Our Technology

Kettle outperforms the industry by using proprietary machine learning algorithms that use more than seven billion lines of weather and ground truth data.

Book a Demo

Laptop computer demonstrating Kettle's Wildfire prediction tool

Kettle is an underwriting platform built for the modern world

We use groundbreaking technology to deliver better protections for people and more stable returns for the insurance industry.

We researched and collected over 3 petabytes of data from alternative sources for wildfire prediction.

This includes such data sources as moderate resolution imaging spectroradiometer (MODIS) and light detection and ranging method (LIDAR).

MODIS captures data in 36 spectral bands ranging in wavelength from 0.4 μm to 14.4 μm and at varying spatial resolutions (2 bands at 250 m, 5 bands at 500 m and 29 bands at 1 km).

LIDAR is a remote sensing method that uses light in the form of a pulsed laser to measure ranges to Earth.

We process the collected raw data using an extract, transform, and load (ETL) pipeline we built using computer vision technology.

More than 7 billion rows of data were generated for the machine learning task, giving Kettle a head start compared to others in the market.

Computer vision and mathematical interpolation are critical in processing the data so it’s digestible for meaningful analytical work.

We apply our proprietary technology, swarm neural networks comprised of 115,456 separate nodes, to the data.

We train these deep convolutional neural networks on factors influencing wildfires, and create ensembles to predict wildfire propensity.

Our technology spans different wildfire characteristics such as actual wildfire events, high wildfire-potential areas, and non wildfire areas.

Kettle’s technology uses two models, the Genesis Model & Contagion Model, to Predict wildfire risk.

Using 42 million grid-level high resolution simulations to generate wildfire ignitions and spreading pattern, Kettle accurately captures the distribution of wildfire risk and property damage.

Using the Genesis Model, Kettle divided California into 320,000 micro grids - each 0.5 square mile.

We train our Genesis Model by discretizing the state of California into grids and analyzing the factors that contribute to a wildfire ignition in a specific month.

The accuracy
score of Kettle’s
Genesis Model is

89.2%

Kettle’s contagion model adds AI components to further improve accuracy and computational efficiency.

The Rothermel Wildfire Model’s average time to compute one wildfire is four hours. Through the use of technology, Kettle shortens the computational time to approximately 20 seconds for a typical wildfire.

We use the accurate results of Kettle’s wildfire simulation model to develop a pricing model and a portfolio optimization algorithm.

We use the pricing model and algorithm to create optimal portfolios of risk for our clients.

Book a Demo

We started with fire, but are quickly moving into all types of catastrophic risk.

Risk will continue to grow in complexity and variables will continue correlating more intricately, and data will increasingly expand. Our system is meant to take the massive stores of unstructured data and use them to create better products.

People standing in a flooded street

Flood

We are using the same weather and satellite data we use for fires, combined with additional data sources, and apply our machine learning models to predict flood patterns and risks.

People standing in a flooded street

Convective Storms

We use our advanced machine learning models to predict wind and hurricane risks.