Mosaic Risk — hail risk intelligence, scored by geography
The hail risk data underwriters price against rests on two things: reports filed by people, and hail inferred from radar. Mosaic Risk rests on hail that was measured where it fell, then extended outward by Hailios Fusion AI, trained on it — and it tells you, cell by cell, which of those you are looking at.
Mosaic RiskMOSAIC
Powered by Hailios Fusion AI · Patent pending
UNDERWRITING · REINSURANCE · CAT MODELING · CAPITALPROVENANCE STATED PER CELL
Selected location
Specimen
Front Range · El Paso County, CO
87/100
High · hail alley
The hail layers underwriters price against were assembled from reports and from radar-derived proxies. Both are useful. Neither is a measurement. A report requires a witness who was outside, awake and inclined to call it in, which quietly encodes population density, daylight hours and road networks into what is then presented as a hazard surface. A radar proxy converts returns into hail on a coarse grid, from a beam passing well above the roofline, and publishes the result without saying how well it did.
Price a book on that and you pay for the sampling bias twice: once where the record over-reports because the town is there, and once where it under-reports because nobody was. Risk starts from the other end. Instruments in the storm record what actually landed, and the layer is extended outward from those points by a model trained on them — with the distance from measurement carried in the data rather than lost inside it.
The differentiator is provenance. Provenance is the whole product.
A hail report is a person. Someone was outside, awake, near a road, and moved to pick up a phone. The resulting map is an excellent record of where people were standing — biased toward population, toward daylight, toward highways, and silent over farmland at three in the morning. It is then used, largely unchallenged, as a record of where hail falls.
Radar does not see hail. It sees returns, and a proxy converts those returns into a hail estimate across a coarse grid, from a beam sweeping thousands of feet above the ground. It is inference, published without an accuracy statement, and it inherits every assumption in the conversion.
An instrument on the ground, in the storm, recording impacts as they land: how many, how hard, how long, timestamped to the second. No observer, no proxy, no conversion. This is what Risk is built on, and the model that fills in the rest of the layer is trained on it.
THE FIRST TWO ARE INFERENCE WEARING A CONFIDENT FONT. THE THIRD WAS THERE.
An underwriter who cannot see the confidence cannot use the layer. So Risk draws it. Cells we measured are crisp squares at full strength; ground-trained cells are drawn softer as they leave the instruments, and each one states its confidence. That confidence is not a guess about a guess: Hailios Fusion AI is scored against instruments that were standing in the storm.
Every scored cell states three things before it states a hazard number: what was measured, what was ground-trained on measurement, and how confident the layer is in the result. Those are not a footnote at the bottom of a methodology PDF. They are fields, they travel with the score, and they are the first thing a cat modeler should interrogate.
Nobody else can show you this, because nobody else has a ground network to grade a model against. A layer checked only against radar and storm reports can tell you what it thinks; it cannot tell you how often it is right. Ours can, and it is the first question we would want a cat modeler to ask us. A hazard surface with uniform confidence everywhere is not confident, it is silent. Measured skill is what makes every cell worth believing, and what lets you load or decline one for a stated reason.
Specimen · one surface, every cell labeled
Not a color on a map. A record with its provenance attached, for a book you have to rate, retain or cede. Read the provenance and confidence fields first — they tell you how much weight the hazard field should carry inside your own model.
