-to-practice pathway for shared analytical components |
Annual Cycle
A joint plenary meeting between the AIRM and the industry is held annually, with members from the AIC, the Council of Bureaux, the AIEC and ANDREWS. The plenary serves as the platform to exchange views on the progress and implementation of the various insurance initiatives. ASEAN
Disaster Risk Financing
SEADRIF is the clearest evidence in the region that pre-arranged, decision-driven risk financing works. It is also the clearest evidence of how slowly it scales without shared capability. Both lessons matter for what a regional decision layer should try to be.
What SEADRIF Has Actually Done
The Honest Lesson from the Implementation Review
Why regional uptake was slower than planned
The World Bank implementation completion report is direct about what did not work. The project assumed a pace of multi-country uptake that proved unrealistic for a newly established regional insurer operating within a multi-tier governance structure, and it did not include a formal onboarding plan, a product-development roadmap, or a sustainability strategy for SEADRIF beyond initial capitalisation. Those gaps limited SEADRIF's ability to attract additional members during implementation, and highlighted the importance of integrating demand diagnostics, regulatory feasibility analysis and readiness work.
Reusable Components a Regional Decision Layer Would Carry
Trigger design library
Parametric index, modelled loss, and impact-based triggers including the people-affected pattern, each with its documented failure modes, basis risk profile and data dependency.
Impact data pipeline
Government-reported disaster impact data, validated and channelled into a payout decision. The SEADRIF–AHA Centre memorandum addresses exactly this — consistency and quality of disaster impact reporting.
Layered risk financing model
Retention, contingency, pool and market transfer as a layered stack, sized against a sovereign's fiscal capacity and its actual loss distribution.
Demand diagnostic
The function named as missing in the implementation review — establishing whether a country has the exposure, the fiscal case and the institutional readiness before onboarding begins.
Regulatory feasibility analysis
Whether the product can be issued, held and paid under the target jurisdiction's own insurance and public finance law.
Payout governance
How a payout is verified, released, deployed and accounted for — with the safeguards that ensure funds move swiftly, transparently and responsibly.
Regional Playbooks
A playbook is a decision sequence agreed before it is needed. Each one below is a pattern that recurs across ASEAN markets, where every participant currently improvises its own version and none is better for the improvisation.
PB-01 · Multi-market catastrophe event HIGH FREQUENCY
A single physical event produces loss in two or more ASEAN markets simultaneously — the June 2026 Philippines and Indonesia earthquake is the reference case.
PB-02 · Cross-border motor claim RECURRING
A vehicle insured in one ASEAN market causes loss in another. The ASEAN Council of Bureaux exists precisely for this class of matter.
PB-03 · Sovereign risk financing design STRATEGIC
A member state assesses whether and how to transfer disaster risk — the decision sequence the SEADRIF implementation review found was missing as a formal capability.
PB-04 · Regional model failure EMERGING
A model used by multiple carriers across multiple markets is found to be materially wrong. Adoption of common vendor models across the region makes this a question of when, not whether.
Data Register and Provenance
A platform that asks an industry to trust it with method has to be auditable on evidence. This register lists every substantive external figure used across the three views, its source, its period, and its data class. Where a figure could not be verified to an attributable source it is not on the platform at all — it is listed here as a declared gap.
