Insurance companies rely on extensive historical data to predict and manage risk and establish rates. Historical data is becoming less reliable as insurance companies face simultaneous challenges of escalating climate change risks and shifting industry regulations.

Challenge

A prominent insurance company approached Kopius* seeking a way to better predict and plan for weather disasters. Doing so would help improve its risk models and allow more accurate premiums for policyholders. Given the unpredictable nature of many weather events, a solution that gave real-time information about whether a weather event had occurred or was about to occur was important.

The Solution

Kopius was engaged to develop a pilot solution using cloud-based machine learning and IoT devices to identify weather events and notify the client in real-time. We trained the machines to recognize sounds picked up by IoT devices at weather ground stations, using 10,000 audio seed files.

Results

With our algorithms and AI process, we could train the machines to recognize 12 unique types of weather events, including metrics tracking the severity of the events.

The pilot included placing 500 devices with machine learning modes at strategically selected locations nationwide. The input from the devices was found to be 85% accurate in the initial pilot.  The pilot informed future strategies for the insurance company and contributed to the technology development roadmap.

*Kopius performed this work under its previously known business name, Valence.