The Global Risks Alliance (GRA)
Data, automation and machine learning now reach every step of insurance: distribution, underwriting, pricing, claims and fraud detection. They can lower costs and widen access, and they can also discriminate, exclude, mislead or fail in ways that are hard to see. This page sets out how technology and model risk develop in insurance, how their effects on customers are measured, and which institutions hold the governance decisions.
Insurtech and AI · Coastal portfolio · Event scenario EV-30
Fig.Protection gap and claims surge. Design wireframe with placeholder names and illustrative values, not a record or a live system.
AI in insurance is the use of machine learning and automation across distribution, underwriting, pricing, claims and fraud detection. It can lower cost and widen access, and it can also discriminate through proxies, drift as data change or concentrate dependence on a few vendors. Supervisors expect governance, testing, explanation and human oversight.
How satellite radar turns a flood into an exposure update. Synthetic scene; no real event or place.
AI model governance and customer effects. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Complex models are hard to explain to customers, supervisors and boards. Performance drifts as data and behaviour change, and errors can scale across a whole book of business before anyone notices.
Rich external data, from credit information to telematics and online behaviour, can stand in for protected characteristics. A model that never sees a protected attribute can still produce disparate outcomes.
Finer segmentation moves premium towards individual expected loss. It can reward risk reduction, but it can also price some people out and erode the pooling on which insurance relies.
Many insurers rely on the same data vendors, models and cloud platforms. A flawed dataset or a vendor outage then affects the market at once, and accountability is split across contracts.
Each term as this page uses it. Where a standard-setting body defines the term, the wording follows that body.
| Term | Definition |
|---|---|
| Proxy discrimination | Unfair treatment of a protected group that arises when a model uses variables correlated with a protected characteristic it does not see directly. |
| Model drift | A decline in a model's accuracy as the data it meets in use move away from the data it was trained on. |
| Price optimisation | Setting individual premiums partly on predicted willingness to pay or likelihood to switch rather than on risk alone; restricted in several jurisdictions. |
| Human oversight | Review of automated decisions by people with the authority and information to change them. |
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Label |
|---|---|---|---|---|
| AI systems in use by business function | count of systems by function | survey cycle | Supervisory surveys of insurers | REPORTED |
| Disparate outcome ratio | ratio of outcome rates between groups | per model review | Fairness testing | ESTIMATED |
| Model performance drift | change in discrimination or calibration metric | monthly | Model monitoring | OBSERVED |
| Automated claims decision share | percent of claims decided without human review | annual | Insurer disclosures and surveys | REPORTED |
| Complaints about automated decisions | complaints per thousand automated decisions | annual | Supervisory complaint data | OFFICIAL |
| Critical third-party model and data dependencies | count of critical vendors | annual | Vendor registers | DECLARED |
The topic as a computational graph: named sources feed the methods, an evidence review gates the outputs, and the decisions stay with the authorities that hold them.
Fig. 02Nodes carry their type, two metric rows and a state; connectors show what feeds what; dashed frames name the owners; the timeline is the cadence. Design of the pipeline, not a record of a run.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Source | What it gives | Licence class |
|---|---|---|
| NAIC artificial intelligence surveys of insurers | Use of AI by function in motor, home and life insurance in the United States | Public reports |
| EIOPA reports on digitalisation and AI governance | Use of big data analytics and AI governance principles for European insurers | Public, EU reuse terms |
| Gallagher Re Global InsurTech Report | Investment, start-up activity and themes in insurance technology | Proprietary, published report free to read |
| AI Incident Database | Reported incidents involving AI systems, including in financial services | Open data, attribution and terms |
| OECD.AI Policy Observatory | AI policies, principles and trends by country | Free, OECD terms |
| IAIS application papers and FinTech publications | Supervisory practice on AI, big data and conduct | Public, IAIS terms |
Insurance Nexus prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Who decides | The question | Evidence this page serves |
|---|---|---|
| Insurer boards and model risk committees | Which AI systems to deploy, with what oversight, testing and human review | Model inventory, validation and fairness testing, drift monitoring and incident history |
| Insurance supervisors | How to supervise AI use under existing conduct and governance rules and new AI law | Survey evidence, model governance documentation, complaint and outcome data |
| Legislators and data protection authorities | Which data sources and automated decisions are permitted in insurance | Evidence on predictive value, proxy effects, transparency and redress |
| Consumer bodies and ombudsman services | Whether automated decisions are fair and explained to customers | Complaints, decisions and outcome disparities by group |
Insurers use AI for customer service, underwriting, pricing, fraud detection and claims assessment. Machine-learning models can find patterns in large datasets that traditional methods miss, but they can be opaque, sensitive to data quality and prone to reproducing historical biases. Variables that seem neutral can act as proxies for protected characteristics. Generative AI introduces further questions about accuracy and accountability.
Regulators in several jurisdictions have issued principles or rules for AI in insurance and finance, emphasizing governance, explainability, fairness and human oversight. Dependence on a small number of cloud and model providers creates concentration risk across the sector. Insurtech firms have also changed distribution, through embedded insurance sold alongside other products and usage-based cover informed by telematics and connected devices, which raises questions about data ownership, consent and the boundary between risk prevention and surveillance.
Approaches to validation, monitoring and accountability, drawing on artificial intelligence and machine learning.
Statistical methods from probability and statistics assess whether models produce disparate outcomes.
Studies of how automated decisions affect access, price and claims experience.
Fairness has several competing definitions, model behaviour can drift over time, and proprietary models limit external scrutiny.
The current service level is R0: information and enquiries. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.