Source Register
| Source | Class | Period | What is drawn from it |
|---|---|---|---|
| MAS media release — AI Risk Management Guidelines consultation | PRIMARY | 13 Nov 2025 | Scope, applicability to all FIs, proportionality principle, consultation close date of 31 January 2026, and the deputy managing director's statement on proportionate risk-based guidelines |
| MAS Consultation Paper P017-2025 | PRIMARY | Nov 2025 | Definition of AI scope covering machine learning, deep learning, reinforcement learning, generative AI and AI agents; the four expectation domains |
| MAS written parliamentary reply | PRIMARY | 5 Aug 2026 | Confirmation by the MAS Chairman that the proposed Guidelines apply to all AI use cases including agentic AI and will be finalised soon |
| MAS media release — PCC framework consultation | PRIMARY | 7 Jul 2026 | PCC structure, eligible activities, consultation dates, the ~US$65bn Asia 2025 nat cat loss figure and the >90% uninsured share |
| MAS Consultation Paper P013-2026 | PRIMARY | 7 Jul – 7 Aug 2026 | Core and Cell structure, statutory segregation, captive, ILS and sovereign risk pool use cases, rent-a-captive model definition |
| MAS Notice 133 and amendments | PRIMARY | 2022–2026 | RBC 2 structure, applicability by insurer class, AT1 and Tier 2 recognition criteria effective 1 January 2026, equity counter-cyclical adjustment and structured product treatment |
| MAS Guidelines on Risk Management Practices — Insurance Core Activities | PRIMARY | Current | Environmental risk management guidelines and transition planning addendum, outward reinsurance, outsourcing, AML/CFT, product development and pricing, IAC outcomes |
| MAS Compliance Toolkit for Direct Insurers and Reinsurers | PRIMARY | Current | Obligation categories: licensing and control, insurance funds, RBC, business continuity, POPS, reinsurance and technology risk management |
| MAS keynote, 21st Singapore International Reinsurance Conference | PRIMARY | 3 Nov 2025 | S$78bn total premiums end-2024, >8% average annual growth 2019–2024, ~19,000 workforce, refreshed ILS Grant Scheme extension to non-APAC risks and renewals |
| MAS Insurance-Linked Securities Grant Scheme page | PRIMARY | Jan 2026 – Dec 2028 | Grant percentages and caps by instrument type, eligible risk classes, eligible applicants |
| MAS Insurance and Risk Financing Initiatives page | PRIMARY | Current | Global-Asia Insurance Partnership, the NTU IRFRC-led natural catastrophe data analytics partnership, SEADRIF domicile and licensing |
| MAS regulation and licensing pages | PRIMARY | Current | Insurance regulation overview, FinTech Regulatory Sandbox, tax incentives and grant schemes |
| IBF — GenAI Jobs Transformation Map | PRIMARY | Current | MAS, IBF and WSG partnership with eleven named financial institutions including Income Insurance, Manulife and Prudential; Skills Framework for Financial Services; Young Talent Programme for AI in Finance |
| LIA Singapore industry results | PRIMARY | FY2025 | S$6.53bn total weighted new business premiums, up 11.3%; 1H2025 S$2.99bn up 7.7%; ILP share 43% of new sales |
| Philippine Insurance Commission data via published reporting | SECONDARY | FY2025 | PHP 502.64bn total premiums up 14.1%; life 80.77%, non-life 16.41%, MBA 3.37%; density PHP 4,384.56; penetration 1.79% below the 2% target; benefits PHP 121.88bn |
| 28th AIRM · 51st AIC reporting | SECONDARY | 27 Nov 2025 | Meeting theme and agenda; Cambodia penetration 1.13% and density US$20.65; climate, digital and ageing priorities; ASEAN Taxonomy reference |
| ASEAN main portal — sectoral bodies | PRIMARY | Current | AIRM establishment 1998, AITRI mandate, joint plenary composition with AIC, COB, AIEC and ANDREWS, IAIS Insurance Core Principles observance |
| ASEAN Insurance Council materials | PRIMARY | Current | Establishment 4 April 1978 in Jakarta; 15 members across ten member countries; accredited ASEAN entity status; 2026 events calendar |
| SEADRIF announcements | PRIMARY | 2026 | AHA Centre memorandum of intent 9 March 2026; SEADRIF-RAISE Manila workshop 12 February 2026; Lao PDR US$1.14m payout in five business days, September 2026; ownership and support structure |
| World Bank — SEADRIF feature and implementation completion report | PRIMARY | 2026 | Lao PDR 2021 first policy and May 2025 impact-trigger policy; US$3.6m premium grant; SEADRIF 2.0 Philippines policies from Q1 2026; Malaysia, Thailand, Brunei and Timor-Leste membership signals; the finding on onboarding, roadmap and sustainability gaps |
| Swiss Re Institute analysis via published reporting | SECONDARY | 2025–2026 | US$424bn global nat cat protection gap in 2025 from US$395bn; Emerging Asia catastrophe resilience score of 5%; Asia ~30% of economic losses against ~5% of insured losses in FY2025 |
