Methodology

Source monitoring

RiskEvents monitors the world's news output in real time, across 65 languages, organised into five source classes: official advisories (government agencies, meteorological services), wire services, trade media, mainstream media, and social/community sources. Natural catastrophe data feeds (USGS, NOAA, GDACS, FIRMS, EMSC) are polled at higher frequency for rapid detection.

AI classification pipeline

RiskEvents runs a provider-aware AI pipeline and is not tied to a single AI provider. Incoming source items are triaged for specialty-insurance relevance and market materiality, then structured, matched to existing events or created as new Signal records, reconciled against source-family evidence, and enriched with geographic scope, likely-affected Lloyd's market classifications, macro-event context and, where available, analyst and ground-report layers. Models and prompts are benchmarked, version-controlled, and evaluation-gated before production use. Outputs are tracked for quality monitoring, and structured corrections are recorded when human reviewers modify classifications.

Impact assessment

Potential impact is an estimate of London specialty-market materiality, not a measure of human severity. It considers likely insured-loss relevance, affected classes of business, geographic or asset footprint, source quality, escalation trajectory, and market context.

Impact assessments are initial estimates based on available information at the time of classification. They may change as events develop and more information becomes available.

Event lifecycle

Events progress through a defined lifecycle: Signal (initial detection, limited corroboration), Developing (corroborated but not yet fully confirmed), Active (confirmed and being tracked), Monitoring (no new updates for 6+ hours), Closed (resolved, no new updates for 48+ hours), and Retracted (determined to be incorrect or no longer relevant).

Promotion from Signal to Developing or Active requires corroboration from multiple independent sources and sufficient AI confidence. Events sourced exclusively from social media cannot be auto-promoted and require at least one non-social source.

Corroboration and source families

The platform distinguishes between original reporting and syndicated content. Multiple articles from the same wire service report are grouped into a single source family. The corroborating source count reflects the number of independent source families, not the total number of articles. The publicly displayed report count shows the total number of source items for transparency.

Review status

RiskEvents is primarily an AI-driven platform. Each event displays a review label: “AI Generated” (detected and classified by AI with no human review) or “Human Verified” (confirmed or corrected by a human reviewer). Reviewer identities are not disclosed.

Intelligence Priority and the evidence ledger

The default feed order, Intelligence Priority, is a composite reading order rather than a single sort key. It blends recency of material updates, assessed impact, depth of corroboration across independent source families, and situation grouping, so that the most decision-relevant items surface first. Switching the sort control to Most recent or Highest impact applies that single literal key instead.

Inside each event, structured intelligence is graded on an evidence ledger with three levels: Known (corroborated by authoritative or multiple independent sources), Reported (asserted by credible reporting but not yet independently corroborated), and Uncertain (conflicting or thin sourcing). Claims can move between levels as evidence accumulates, and superseded claims remain visible in the revision history.

Status and colour legend

Colour carries exactly one meaning each across the product. Vermillion marks High impact only; amber marks Medium impact only; Low impact renders as a quiet outline. Lifecycle states are written as words (Signal, Developing, Active, Monitoring, Closed, Retracted) rather than encoded in colour, with a small pulse marking live states. Peril categories use their own fixed palette, applied to dots, chart marks and map pins, and nowhere else.

Glossary

Core Market feed

The default event feed: every peril class relevant to specialty insurance, with Conflict & Political Risk excluded unless explicitly included.

C&PR desk

The separate Conflict & Political Risk feed covering war, terrorism, sanctions and political-risk events.

The Box

The line-of-business view, named for the underwriting box at Lloyd's: pick one or several lines and see their complete lifecycle picture. The Box is a destination, not a feed, so it always includes both the Core Market feed and the C&PR desk for its line -- no toggle applies there.

Lens

The time window applied inside The Box (all time, 7 days, 30 days, or a custom range).

Situation

A macro-event grouping related events into one narrative -- a hurricane and its landfalls, a conflict and its incidents.

Line of business (LoB)

A class of specialty insurance business (Marine, Energy, Cyber, Property and so on) used to slice events by underwriting relevance.

Source family

A group of articles tracing back to one original report, counted once for corroboration purposes.

Intelligence Priority

The default composite feed order described above.

Known / Reported / Uncertain

The three evidence-ledger grades described above.

Limitations

RiskEvents is an intelligence tool, not a definitive source of truth. AI classification may contain errors. Impact assessments are indicative, not actuarial. Coverage spans the world's news output in 65 languages but may still miss events reported only in closed, paywalled, or unindexed channels. Users should always verify critical intelligence through independent sources before making underwriting or exposure management decisions.

Questions about our methodology? Contact us.