ICEYE turns radar observations into AI-assisted catastrophe insights

ICEYE builds and operates radar satellite systems and supplies data products that describe conditions on Earth. Its natural-catastrophe business combines observations, machine learning and specialist analysis to help organisations understand floods and other events. For an insurer or public agency, the important distinction is between purchasing satellite imagery and receiving a prepared hazard dataset. The latter can shorten an analysis workflow, but it does not decide the meaning of an individual claim.

01 / ProductRadar observations become a hazard information product

The current Flood Solutions page describes flood depth, extent and duration, with delivery through an application, an API and partner integrations. Radar provides an observation source that can operate through clouds and darkness. The resulting hazard layers are not simply pictures: they combine observations and other inputs to express where flooding occurred and how an event evolved. That distinction makes ICEYE relevant to teams that need structured information rather than another imagery viewer.

The Flood Insights brief explicitly describes machine learning alongside radar imagery, third-party data and in-house meteorology and geospatial specialists. This is a substantive AI relationship grounded in a measurement workflow. It is also a useful boundary on the claim: the documented approach combines methods and expertise. Calling it AI-powered does not mean every value is directly observed by a satellite or that the analysis is independent of supporting data quality.

ICEYE’s broader catastrophe solutions portfolio covers more than flood, including hurricane and wildfire intelligence. The company also supplies radar imagery and sovereign satellite capabilities. These offers sit within ICEYE’s portfolio but should not be folded into a single software entitlement. A procurement brief should say which hazard, geography, event phase and delivery product it covers, because those choices determine how the data will be used.

02 / AudienceDesigned for organisations coordinating many exposures

Insurance teams may need to understand how a wide event intersects with a portfolio before individual reports are complete. Public agencies may need broad evidence to support recovery planning. ICEYE’s insurance overview presents event response, loss sizing and claims-related uses. The common requirement is a timely geographic view across many locations. The opportunity is to organise review more intelligently, not to replace every local observation with one remote dataset.

A household seeking a definitive statement about interior damage has a different need from a catastrophe team estimating the scale of an event. Depth or extent at a building location cannot by itself establish contents damage, policy coverage or repair cost. Organisations should define what the data is allowed to influence. A triage indicator can tolerate different uncertainty from evidence used to determine a payout, and the review process should reflect that difference.

Guidewire is relevant when the challenge is connecting new hazard information to an insurer’s existing systems and decisions. Hexagon offers an adjacent perspective on geospatial measurement and analysis. The comparison is about the question and workflow: observed event information, tools for interpreting geographic data and the system managing insurance operations are complementary components, rather than identical feeds that can be substituted without further evaluation.

03 / WorkflowA proposed retrospective flood-review exercise

Start with a completed flood event for which an authorised team has reliable portfolio and field records. This proposed workflow tests whether ICEYE information would have improved the order and quality of human review. It does not use real policyholder information in this article, and Sequenced has not run the exercise. An initial retrospective study makes it possible to investigate mistakes without allowing a model output to determine an individual’s outcome.

1. 01

   Establish the event timeline

   Record when the organisation first learned about the event, when reliable field reports arrived and when different internal decisions were made. Separate those moments from the time of the flood peak. A dataset can be accurate in hindsight while arriving too late to help the decision it was purchased to support.

2. 02

   Match locations carefully

   Use approved exposure locations and preserve geocoding uncertainty. A hazard layer and a portfolio address can appear to disagree because the address point is misplaced or represents a different building. Review a sample of these mismatches before attributing them to the hazard model, especially where multiple structures occupy the same property.

3. 03

   Compare successive releases

   Retain each release with its event and delivery timestamps. Determine whether revisions change the set of locations needing attention. Evaluate the early, less complete picture and the later assessment separately. Combining them into one final dataset would conceal how useful the information actually was at the time of a decision.

4. 04

   Keep claims judgement separate

   Have trained staff compare the hazard indicators with independent evidence and document disagreements. Assess whether the information improves outreach planning or review prioritisation. Any claim determination should follow the organisation’s authorised process and required evidence, with the geographic layer treated as one input rather than a substitute for coverage analysis.

A valuable pilot output is a map of uncertainty as well as a map of apparent impact. Retain areas without adequate evidence, uncertain locations and records that cannot be reconciled. This avoids a common analytical mistake: treating a blank area as confirmed absence of flooding. A review queue that clearly distinguishes strong evidence, weak evidence and missing information is more useful than an apparently complete map that hides those differences.

04 / PricingAgree the product, geography and delivery relationship

The reviewed public sources offer a sales-led route to catastrophe products; they do not establish a universal published per-property tariff. ICEYE’s September 2026 Insurity announcement adds an important route for SpatialKey customers in the United States and London insurance market: access through a single commercial relationship with Insurity. That is a stated partner channel, not evidence that every existing customer automatically receives every dataset at no extra cost.

The ICAS contract guide concerns access to the radar constellation, not a price list for the insurance insights product. It documents contract-controlled service options, delivery configurations and whether pricing information is attached. This is a practical reminder to keep interfaces separate: a raw-imagery API entitlement should not be assumed to include flood analytics, and a prepared-insights subscription should not be assumed to grant satellite tasking rights.

A useful cost comparison includes the work needed to consume each route. A prepared event layer may reduce specialist processing, while a raw imagery licence may offer flexibility to a team that already has that capability. Existing platform integration can simplify adoption, but it may also define how much history or metadata is exposed. Ask to see the actual deliverable and access conditions before estimating savings from the shorter workflow.

05 / DistinctionsObserved events can complement forward-looking risk models

The main distinction is between understanding a possible future event and observing an event that is occurring or has occurred. A probability model helps frame risk over a period; an event layer helps determine what happened during a particular flood. Neither eliminates the other. An organisation that conflates them may mistake a surprising observation for a model failure or use a historical event footprint as if it represented the full range of future risk.

ICEYE also connects a specialised sensing capability with industry-facing delivery. That can make radar-derived information useful to people who are not radar analysts. The abstraction is helpful when its limits remain visible: users need to know what a depth value represents, when it was revised and how much local support exists. The organisational benefit depends on communicating those definitions to claims, portfolio and recovery teams consistently.

06 / QuestionsTiming claims and local applicability need checking

The current flood page describes six-hour extent updates during an event. The older 2024 product brief gives a different cadence for a named rapid-impact component and notes selected-country availability. These should not be blended into a universal SLA. Ask which current product applies in the proposed countries and what delivery commitments are in the agreement. A public cadence description is not proof that every event or location will receive an equally complete observation.

The sources reviewed do not establish independent error rates for the proposed portfolio. They also do not settle how much performance changes across building types, terrain or local reference-data quality. Require a representative sample, explicit uncertainty information and a process for disputed observations. Keep financial decisions and customer communications within the organisation’s review controls; a technically integrated data feed is only one part of a dependable catastrophe response.

07 / DecisionStart with the event decision you need to improve