[ Summary ]
Our team helped redesign underwriting for a top US life insurer by leveraging advanced AI models to refine risk assessment. By identifying nuanced patterns in medical data, the insurer was able to accurately price premiums and approve previously rejected applications, enhancing profitability and underwriting quality.
Life & Health Underwriters have to classify risks broadly and tend to apply broad pattern. For example, an insurer may view, hypothetically, all diabetics to be of the same risk perhaps even, in the most extreme use, uninsurable. In fact, there will be many Type 1 diabetics who are healthier and have lower mortality risks than the average population and many Type 2 who are unhealthier than average population. However, this is not categorically true in every single case for either the Type 1 diabetic or the Type 2 diabetic. To make underwriting even more challenging, there will be undiagnosed diabetics who may be showing symptoms of the disease, hidden within the medical information provided but have note received an official diagnosis at the type of application. Diabetes is used as an example to illustrate but the same fact pattern is true for many diseases.
More refined understanding of existing data to price to underwrite more accurately and price risk better in the future, allowing both higher premium and more profitable premiums.Specific client example: A top 5 US life insurer had this exact problem with a disease category. That is, treating insurance applicants in a homogenous manner missed opportunities. Compounding this missed opportunity was the risk of missing evidence of an undiagnosed disease harming profitability, A top 5 insurer has vast amounts of data but if large insures lack ability to determine patterns in the data, the quantify of data value is not that valuable. Humans alone can not completely discern all needed patterns.Lazarus AI
Lazarus AI brought our proprietary business trained models to this underwriting problem. Our models used medical records with other application input and we did some refinement to identify mortality risk. After this limited training, Lazarus AI foundation models delivered value beyond what employees alone could do. In the future no additional tuning will be needed for this disease.
For people diagnosed with the disease, the insurer found they could write some previously rejected applications while also pricing accurately. The underwriters were also alerted to markers such as co-morbidities allowing them to probe deeper on applications where there were indicators of undiagnosed disease.
Lazarus AI delivered value by increasing premium, increasing quality of underwriting through AI augmentation.