Page 204 - India Insurance Report 2023- BIMTECH

192 India Insurance Report - Series II

Intelligence solutions in real-life situations.

The Typical Life Cycle includes:

a) Data Collection;
b) Data Preparation;

c) Exploration of Data;
d) Model Preparation;
e) Deploy the Model (IT Team Is Required to Carry This to Its Various Systems);
f) Monitor for Model Drift;
g) Retrain the Model.

Two Challenges are typically seen post the Model creation : Challenge 1 - How Soon can the IT
Team of the Company deploy the Model into the Core Insurance System or its allied systems? Challenge
2 : How to Manage Model Drift?

Challenge 1 : The IT departments are usually busy with implementing various other changes in the
system, and implementing a model developed by data science takes a lot of time. A reasonable estimate
for implementing the model on the IT Systems should be between 3-5 Months. In today’s competitive
world, this kind of time frame to get change done is largely unacceptable. Additionally, insurance core
systems can handle linear models well, but they cannot set up nonlinear pricing models or an ensemble
of models. Hence, there is space to accommodate a genre known as “Model Ops”. Increasingly, one will
see Insurance Companies move the “Model Ops” way.

What is Model Ops? Model Ops is built around the principle of DevOps, which means continuous
integration and continuous delivery are built into it. Imagine a Data science team builds a model and it
can be integrated into the core insurance without taking 3-6 Months. This is what Model Ops tries to
achieve. Once the initial setup is done, the model, once finalized by the data science team, can be
enabled on production in a few hours.

Challenge 2 : Model Drift. It refers to the degradation of model performance over time due to
changes in data and relationships. Model drift could be so sharp that it may render the original model
ineffective. Therefore, one has to govern the AI model very closely. Once a model drift has occurred,
the model needs to be tweaked to bring it back to optimum performance. Some of the technology
solutions today have built-in methods of identifying model drift and will alert the tech and data science
teams about the model drift, thereby preventing the Insurance company from making costly mistakes
and driving greater efficiency.

The benefits to customers include lower costs, and to an insurance company, it brings faster changes
in pricing and underwriting models.

2. Customer Data Platform (CDP) In Insurance

In an industry where retail products are largely standardized and little differentiation exists on the