Hello everyone. My name is Samis Joshi. I'm working with Cognizant as a senior, a technical architect on Duck Creek Technology Cloud platforms. I'll Today we're walking you through how robust ML ops practices are transforming insurance operations. And then over the next few slides, I'll show you how AI driven strategies are delivering up to 65% efficiency gains and reducing. Processing times by as much as 99% across insurance functions. So let's dive in. So here's our agenda for today. First, we'll look at the broader AI transformation in insurance. Then we'll dive into ML Lops architecture and examine specific technical deep dives like modeling computer vision and telematics. Next we'll touch on compliance and biases detect, followed by an implementation roadmap and ROI analysis. Finally, I'll summarize the key takeaways and share the actionable next steps. Let's start. Insurance as an industry has come a long way in AI adoption. Before ML ops only about 18% of the carriers are experimenting with AI. Today, thanks to ML ops that option has grown to 54%. The results are tangible 65% higher operational efficiency, and a 60% improvement in underwriting accuracy. This isn't just about automation guys. It's about enabling cons, consistent data driven decisions that scale. Let's rewind and see why transformation was needed. Traditionally, policy applications involve paperwork and manual data entry. Underwriting was often subjective and policy issuance took three to four days. Claims assessments could stretch for weeks, and risk evaluation was limited to annual reviews. The net effect was higher cost. Long cycle times and inconsistent customer experiences. So now here comes into picture the ML ops. So ML lops brings order and automation. So here we are talking about continuous integration and deployment pipelines that cut the deployment cycles by 78%. Feature store architectures, architectures that reduce feature engineering time by 67%. And model registries that ensure traceability and compliance. This integrated architecture allows insurers to respond quickly to market and regulatory changes. That's interesting. Proceeding with transformative transf transformative results on the policy issuance. So consider policy issuance. So traditionally it took three to four days. With ML lops, it's down to just 15%. Now with mops. So that 99% re time reduction in time while maintaining a 93.6% accuracy rate. The key lies in the automated risk assessments, document processing and continuous monitoring that ensures model remains accurate even as risk profiles evolve. Proceeding to next step comes the catastrophe modeling. That's a very important point in here. Catastrophe modeling is critical for insurers but traditional approaches relied on static data and quarterly updates with machine learning operations model in just real time weather data, satellite image imagery, and even assemble approaches that mix physics and machine learning together. The outcome. It's a 31% boost in the flood prediction accuracy, nearly 30% better cyclonic forecast, and a 43% improvement in claims readiness. And that's a very good number. If you pro, if you look into the computer vision. In case of the property and casualty insurance let's start on this point. In this case, if you see computer vision has ref re redefined the property inspections with drones running Edge, deploy on the running on the edge deployed models, insurers can capture pre-pro, insurers can capture pre-process and assess property damage in real time. Accuracy has reached 89% and inspections that once took days are now completed in minute. Importantly, champion challenger frameworks. Ensure models are continuously valued and improved proceeding with the telematics. This basically brings the real time example where I'll tell you how telematics is helping in assessing the real time driver risk assessment. Driver behavior is another area that is being shaped by ai. Telematics use uses OBD two. Which is on onboard diagnostic devices and smartphones to capture over 300 driving features. Kafka based streaming pipelines processes, terabytes of data daily and containerized inference. Inference engines adjust seamlessly to traffic loads compared to traditional demographic based risk assessments. Predictive power has improved by 83%, and that's again, a very good number now proceeding to regulatory compliance and bias distinction. So no AI transformation is complete without addressing fairness and compliance. So automated adversarial testing detects nearly 80% of potential bias before deployment. Model validation pipelines, automate documentation and regulate regulatory reporting cutting compliance cycles by 67%. So techniques like lime, LIME, local interpretable model, agnostic explanations, and SHAP shapely edit additive explanations. These enhance explainability, which is crucial in market regulated like insurance now proceeding with the. How the architecture looks like the integrated data architecture. So to achieve accurate models, insurers must unify data. So here eight to 12 data sources come together, like customer data, property details, IOTs, telematics, imports, credit scores, even alternative signals like satellite imaginary imagery. So by integrating these sources. Through automated feature engineering predicting accuracy improves by more than 36%. Now, if we come to the key components for the MLF implementation, ML LOPS implementation there are four components that make it make this possible. Automated deployment pipelines using GitHubs. Monitoring frameworks for real time drift detection experimentation platforms for AB testing and centralized feature stores with versioning Together, these form a mature ML ops ecosystem that balances speed, accuracy and compliance. So now if we look into the implementation roadmap, so transformation doesn't happen overnight. So phase one is building the foundation registries, versioning, and pipelines over three to four months it, it takes to build all these things. Now, phase two, operationalize. Workflows, like automated training and monitoring. The typically, this typically takes two to three months. And then phase three, that adds optimization, automated retraining, AB testing, and compliance automation over three to four months. So within a, within less than a year, insurancers, insurers can go from POC to production, and that's a very good timeframe to achieve this. So next we'll proceed with the ROI analysis. The results speak for themselves, which on my slide. Before ML ops model deployment, claims processing and process and policy issuance times are much higher after adoption carriers. Certain big carriers saw improvements within nine to 12 months, and these aren't just marginal gains. They are game changers that reduce operational costs and scale as new models come online. Proceeding to the key takeaways. First is ML lops is driving a 300% growth in AI adoption across insurance. Secondly, end-to-end automation insurers accuracy above 93% while enabling rapid iter. Third, compliance and fairness frameworks cut regulatory cycles by 67%. For your next steps assess your current ML maturity and pri prioritize high value use cases, establish foundational components like registries and feature tools, and integrate compliance from day one. Doing this as it was mentioned, we can go from POC to production. In within nine to 12 months. Thank you guys. Thank you for your time and attention. I hope this session gave you a big picture and technical depth on how ML Ops is revolutionizing the insurance operations. Thank you and good day.