We only reach out when the evidence is strong.
A policy needs a 35% or higher confidence of real loss before we draft anything. Below the bar, we stay quiet, and you can report the usual way at any time.
A fictional insurer, built by AI Accelerator as a demonstration. Every name, figure and customer is invented.Back to AI for insurers →
Signal reads radar, satellite and weather-station data while an event is still happening, matches it to the homes, farms and cargo we insure, and gets in touch first. A claims handler checks every message before it goes.
Figures from the Tarnmoor Valley pilot simulation. Illustrative.
Simulated event on a fictional book of 190 policies. Click any home, farm or vessel on the map to see what that customer receives.
A claims line that answers in seconds still waits for the customer to call. People are clearing up, checking on family, getting the roof covered. On a farm, damage can take days to show. Signal closes that gap from our side.
The same event data that tells a forecaster where hail fell tells us which policies sit underneath it. We put a person between the model and your inbox.
We watch radar, satellite and ground stations across every area we insure, around the clock.
Radar reflectivity · hail size estimates · roof condition from satellite imagery · wind fields · port notices
The event footprint is laid over our book. Each policy gets a confidence score for a real loss.
Location · cover type · building and farm data · score from 0 to 100%
Only policies above our confidence bar get a draft. A claims handler reads each draft and presses send.
Confidence bar 35% · claims handler review about 12 minutes a file
You hear from us first, with a visit to book or a loss report already drafted. You decide whether to go ahead.
App · text · your broker, if you have one
The first version of this idea messaged everyone under the storm. Some people were told about a claim they didn't have. These four rules came out of fixing that, and they don't change.
A policy needs a 35% or higher confidence of real loss before we draft anything. Below the bar, we stay quiet, and you can report the usual way at any time.
The model drafts. A Marrowby claims handler checks the policy, the wording and the event data, then sends it or bins it. No message goes out automatically.
A storm doesn't decide cover. Your policy does. So our first message offers help, without promising anything we can't stand behind.
On a farm policy, Signal fills in a draft loss report from the event data. You check it against the buildings and decide whether to send it. It is your call, every time.
Signal is included on Marrowby home, farm and marine cargo policies in the valley. There's nothing to install and nothing to pay.
The footprint that starts your conversation also tells the rest of Marrowby what just happened. Every team moves on the same picture, and each keeps one decision for a person to make. Figures follow the June replay.
Your firm gets a Signal feed: which of your clients sit under the footprint, what we've sent them, and who has replied. You can call first, with the facts in front of you.
| Client | Cover | Confidence | Status |
|---|
Set the size of an event and how strict the confidence bar should be. The model shows who we'd reach, who we'd miss and what it costs in claims handler time. Assumptions and gaps are listed underneath, so you can argue with them.
We're starting small and adding perils only when the last step works. Each stage has a named lead, the thing it depends on and the number it has to hit.
Live for hail in Millford and Eastbrook. Every draft reviewed by a claims handler.
Add named storms and wind. Brokers get the Signal feed for their clients.
Port closures for marine cargo. Leadership gets an indicative loss figure within a day.
Pick one, or type your own. Answers come from our Signal policy notes, so they're the same ones our claims handlers give.