Page 38 - Insurance Times July 2024

With Generative AI: Data Ingestion and Pre-Processing - The solution can
Our Underwriting Co-Pilot, with its advanced LLM Models integrate with various internal and external applications
meticulously matches and aligns each data point with the to ingest data about applicant. This step can also
company's respective underwriting guidelines. The include extraction of data from scanned documents,
application is designed to highlight any parameter that which an applicant might have submitted for policy
deviates from the norm and pairs/overlay it with the approval, and may also include cleaning of data, data
relevant section of the guideline for a seamless review. format normalization, etc.
Embedding Generation - The textual information from
Assessing various risk parameters such as profile risk, the underwriting manuals and policy documents are
financial risk, geographical risk, Morbidity risk, Political processed to generate vector embeddings. This
Exposure, etc. the system acts like a well-orchestrated transformation allows the data to be processed and
symphony. Furthermore, the LLM models don't just evaluate understood by machine learning models.
- it predicts. Recommending the Next Best Action, it
Feature Engineering and Model Deployment - Key
presents a concise risk narrative for each application.
attributes relevant to risk assessment are extracted and
processed to a format suitable for training and
However, even with all this automation, it is designed to prediction. The pre-trained LLMs can be fine-tuned with
always respect the Underwriter, who have the option to insurance domain specific data to improve decision
reject the recommendations and modify the next steps based making accuracy.
on his/her experience and a particular case. Incorporating
all this feedback the LLM Model continuously learns and Assessment and Recommendation - Advanced
adapts, and able to provide more nuanced algorithms/ML Models analyses every data point,
recommendations. It's this collaborative intelligence that matches with the UW rules and generates risk scores.
enables the system to evolve. The Gen AI model, by collating, all the information,
recommends the decision along with a detailed
Case in Point: explanation and associated risks.

Consider an application where there is a substantial Feedback and Re-training - UWs can provide feedback
mismatch between the applicant's income and the on the LLM recommendations and the same is used to
refine model output for future cases.
requested coverage. The module aids underwriter by
providing the following recommendations:
The solution also includes components like load balancer
Decision: Decline the case (distribute incoming requests to ensure system reliability and
availability), container orchestration (to manage
Reason for Decision: The applicant's profile presents a high deployment and scaling of containerized application),
financial risk due to a significant discrepancy between database management (MongoDB, PostgreSQL, etc.), API
annual income and the proposed sum assured
Gateway and so on. In terms of Data Security and
Compliance, the solution uses cloud native services to ensure
Next Steps: Inform the applicant of the decision and the that all data stored or processed by this solution is in line
reasons for the rejection and advise to consider applying for
a lower sum assured that aligns with their financial profile with the regulations put forward by IRDAI and other
regulatory authorities in India.
This solution also provides detailed feedback on all the
The benefits our solution brings to the table are also very
parameters and risks it has identified and highlighted, which
can be reviewed by the Underwriter to check the accuracy robust and tangible:
of assessment. Increase in Underwriter Productivity - With this
technology, UW can process much more applications in
lesser time
How does this work?
Consistency in Decision making - The application
The Underwriter Co-Pilot is a cloud-based solution, which is
designed to be cloud-agnostic and can work with multiple provides standardized evaluations, ensuring decisions
proprietary and open-source LLMs available in the market. are based on uniform criteria
The key components involved in this solution are: Improved Accuracy - The ability to process vast

34 July 2024 The Insurance Times