| Scored unit | A geographic cell. Every cell carries the same fields whether it sits under an instrument or between them — there is no quiet gap where the layer stops explaining itself. |
|---|---|
| Hazard | Long-record hail hazard expressed through what Tessera measures: impact energy and impact concentration, with stone size as a measured floor where sensors stand, ground-trained and labeled everywhere else. |
| Provenance | Measured, ground-trained, or modeled. Stated per cell, every time. This is the field an underwriter should read first, and the reason the rest of the table is worth reading at all. |
| Confidence | How sure the layer is about the cell, and what that confidence rests on — instrument density in the neighborhood, depth of the record behind it, agreement between independent sources. |
| Distance to measurement | How far the nearest measured cell actually is. Provenance you can audit rather than take on trust. |
| Record depth | How far the measured record and the longer reported-hazard record each reach behind the cell, stated separately — so a short measured history is never mistaken for a long one. |
| Coverage | Available in the US, Canada, Europe, Australia and Japan. Across the US, scored by Hailios Fusion AI on six years of measured hail and 26+ years of reported record. Elsewhere, built on historical radar from January 1, 2025 onward, plus Tessera measurements wherever one stands; ask us to confirm availability for your locations. The sensors themselves deploy in 150+ countries. |
| Delivery | Score your own portfolio location by location. |
| Commercial shape | Annual subscription, per portfolio. Talk to us. |
NO FIELD ABOVE IS AN OPINION WITH A DECIMAL POINT ON IT.
NA
North America — our home market and the deepest instrumented record we hold. Across the US, Risk runs on Hailios Fusion AI; in Canada, on historical radar from January 1, 2025 onward, plus Tessera measurements wherever one stands.
EU
Years of measurement inside the European hail belt, not adjacent to it. Risk is available today on historical radar from January 1, 2025 onward, plus Tessera measurements wherever one stands. Training Hailios Fusion AI on that record is the next step.
AUS
Years of measurement across one of the most hail-exposed insured markets in the world. Risk is available today on historical radar from January 1, 2025 onward, plus Tessera measurements wherever one stands. Training Hailios Fusion AI on that record is the next step.
JPN
Risk, available today on historical radar from January 1, 2025 onward, plus Tessera measurements wherever one stands.
MEASURED WHERE WE STAND · GROUND-TRAINED EVERYWHERE ELSE.
Order either for a single address, or run a whole portfolio through them as a list.
One location, one reference date
Any US address · CAN · EU · AUS · JPN
How exposed is this location? Its risk level, how often hail reaches each threshold and a regional comparison, with the history included. In the US it is ground-trained by Hailios Fusion AI.
One location, two years
Any US address · CAN · EU · AUS · JPN
What happened here over the past two years? Every event with its date and severity, for a purchase, a renewal or a date of loss nobody is sure of. In the US it is ground-trained by Hailios Fusion AI.
Outside the US, both reports are built on historical radar from January 1, 2025 onward, plus Tessera measurements wherever one stands. Ask us to confirm availability for your locations.
Same measured surface underneath all three, assembled by the same AI layer. Risk is the decades view; the rest of Mosaic handles the single storm, and the one that hasn’t arrived yet.
Powered by Hailios Fusion AI
Mosaic Reports
This address, this date, this degree. One report an adjuster can file, a broker can hand over and a reinsurer can audit later without re-arguing the weather. Risk says what the territory costs; Reports says what happened on Tuesday.
Explore Reports →
Mosaic Alerts
Alerts fire when hail above your threshold is heading for a given site — up to 30 minutes of warning for the operators and insureds sitting inside the cells Risk scored.
Explore Alerts →
Risk is only as good as the ground truth it answers to. Hailios Fusion AI carries the layer beyond the instruments; the instruments are what keep it honest, and the layer sharpens in exactly the places where a sensor goes up.
That is the roadmap in one line: coverage deepens wherever another Tessera goes up. If your portfolio concentrates somewhere we have not instrumented yet, that is a conversation about deployment, and it is a cheaper one to have before the season than after it.
TESSERA MEASURES. MOSAIC MAKES SENSE OF IT. THE LONG PATTERN IS WHAT FALLS OUT WHEN YOU LEAVE THE INSTRUMENTS RUNNING FOR YEARS.
Tell us what you underwrite, cede or model, and where the exposure concentrates. We’ll show you what the layer says about those cells — including the ones where it says it isn’t sure.
ANNUAL SUBSCRIPTION, PER PORTFOLIO · A REAL PERSON WILL REPLY, USUALLY THE SAME DAY
Talk to us