| Aon catastrophe reporting via published reporting | SECONDARY | H1 2026 | Tropical Storm Penha; June 2026 Philippines–Indonesia earthquake; Indonesia January–February 2026 flooding |
| GlobalData analysis via published reporting | SECONDARY | 2026 | Philippines catastrophe protection gap of approximately 98% against a global average of 58% |
| Peak Re insight | SECONDARY | 2012–2022 | Emerging Asia penetration 3.0% to 3.6%; China per capita US$489 against US$86 elsewhere in emerging Asia; ~90% uninsured share of Asian nat cat losses |
| Southeast Asia life and health market analysis | SECONDARY | 2025–2030 | Life GWP US$104.8bn 2025 to US$118bn 2030; health GWP US$12.3bn to US$13.8bn; ~2–3% annual growth; uneven country trajectories |
| Southeast Asia motor market analysis | SECONDARY | 2025–2032 | US$13.19bn 2025, US$15.11bn 2026, US$19.91bn 2032, 4.71% CAGR; ~57% third-party liability share |
| APAC insurtech market analyses | SECONDARY | 2026–2035 | Two divergent estimates, both reported: US$18.7bn 2026 at 16.4% CAGR, and US$28.08bn 2026 at 19.39% CAGR |
| Insurance AI vendor landscape reporting | SECONDARY | 2026 | Category structure across intake, workbench, pricing and claims; Cytora Autopilot launch and Zurich 90-day five-country rollout; Federato funding; Akur8 scale; straight-through processing and underwriting cycle figures; the BCG-cited 70% IT execution constraint |
| Chambers Insurance & Reinsurance 2026 — Singapore | SECONDARY | 2026 | Singapore's principles-based AI approach; FEAT and Veritas lineage; M&A and portfolio transfer approval requirements; insurtech hub positioning |
| CapitalMarkets.SG licence register cross-reference | SECONDARY | 18 Jun 2026 | 337 insurance companies and brokers holding active MAS licences, cross-referenced against the MAS Financial Institutions Directory |
Declared Gaps
| Gap | Why it is declared rather than estimated |
|---|---|
| Penetration for six ASEAN markets | Malaysia, Vietnam, Thailand, Brunei, Lao PDR and Myanmar — no current attributable primary figure established in this build. Regional averages would produce a number that looks precise and means nothing |
| Density for eight ASEAN markets | Only Cambodia and the Philippines carry an attributable current density figure. Deriving density from a premium estimate and a population estimate compounds two uncertainties into one confident-looking output |
| Singapore penetration and density | Total premiums are published; the penetration ratio requires a GDP denominator on a stated basis and period, which was not established here. The premium figure is shown; the ratio is not |
| AI supervisory instruments in eight markets | Only Singapore and Malaysia had an AI-specific financial sector instrument identified. "Not verified" means not established in this build, not confirmed absent |
| Sector readiness percentages | Every readiness figure on this platform is marked illustrative. No survey has been conducted. The mechanism is real; the numbers demonstrate the mechanism |
| Correlation classifications | Reasoned from published exposure characteristics, not from a calibrated correlation model. Marked illustrative and unsuitable for pricing, capital or pool design |
InsurEDIC³ Platform Architecture
Eight layers. The AI Control Tower sits at the top as the orchestrating brain — it does not merely observe the layers beneath it, it grants and withdraws their authority at runtime. Layer 1 is the foundation at the base: the participant's own systems, which this platform reads from and never replaces.
The Eight Layers
How the Control Tower Governs an Agent
Agent Orchestration Map
Agents are organised into squads by decision domain. Every agent below runs inside an authority envelope set by the participant, and every one of them can be narrowed, suspended or withdrawn at runtime by the Control Tower. No agent binds risk, issues a policy, moves money or settles a claim — those remain regulated acts performed in the participant's own systems by the participant.
Autonomy Ladder
Squad 1 · Regulatory and Compliance
Squad 2 · Underwriting and Pricing
Squad 3 · Claims
Squad 4 · Capital, Finance and Risk
Squad 5 · Climate and Catastrophe
Squad 6 · Governance and Capability
Why seven agents are permanently restricted to recommend-only
Capital position, pricing fairness, claims friction equity, fraud referral, boundary audit, precedent curation and model failure response are all domains where an autonomous action would either affect a customer adversely without human judgement, or where the agent would be marking its own homework. These are not configuration defaults that a participant can raise — they are architectural ceilings. An agent that could autonomously clear its own boundary audit or autonomously decline a claimant is not a governance improvement over a human doing it badly; it is the same failure at machine speed with a thinner accountability trail.
