Creating a Single Source of Truth Using Azure
Trianz was engaged by a reinsurance company to help with the development of predictive analytics applications. As part of its business, the company acquires existing books of reinsurance policies and manages claims associated with them, and they needed analytics capabilities in order to manage policy claims more effectively.
Business Objectives with Predictive Analytics
By developing a predictive analytics solution, the objective for the reinsurance company was to enable its legal team to detect fraudulent claims more effectively. The reinsurance firm also needed a way to predict claim settlement amounts before dues are paid, assisting with financial planning and cashflow.
Finally, a new alerting system was required to proactively alert employees upon receipt of high-risk or high-value claims, ones which would require a significant cash windfall for full settlement.
Data sources were mixed and siloed on the IT network. The company shared information using large documents with attachments and non-standardized file formats, creating an accessibility challenge. The reinsurance firm had no data warehouse to act as a single-source-of-truth (SSOT), leading to data credibility concerns and technical bandwidth limitations during data analysis.
Most critical were the manual data input workflows, where hundreds of documents and files were manually uploaded to the system to be analyzed. No automated ingestion method existed on the reinsurance firm’s IT network.
Technology Components Highlighted
Trianz worked closely with the reinsurance provider to identify suitable technology components for its predictive analytics project:
Azure Data Lake was chosen to act as a data repository and SSOT from which all IT resources could reference and verify the validity of data.
Azure Data Factory was chosen as an extract-transform-load pipeline solution operating on an economical pay-as-you-go model.
Databricks was chosen as a data management platform using the Lakehouse architecture for simpler data infrastructure operations.
Azure Data Catalog was selected to store metadata relating to data sources and simplify data asset discovery.
Azure DevOps was used to orchestrate agile software development, create and manage continuous integration and deployment (CI/CD) pipelines, and facilitate source control.
Python is the coding language used to build the predictive analytics algorithm model for the reinsurance firm.
R-Shiny is an open-source, code-first data science platform that helped with deployment of data-based web applications, referred to as Shiny Web Apps.
Docker was used to host serverless containerized applications and resources in tandem with Kubernetes.
Spark Natural Language Processing (NLP) was adopted to annotate conversational audio and text files with relevant metadata using machine learning (ML) and Python.
DataRobot Prediction Server sits in a Docker container acting as an execution environment for DataRobot model packages (.mlpkg files).
Word2vec is a technique for NLP. It consists of an algorithmic neural network that learns word associations to create meaning using cosine similarities between vectors.
Jira was adopted to track issues and keep projects on track for DevOps and data science teams, using agile development methodologies.
Enabling Predictive Analytics and Reducing Manual Workflows
Trianz worked with the reinsurance company to recruit a team of cloud architects, data scientists, data architects, and data engineers. The team consisted of native and overseas personnel in different time zones, meaning a point of contact was always available for the reinsurance company.
As part of the roadmap, Microsoft Azure technology components were developed first, forming the data science foundation for applications and datasets. The predictive analytics application was enhanced to include entity recognition, NLP, custom analytics models, and robotic process automation (RPA). Finally, a cloud-based data mart was established with production-ready data models and processes to support data querying operations, analytical reports, and data monetization initiatives.
The Agile development methodology was customized to suit the reinsurance company’s needs. This supported data pipeline development, predictive analytics modelling, and CI/CD deployment and testing during the development phase.
Transformational Effects of New Data Mart and Analytics
After adopting the new technologies, the reinsurance firm experienced tangible transformational effects:
RPA saved time and effort for the data science team by consistently and effectively automating previously manual processes and workflows.
New data models enhanced scoring capabilities to improve predictive forecasting, mitigating risk, and smoothing cashflow when dealing with reinsurance claims.
Entity explorer graphs were used to identify people and create relations between people, places, and business datasets. The purpose was to detect fraud and mitigate risk to the business.
NLP can now drill through phone conversations, email conversations, and other datasets to proactively alert legal teams when claim numbers increase sharply and suddenly, which also serves to mitigate risk.
Contact Us Today
Contact Us Today
