Deploying Deep Learning in Claims Reserving

By February 6, 2019 February 17th, 2020 Blog
Deep Learning in claims reserving

P&C insurers are realizing the value of leveraging Deep Learning to improve segmentation and listing of risks. Promising statistics and research, points to the success of deploying supervised and unsupervised Machine Learning for multiple insurance use cases. There is definitely a high potential and power by introducing sophisticated automation and artificial intelligence for Claims Reserving, which would enable rethinking the analysis of claims liability.

Understanding Claims Reserving

When a loss occurs related to property and casualty or when multiple parties or properties are involved, the traditional method used was to allocate a claims reserve amount. This Amount is flag marked for eventual claim payments. RBNS (Reported but Not Settled) & IBNR (Incurred But Not Reported) Reserves are other reserves used for Potential future claims. Hence, Reserve is represented as a liability in the balance sheet.

Problem with Rule based Claim Reserving

Traditional methods of setting aside a specific amount for claims reserve can prove to be problematic and might result in over reserving or under reserving of allocated funds for compensating the claimants. Rule based Claims Reserves are determined through the data points listed:

  • Closed Claims/ Historical data and information.
  • Observations and assumptions
  • Creation of models
  • Statistics and estimates

Insurance Underwriters have been using algorithms such as average costs per claim, bootstrap, Over-dispersed Poisson, separation method and so on to arrive at claim calculations.

However, not all such methods are as effective as using input data and patterns using Deep Learning.

Deep Learning for Claims Reserving

In the competitive world of insurance business, companies are realizing the intense need for providing improved standards of services and the importance of building trust and reliability as well as avoiding pitfalls. Insurance businesses can enhance methods of forecasting claims and enhance customer experience through using Deep Learning for claims reserving processes.

Deep Learning offers new ways of analytics to study data and for extracting models and patterns that are far more accurate and time saving. P&C insurers can improve decision making related to claims reserving, leveraging the features used for finding potential payouts through models created on the machine learning platforms. 

Deep Learning Approach and Benefits

Loss estimates are computed by using statistical case estimates. Automated processes could use data pointers as described below and free up the bandwidth of claim adjusters. This method uses best of both traditional reserving techniques and enhanced deep learning case estimation algorithms. In addition, the use of Age-Period-Cohort (APC) algorithms will help to reflect better trends and correlations across multiple factors like underwriting companies, accident years, products and segments.

For example, a BI (bodily injury) claim uses historical claims data, BI reports information, Adjustor notes, and images from the claim files and so on to derive the approximate reserve. Machine learning algorithmic methods can leverage inputs from other models such as Segmentation, Subrogation & Fraud. Additionally, it would use factors such as claimant behavior, event scenario & demographics.

Additional Parameters to consider are nature & severity of injuries, age, current earnings, costs for care and proximity of the claimant’s residence to healthcare providers / medical specialists, etc. Machine Learning models will leverage such data points more effectively & yield better benefits.

  • The benefits derived through Deep Learning are high especially in claims, which have greater uncertainty such as litigation potential.
  • End-to-end processing and claims reserving will be faster with lesser turnaround time.
  • Up to 20% of the claims can have a near real-time reserving
  • Adjustment of reserving through reinforcement learning
  • Aggregated Reserving for at different grains of Agent, LOB, Product, State & Segments, etc.
  • Manual effort could be reduced up to 15% through the usage of Cognitive RPA

Deep Learning helps to arrive at IEULRs (Initial Expected Ultimate Loss Ratios) combined with human intelligence. This helps classification / segmentation of BI Claims V/S Non BI Claims and historical analysis of reserve ration against actuals.

Deep Learning has the capability to offer multiple opportunities for P&C insurers to the better augment, particularly in Claims reserving. Data that can be omitted or overlooked through human error can be analyzed in detail using the tools of artificial intelligence and help reduce costs/liability for the insurance company. Without ignoring the traditional reserving methods, Claims reserving process will definitely be augmented using AI methods. Modern Technology would be the major factor influencing this change.

Citations: 1. 2.

Gopal Swaminathan

Gopal Swaminathan

Gopal is a dynamic leader with 18 years of Data & Analytics experience, enabling business outcomes through information insights. Since 2008, his focus has been P&C Insurance companies. Therein he drives ROI maximization and enhancing customer experience using Advanced Analytics, Artificial Intelligence, and Machine Learning solutions. As a Director of Products, he spearheads the Product Development for Insurance Analytics. Additionally, he is responsible for Business Development of existing and new accounts in the Greater Phoenix Area.Read More Posts


  • Amey Kulkarni says:

    Great Blog Sir. In the industry like insurance where insurers don’t know the upfront cost for their service and have to rely on historical data and expert judgement to offer a sustainable price to settle down the claims, implementation and evolution of deep learning can certainly give an upper hand in the competitive market.

  • Sridhar says:

    Excellent Analytical work and deductions – keep it going Gopal!

Leave a Reply