Frameworks and Methods Library
The core proposition of InsurEDIC³ is that the insurance industry keeps paying, separately and repeatedly, for the same intellectual assets. This library is the shared alternative: frameworks, tools, processes and methodologies that any participant can take, apply to its own book, and adapt — without building them, and without disclosing anything.
Decision Frameworks
| Asset | Applies to | What it gives a participant |
|---|---|---|
| Decision Object Model | All | A standard structure for what a decision is — owner, trigger, options, rationale, materiality, decision-maker, outcome — so decisions become auditable objects rather than events that happened |
| Signal-to-Decision Spine | All | The path from raw signal through insight to decision to action to outcome, with the handoff points named and owned |
| Decision Materiality Scale | All | A common scale for classifying decision materiality, which is the input proportionality depends on and which most firms currently invent ad hoc |
| Autonomy Ladder | All | The four-level authority model — observe, recommend, act with approval, act within bounds — with the criteria for moving between levels |
| Retention Optimisation Framework | Captives, PCC cells | How much risk to retain at what attachment, against what loss distribution, compared to commercial market pricing at this point in the cycle |
| Layered Risk Financing Model | Sovereign, corporate | Retention, contingency, pool and market transfer as a layered stack sized against absorptive capacity |
| Demand Diagnostic | Risk pools, new markets | Whether exposure, fiscal case and institutional readiness support a transfer decision — the capability the SEADRIF implementation review identified as missing |
| Affordability-Constrained Product Design | Protection gap | Designing to a price point the segment can bear rather than pricing a design it cannot |
| Multi-Jurisdiction Control Crosswalk | Regional operators | One control set mapped to several supervisors' instruments, so one piece of evidence satisfies several regimes |
Governance Instruments
| Asset | Maps to | What it gives a participant |
|---|---|---|
| AI Inventory Schema | Proposed AIRM Guidelines | A structure for the AI inventory that satisfies the expectation, including third-party and embedded models most inventories miss |
| AI Life-Cycle Control Set | Proposed AIRM Guidelines | Controls across development, deployment, monitoring and retirement, each with the evidence artefact it must produce |
| Three-Lines-of-Defence AI Mapping | Proposed AIRM Guidelines | Which line owns which AI control, and where the boundaries sit for a model built by one line and used by another |
| Board AI Reporting Pack | Proposed AIRM Guidelines · IAC | What the board actually needs to see to discharge oversight, at a length a board will read |
| Third-Party AI Evidence Standard | Third-party risk · AIRM | The artefacts a participant must obtain from any AI vendor, expressed as contractual requirements |
| Model Card Template | AIRM · FEAT | Intended use, limitations, training provenance, performance, fairness results and known failure modes in one document |
| Decision Ledger Schema | All supervisory instruments | What must be recorded about a decision for it to be reconstructable years later under examination |
| Obligation Register Structure | All supervisory instruments | How to hold the obligation set so that a regulatory change updates the affected controls rather than triggering a full re-mapping |
Analytical Methods
| Asset | Domain | What it gives a participant |
|---|---|---|
| Fairness Testing Battery | Conduct | Proxy scan, outcome parity, friction asymmetry, explanation adequacy and elasticity separation — the method is shared, the results never are |
| Model Validation Methodology | AI governance | Independent validation protocol scaled by materiality class, addressing the capacity constraint rather than restating the policy |
| Drift Detection Configuration | AI governance | What to monitor, at what threshold, with what automatic authority consequence on breach |
| GenAI Risk Control Set | AI governance | Hallucination, prompt injection and data leakage controls, following the Project MindForge focus areas |
| Hazard Model Library | Climate · cat | Documented, transparent hazard models and vulnerability functions — the highest fixed cost and lowest competitive value capability in the industry |
| Climate Pathway Sets | Climate | Scenario pathways for climate scenario analysis across underwriting book and investment portfolio |
| Basis Risk Quantification | Parametric | How to quantify the gap between trigger and actual loss, which is the failure mode that kills parametric products commercially |
| Thin-Data Loss Characterisation | Protection gap | Characterising a loss distribution where there is no claims history because there was never any cover |
| Fraud Pattern Taxonomy | Claims | Shared typology of fraud patterns including synthetic identity and staged accident schemes, with detection logic — no participant's claims data attached |
Operating Playbooks
Industry Learning Exchange
Industry players should be able to learn from each other. The obstacle has never been willingness — it is that the obvious mechanism, pooling data, is commercially impossible and legally fraught. The exchange is built on a narrower and more durable premise: share what went wrong and what worked, never what it was worth or to whom.
