The Industry Landscape
Subrogation Demands a Digital-first Approach

Insurance companies invest substantially in subrogation to recover funds and reduce financial impact. Yet, this process remains labor-intensive, requiring meticulous reviews of claims data and legal documents to determine liability and pursue recovery.

Technological advances now empower insurers to analyze large volumes of data including unstructured notes, adjuster comments and medical records using advanced ML and Natural Language Processing (NLP) to identify potential recovery opportunities and prioritize high-value claims for further investigation. Against this backdrop, insurers are partnering with data analytics and digital-first industry experts to swiftly deploy advanced AI / ML solutions, streamlining subrogation workflows for enhanced decision-making and financial outcomes.

The Client's Challenge
Unlocking Subrogation Value in an Unstructured Claims Ecosystem

The challenge for our client, a property and casualty insurance leader, was identifying recovery opportunities amid a deluge of manually processed claims. Key issues encompassed:

Fragmented processes, high data volumes and scattered operations across multiple vendors impeding seamless functioning and necessitating extensive coordination efforts

Unstructured data from numerous sources limiting visibility into critical information; manual data extraction and indexing processes added complexity and hindered timely decision-making

Manual and labor-intensive approach of subrogation handlers, resulting in missed subrogation opportunities and exacerbating operational inefficiencies and financial losses

Consequently, the insurer sought end-to-end digital transformation of its subrogation function.

The Solution
A Smart Model Combining Digital and Human Intelligence

Stepping in as an end-to-end data, analytics and AI partner, WNS Analytics devised a comprehensive approach, deploying our proprietary Subrogation-as-a-Service solution across the subrogation value chain. Our teams of domain and AI analytics experts came together to create a holistic solution, leveraging their combined expertise to design advanced insurance models. By integrating domain knowledge with cutting-edge AI, they addressed industry-specific challenges, ensuring compliance and improving operational efficiency.

The solution encompassed:

Key enablers of the solution included:

• A Rigorous Data Quality Framework to ensure the accuracy, completeness, timeliness, consistency and uniqueness of the data extracted from the insurer’s database. This involved implementing stringent data cleansing and validation processes

• An Effective Machine Learning Operations (MLOps) Framework to deploy and manage ML models across the lifecycle, following a classic Continuous Integration / Continuous Deployment (CI / CD) process for scalability and reliability

• An End-to-End Model Pipeline to extract, transform and score the claims data, facilitating seamless integration with the insurer’s systems and work󹀹ows for rapid processing and identification of recovery opportunities.

  The pipeline included:

  >Claims data extraction and contextualization model using advanced NLP algorithms

  >A recovery model to automate identifying recovery opportunities by integrating into the claims processing work󹀹ow

• An Application Programming Interface (API) Framework to access the subrogation prediction model; the API was integrated into the client's infrastructure, providing a seamless interface for utilizing the model's predictive capabilities through AWS Lambda

Recommended future enhancements will include implementation of SKENSE – our award-winning, proprietary, AI-led cognitive data capture and processing platform – for curating nuanced and actionable insights from unstructured documents for better decision-making.

Methodology

1. Data Analysis and Pre-processing

Conducted a comprehensive Exploratory Data Analysis (EDA) and applied pre-built text-processing algorithms to distill insights from unstructured data

2. Model Development

Leveraged a pre-built advanced ML algorithm and model evaluation metrics from the utility library to develop predictive models and deployed the best performer to identify potential recovery opportunities

3. Model Deployment

Deployed trained models using Amazon SageMaker as an API, with endpoints hosted on AWS, for seamless integration with the client’s claims processing workflow

4. Real-time Scoring

Automated daily data extraction from the claims center, uploading subrogation predictions to Amazon S3 and flagging them for handlers’ review, thus ensuring timely insights into potential recovery opportunities

5. Stakeholder Collaboration

Shared identified claims with subrogation handlers for review and action, facilitating faster decision-making and enhanced coordination among client teams

AWS components and services used for the solution:

Amazon S3 to store raw data, scored data and model objects

Amazon SageMaker to train, test and create the model (using Notebook Instances and SageMaker Studio)

SageMaker Endpoint and AWS Lambda to deploy the model and score claims

The Outcome
Predictable Recovery Value for Scalable Operations

The effective deployment of advanced analytics as part of Subrogation-as-a-Service led to significant improvements in claims recovery rates, indemnity benefits and operational efficiency for the insurer, with fewer false positives and manual interventions. Directly integrating with the claims center streamlined claims processing productivity, eliminating the need for manual interventions and accelerating the identification and processing of recovery opportunities.

Our outcome-based commercial arrangement on a “no win, no fee” basis ensured there was no upfront investment by the client. WNS further put skin in the game by underwriting the outcomes.

Benefits delivered