The Floow Crash Detection Service empowers the precise identification of crashes involving mobile phone users while they’re driving a vehicle on the road. It is therefore the cornerstone of our FloowClaims product suite.
Our app-based Crash Detection feature seamlessly operates on both IOS and Android platforms. The service operates in the background, sampling data at high frequency from phone sensors, without the need of user intervention.
Employing cutting-edge signal processing and machine learning, The Floow Crash Detection service analyses real-time data to discern pivotal road events. By analysing factors such as acceleration, directional shifts, and vehicle dynamics, our service identifies potential crashes and generates notifications for action.
The insurance sector grapples with soaring claims costs across diverse domains, particularly in the motor industry, where claims continue to surge at an alarming rate. In the United States, the average claim cost has escalated to $7,5791, while in the United Kingdom, it stands at £5,349 ($6,487)2.
Against this background, the Floow’s crash detection technology delivers substantial value to our clients and users. For end-users, a crash is a source of extreme stress, addressable with prompt, data-driven support around recovery and claims reporting assistance. For insurers, real-time crash notifications deliver immediacy of service to customers in need and significantly reduce operational costs.
The challenge we have cracked in creating, maintaining, and enhancing our crash detection system is twofold: the availability of labelled data and the establishment of ground truth for model validation and benchmarking. From a data science perspective, crash detection poses a problem of imbalanced classification. As a result, the selection of appropriate metrics assumes paramount significance in benchmarking. To this end, we prioritize precision and recall3 as our metrics, with precision given more importance to minimize false alarms, aligning with invaluable client feedback. Benchmarking the crash algorithm is also a challenging process, necessitating manual review of the predictions to ensure acceptability. Notably, not all crashes get recorded or captured due to factors such as low battery or the absence of the phone in the vehicle. Furthermore, even when data is recorded, end-users may not have reported the crash for various reasons. Hence, we routinely perform manual reviews of the crash service’s output to ensure its alignment with our real-world benchmarks.
At present, our performance metrics consistently exceed 80% in both precision and recall across multiple clients. Clients who have integrated our crash service into their operations have witnessed a substantial uptick in customer satisfaction, even among those contacted as false alarms (false positives).
If you’d like to learn more about our crash service and how you can seamlessly integrate it into your existing operational workflows, please reach out to us at email@example.com or connect with one of my colleagues at Insuretech Connect in Las Vegas.
1 Highway Loss Data Institute
2 WTW. UK motor claims inflation on the rise with added pressure from delayed injury settlements – WTW (wtwco.com)
3 Precision and Recall – Precision is a measure of false alarms while Recall is a measure of misses. In other word, Precision is a measure of what percentage of the detected crashes are actual crashes while Recall is a measure of what percentage of all crashes were detected.
Share this article