Four Exchange Mechanisms
1 · Anonymised pattern contribution
A participant contributes the shape of a decision that went well or badly — the trigger, the option set considered, what was missed, what the corrective was. No amounts, no counterparties, no dates precise enough to identify the event, no lines of business specific enough to identify the contributor.
"A rate adequacy signal was visible eleven months before the segment turned unprofitable, but sat in a report nobody owned. The corrective was to make rate drift an owned decision object with an escalation threshold." — no segment, no numbers, no firm.
2 · Method peer review
Participants review and improve shared methods, the way an actuarial profession improves a standard. A better fairness test contributed by one participant is a better fairness test for everyone, and it reveals nothing about the contributor's book.
No carrier competes on the quality of its proxy discrimination scan. Every carrier competes on price, appetite, distribution and service. Improving a shared test costs a contributor nothing it was monetising.
3 · Minimum-cell-size aggregates
Where a genuine benchmark is useful — readiness against a control set, adoption of a control, certification completion by role — participants contribute a scored result and the platform publishes the aggregate. Minimum cell size five, no attribution, suppression rather than estimation below threshold.
No financial benchmarks. No loss ratios, expense ratios, rate movements, retention or premium figures. Those are the numbers that would let a participant infer a competitor's position, and no cell size makes them safe.
4 · Facilitated practitioner forums
Structured sessions where practitioners work a real problem together under a common framework, with a facilitator and a rule set. The oldest and still most effective knowledge transfer mechanism in the industry — the ASEAN Insurance Council has run on this principle since 1978.
The output is not a conference summary. It is a versioned addition to the shared method library that every participant can apply the following week.
The Line That Cannot Move
| Never exchanged, under any mechanism | Why |
|---|---|
| Loss ratios, expense ratios, combined ratios | The core competitive metrics. No aggregation makes them safe in a market with a small number of large participants |
| Rate movements or pricing levels | Exchange of pricing information between competitors raises competition law issues independent of any commercial concern, and this platform will not host it |
| Appetite, capacity or retention positions | Directly commercially sensitive and directly inferable into a competitor's strategy |
| Claims records or claimant information | Personal data with its own legal regime, and of no legitimate value to a competitor |
| Distribution or commission economics | The most closely held commercial information in the industry |
| Anything identifying a specific transaction, client or counterparty | Anonymisation that leaves a transaction identifiable is not anonymisation |
Data Sovereignty Charter
No insurer will put its decision-making on a platform shared with its competitors unless the data question is answered completely, verifiably and in advance. This charter is that answer. It is written to be tested, not to be reassuring.
The Boundary
- Policyholder and claimant personal data
- Policy, claims, exposure and accumulation files
- Pricing models, rating factors, technical rates
- Underwriting appetite and declinature rules
- Reserving assumptions and reserve balances
- Capital position ahead of regulatory filing
- Reinsurance structures, terms, counterparties
- Distribution and commission economics
- Financial results ahead of publication
- The participant's own model inventory and validation evidence
- Individual decision records and their outcomes
- Strategy, M&A activity and corporate plans
IN
ARCHITECTURE
- Frameworks, taxonomies, decision templates
- Control sets mapped to regulatory instruments
- Model architectures and reference implementations
- Validation and fairness testing methodologies
- Hazard model libraries and climate pathways
- Regulatory intelligence and obligation tracking
- Agent designs and authority envelope patterns
- Playbooks and decision sequences
- Training curricula and certification
- Anonymised aggregates, minimum cell size five
- Anonymised failure and success patterns
- Published market and regulatory data
Nine Commitments — Each One Testable
| No. | Commitment | How a participant can verify it |
|---|---|---|
| 01 | No training on participant data | Model provenance records state the training corpus for every shared model. A participant may commission an independent audit of that provenance at the platform's cost, once per year |
| 02 | No cross-participant computation | The platform performs no computation that reads more than one participant's data together. There is no cross-book analytical surface in the architecture, so there is none to audit around |
| 03 | No operator access to decision content | Platform support operates against configuration and telemetry, not content. Any access to participant content requires a per-instance, time-boxed grant by the participant, logged and visible to the participant |
| 04 | Egress transparency | The participant can see, at any time, every category of information that has left its instance and when. The Boundary Auditor agent runs this check continuously and reports to the participant, not to the platform |
| 05 | Aggregate contribution is opt-in and reversible | Contribution to any aggregate is per-metric opt-in, and withdrawal recomputes the aggregate without the participant's contribution |
| 06 | Minimum cell size with suppression | No aggregate is published below five contributing firms. Below the threshold the cell is suppressed and marked as such, never estimated or filled from an adjacent cell |
| 07 | No financial or pricing benchmarks | Loss ratios, expense ratios, rate movements, retention and premium figures are outside the aggregate set entirely. This is an architectural exclusion, not a configuration default |
| 08 | Full export at any time | Frameworks, control sets, model inventory, evidence packs and the participant's complete decision history export in open, documented formats on demand, with no exit fee and no notice period |
| 09 | No silent scope change | Any change to what leaves a participant instance requires affirmative written consent. It is a contractual amendment, not a terms-of-service update, and silence is not consent |
The Question Every Prospective Participant Asks
"If you learn nothing from our data, what improves over time?"
