Back To Top

Property and Casualty Subrogation Analytics


The property and casualty (P&C) insurance industry is being rapidly transformed by the advent of AI. Improved computing power in the cloud, cutting-edge algorithms, access to massive amounts of data, and tools that make implementation easier are some of the key tenets that are helping insurers achieve lowered costs. By adopting AI-powered digital systems, data is collected, aggregated and analyzed.  As more data is collected over time, the AI systems learn the rules of the business. This helps the AI system improvise the involved processes and helps optimize them. 

AI, argues a recent report from Allianz, will transform the P&C insurance industry in an unforeseen way — such as including adjustments to the way risks are underwritten and how coverages are understood will evolve. Assuming complete adoption of AI across all industries reviewed, the report claims AI is expected to increase corporate profitability in 16 industries across 12 economies by an average of 38 percent.

In the context of the P&C insurance industry, subrogation means that an insurer can legally pursue any third party that caused a loss to the insured. This is done to recover the amount of the claim paid by the insurer to the insured, typically, for the losses incurred. Hence, any amount recovered through this process goes directly towards the insurance company’s bottom line. Thus, a clear understanding of subrogation and the potential to incur losses or avoid them is important because it helps recover payments related to any losses and reduce the overall loss amount. This also reflects the company’s performance and can help the insured (or the policyholders) in reducing their premium outgo.


First, the P&C industry has had a precedent of using paper-based systems and manual methods while applying for property or casualty insurance or filing a claim. Slow response times from customer service and negotiating reams of paperwork were common, making the penetration, implementation and deployment of digital processes extremely challenging. Similarly, a paradigm shift in the culture, making the transition from legacy systems in a seamless way while retaining the huge volumes of data, and training has also been a challenge.

Operators in the P&C industry and insurers are part of an industry that remains highly regulated. Owing to changes in the regulations, processes also need to be updated which entails a lot of effort and time. So, even if the subrogation potential is identified, by the time the claims are referred to the counsel, the limitations in the process and regulations can become a hurdle for a deadline on the horizon. There have been subrogation potential opportunities which have been given up to avoid incurring the subrogation attorney’s fees. Hence, when the claim is sent to the counsel, there is very little time left for investigations and processing with the third party. Considering that it is an investment in time and money, the claims are prioritized based on the dollar amount in question, resulting in missed revenue.

Insurers in the P&C industry process numerous claims that comprises several structured data fields, such as a unique claim number, date of the loss, and amount of the original settlement. However, key facts about each incident that affect subrogation potential are usually added as free text notes to the claim and related documentation. These notes can comprise police reports, witness statements, adjuster notes, recorded statements, medical records, and other relevant information that are usually modified by different people. Adjusters perform numerous tasks for each claim, including interviewing all the parties involved in the incident, analyzing police reports, evaluating damages, and negotiating claim settlement. After these procedures are complete, identifying potential subrogation opportunities follow. This requires coherent analysis of all notes related to the claim at once. 

As claim notes get added over time, it becomes increasingly difficult for adjusters to analyze the entire claim every time there is a new update. In addition, adjusters frequently need to handle multiple claims simultaneously, making the task even more challenging. Identifying an opportunity for subrogation in this manner also allows room for errors in judgement or lack of motivation. Often, adjusters miss the liability exposure of third party. As the fault or the individual responsible for the fault is not determined immediately, the final decision about who will pay has to wait until the investigation is complete.The investigations can also prove lengthy and costly, with the result that many insurance companies neglect to pursue their subrogation right. 

Thus identifying potential for subrogation can become very tedious and expensive if it is not automated. Missed opportunities for subrogation translate into loss of revenue  and affects profitability. It is estimated that the insurance industry misses subrogation opportunities worth about $15 billion each year.


At Abzooba, we automated the subrogation potential for a leading end-to-end payment integrity organization operating in the healthcare insurance domain. is a Auto-ML AI Ops platform that excels in data collection to inference generation.’s Auto-ML AI Ops framework allows automating variable selection, data partitioning, model performance testing, model tuning and so on instead of manual coding. 

The Auto-ML framework made available by leverages the latest machine learning and Deep Learning tools while preparing prediction models. We studied  available claim documentation, primarily, claim notes from adjusters to understand the context of the problem such as what is damaged, what caused the damage and who is responsible for the damage. By using claims processing solution, we extracted key information from the adjuster claim notes and used them as predictors for identifying subrogation opportunities. 

Data Collection and Analysis

A major part of the data transformation journey while creating models involved setting up the required infrastructure and collecting raw, continuous data (unformatted, unparsed) from insurance databases, specifically, adjuster notes and claim documentation. This information is obtained from adjuster notes that were made available in a digital format.These notes include vectors embedded into a Document Vector (DV) for easy identification and processing. Every time the adjuster claim notes are updated, the claim is scored again for its subrogation potential.

Data Cleansing addressed data cleansing and also enabled exploratory data and statistical analysis through functions provided as part of the Python library stack. This helped to quickly classify the data and categorize it based on attributes.  Along with the deep linguistic analysis capabilities offered by, data analysts were able to accurately extract 700+ attributes covering key information about each claim, such as type of collision, vehicle point of impact, driver actions, liability and injuries of all parties, and many more.

Data Preparation can read factors from a varied recommendation text connection and generate an output. This was supplanted with additional data collected (with the aid of libraries) while preparing a predictive model. The DV was further used to determine subrogation potential which continuously identifies and learns new trends on existing and new data-sets thereby increasing the engine accuracy. This predictive model provides a subrogation-likelihood score, key reasons for making the claim subrogatable, and the expected amount of recovery.

Model Review

However, although a model may be built as intended, it might produce inaccurate outputs when compared to its design objective and intended use or a model may be used incorrectly or inappropriately, or its limitations or assumptions may not be fully understood. To address these possible risks, the model development and implementation were aligned with industry best practices and strict quality control.’s Auto-ML AI Ops framework was leveraged to collect and analyse this extensive data repository and analyse details. 

Before these models were deployed to production, they were independently reviewed. This ensured that their behavior was aligned with their design objectives and business uses. From all these variables obtained, models were created. The data versioning and connectivity libraries allowed keeping track of the data sets as incremental changes were made to suit the models. These models were also trained and evaluated with both standard and custom metrics to provide an effective challenge to each models’ development, implementation, and use.’s Auto ML-AI Ops framework also allowed quick reproduction of the model development process, thus enabling model validators monitor and review the model and its potential limitations more closely. 

Inference Generation

Based on the models generated by, the organization was able to identify subrogation opportunities based on the analysis of historical claim records. It enabled the organization to make informed decisions on the cases to pursue subrogation first in order to minimize effort and maximize returns. Additionally, since the solution predicts claims with subrogation potential and presents key claim facts in a tabular format —  easily accessed via web reports — claim adjusters can efficiently analyze claims. 

Early identification of subrogation opportunities also means a higher probability of payment recovery, quicker settlement with the third party, ability to plan for and maintain optimal levels of reserve funds to cover anticipated payouts and reduction of inadequate reserves.  This, in turn, improves revenue and profitability. 

Speak to AI Expert