Method improves. Every participant applying the same framework to a different book surfaces different weaknesses in the framework — and the weakness is reportable without the book being visible. A carrier that finds the fairness battery misses a particular proxy pattern reports the gap in the method, not the finding in its portfolio. Over three years the shared method set becomes considerably better than anything one firm would have built, and no firm has disclosed anything. That is a slower improvement curve than pooling data would give. It is the only curve that is commercially and legally available, and this platform is honest that it chose the slower one deliberately.
"What stops you changing your mind once we depend on you?"
Three things, in descending order of reliability. First, architecture: there is no aggregated data store, so there is nothing to change your mind about — reversing this would require rebuilding the platform, not amending a policy. Second, the export commitment: a participant that can leave with everything at any time, at no cost, is not dependent in the way that makes exploitation profitable. Third, contract: the undertaking is contractual, and change requires affirmative consent. The order matters — a promise backed only by contract is worth exactly as much as the counterparty's incentive to keep it.
Shared Platform Economics
The operating model rests on a single observation: the decision infrastructure an insurer needs is almost entirely non-differentiating, and almost entirely fixed-cost. Every participant is currently paying the full fixed cost of an asset that gives it no competitive advantage. That is the inefficiency the platform monetises, and it is the participant who keeps most of the gain.
Build Alone Against Shared Platform
| Control architecture design | Designed from first principles, once per firm, by people who have not done it before |
| Framework and method development | Fairness batteries, validation protocols, hazard libraries, trigger design — each invented locally |
| Regulatory interpretation | Every firm reads the same instrument separately and reaches a slightly different operational answer |
| Specialist hiring | Competing for the same scarce model validation and AI governance talent as every peer |
| Vendor negotiation | Negotiating AI vendor terms alone, without the leverage to obtain telemetry access or validation artefacts |
| Time to capability | Measured in years, against a transition period measured in months |
| Learning | Only from own mistakes, and only after they have cost something |
| Competitive advantage gained | NONE No customer chose an insurer because its model validation protocol was proprietary |
| Control architecture design | Designed once against the published instruments, maintained centrally as they change |
| Framework and method development | Built once, improved by every participant's application to a different book |
| Regulatory interpretation | Interpreted once, published to all, updated as instruments move from consultation to final |
| Specialist hiring | Scarce capability accessed through the platform and built internally through the Academy, in parallel |
| Vendor negotiation | A common evidence standard that vendors must meet to be usable by any participant |
| Time to capability | Weeks to a working control set, then adaptation to the participant's own profile |
| Learning | From anonymised patterns across the industry, before the mistake is made |
| Competitive advantage retained | ALL OF IT Pricing, appetite, distribution, service and claims handling stay entirely proprietary |
What Is Differentiating and What Is Not
Non-differentiating — should be shared
Nothing on this list has ever won a renewal, a broker relationship or a customer. Every item on it costs real money to build and real money to maintain.
Differentiating — stays private
This is where insurers actually compete, and none of it touches the platform. A participant using InsurEDIC³ competes with a participant using InsurEDIC³ on exactly the dimensions they competed on before — with better instruments on both sides.
Participation Model
Resource Access
The frameworks and methods library, regulatory intelligence, playbooks and the Academy. No platform instance. The entry point for brokers, MGAs, captives and small carriers.
Platform Instance
A participant-controlled instance with the decision ledger, agent fabric and Control Tower, integrated read-only to the participant's own systems. Full sovereignty boundary.
Instance plus Capability Programme
Platform instance with a structured capability-building programme — role pathways, certification and facilitated practitioner forums, sized to the participant's AI estate.
Consortium and Regional
For associations, regional pools and groups of smaller carriers participating collectively — the structure that makes the platform reachable for markets with low premium bases.
AI for Insurance Academy
The proposed MAS Guidelines on AI Risk Management name capabilities and capacity as an expectation domain in its own right. It is the one domain software cannot satisfy. The Academy exists because a platform that hands an under-skilled organisation more powerful tools has made the risk worse, not better — and because the scarcest thing in this industry is not models, it is people who can judge them.
Track A · Foundations
Track B · Decision Intelligence
Track C · Governance and Regulation
Track D · Domain Application
Track E · Leadership and Board
Alignment with the National Skills Architecture
MAS and IBF, supported by Workforce Singapore, partnered eleven financial institutions — including Income Insurance, Manulife and Prudential — to pilot workforce transformation initiatives and prepare the sector workforce for an AI-enabled future, unpacking what AI and generative AI adoption means for specific job roles. IBF also maintains the Skills Framework for Financial Services, the Data Analytics and Automation and Generative AI Jobs Transformation Maps, and the Young Talent Programme for AI in Finance. At regional level, the AIRM operates capacity building through AITRI, and the ASEAN Insurance Education Committee coordinates regional insurance education.
The Academy is designed to map onto that architecture rather than compete with it — role-based pathways, defined technical competencies, and applied assessment. No accreditation, recognition, funding eligibility or affiliation with IBF, MAS, WSG, AITRI or the AIEC is claimed or implied. Any such recognition would be applied for on its own merits.
Learning Pathways
A module list is not a capability programme. A pathway is a sequence for a specific role, ending in an applied assessment on that person's own decision problems, in their own instance, on their own data — which never leaves it.
Role Pathways
| Role | Sequence | Duration | Applied assessment |
|---|---|---|---|
| Underwriter | FND-01 → FND-04 → DEC-01 → APP-01 → DEC-04 | 6 days | Design the authority envelope for an AI-assisted underwriting workflow in their own line, and defend the risks they chose to keep human |
| Claims manager | FND-01 → FND-04 → DEC-02 → APP-02 → DEC-04 | 6 days | Run the friction equity battery on their own claims cohorts and produce the remediation case for what it surfaces |
| Actuary / pricing | FND-03 → DEC-03 → APP-03 → GOV-03 | 8 days | Validate a pricing model to materiality tier two standard and defend a rate movement it produced to a simulated supervisory panel |
| Risk and second line | FND-01 → FND-03 → GOV-01 → GOV-02 → GOV-03 → DEC-04 | 8 days | Stand up an AI inventory and materiality classification for their own firm's estate, including shadow and embedded models |
| Compliance | FND-01 → GOV-01 → GOV-02 → GOV-04 → DEC-01 | 5 days | Produce the evidence pack for one named obligation, assembled from the decision ledger, to examination standard |
| Technology and data | FND-02 → FND-03 → DEC-02 → GOV-02 → GOV-04 | 6 days | Map their own AI estate against the life-cycle control set and identify every control with no owner |
| Catastrophe and climate | FND-03 → DEC-03 → APP-04 → GOV-03 | 8 days | Design a parametric trigger for a thin-data peril with basis risk quantified, and stress it against historical events |
| Board and executive | FND-01 → LDR-01 → LDR-02 | 3 days | Chair a simulated board AI risk review, including an agent that acted outside its envelope, and decide what to do about it |
Progression Model
The Capacity Problem This Is Actually Trying to Solve
Every insurer in this market is hiring for the same two roles
Model validation capacity and second-line AI competence are the two constraints that appear in every readiness assessment, and they cannot be solved by hiring because the pool does not exist at the size the sector needs. The industry is bidding against itself for a small number of people, which raises cost for everyone and closes no gap in aggregate.
Competing for the same scarce hires. It reallocates capability between firms and adds none to the sector.
Outsourcing validation to consultancies. Fast, expensive, and leaves the firm without the capability once the engagement ends.
Converting existing actuarial, audit and risk professionals into AI-competent ones. They already have the judgement and the domain knowledge; what they lack is the specific competence, and that is teachable.