Claims process is a core part of operations for any insurance carrier; it’s one of the most customer-centric, critical, time consuming, labor and capital-intensive part of the business. It’s also the most vital part of customer journey within insurance lifecycle, it’s where carriers’ real purpose and resolve towards protecting it’s insured is tested. Most carriers are aware of this fact, therefore, It’s not surprising that carriers frequently make a considerable investment both in terms of effort and capital to bring advances to their claims process in order to improve customer experience while simultaneously reducing expenses and optimizing losses.
The challenge: Reassignments
One of the much ignored and critical areas of improvement is claims assignment to an appropriate adjuster. The adjuster assigned to the claim then has the ultimate responsibility of ensuring that the claim is settled quickly and accurately. There are some obvious factors like workload, availability, skill, and location, that go into the decision making process for adjuster assignment. The process of assignment is, however, far from being optimal, leading to reassignments which sometimes span across multiple adjusters. The resulting errors and delays often cost the carriers both in terms of accuracy of settlement and NPS (Net Promoter Score). Apart from that, lack of proper due diligence can lead to overpayments and fraud.
The dichotomy between workload balance and assigning the right adjuster is not easy to overcome. There have been attempts at using linear rule-based algorithms to match skill-sets and workload using initial claim triage. However, in the absence of real insights on how a claim may pan out, as the claims moves through various stages in its lifecycle and new information is received, the rules may point to a different adjuster during this journey, consequently, the claim may change hands several times before being assigned to the right adjuster.
A Deep Learning Based Approach
The key question is, can deep learning provide predictive insights to enable better adjuster assignment? Using deep learning, it is possible today to train an algorithm to use historical claims data to accurately predict the journey a claim might take before being settled. Predictions like complexity of claim, requirement of specialized services, current caseload, and access to vendors and third party experts can help stitch together a clearer picture of what path may result in settling the claim faster. The insights on the claim can then be matched with appropriate adjuster skills and experience based on history of similar patterns and types of claims to assign the case appropriately. One of most compelling advantages of this approach is that predictions are in near real-time and become increasingly accurate over time (Reinforcement Learning). In other words, the system is dynamic and self-learning and can adjust changes in claims complexity, adjuster skills and other factors over time.
Here are two extremely simple approaches that can be used to increase accuracy in adjuster assignment. In the first approach, deep learning can be used to segment exposures into three distinct segments based on predicted payout; no payment, under $1000 payment, and more than $1000 payment. This exposure level segment then can be used to determine the adjuster/group that’s appropriate for the level of payout.
In the second approach, a deep learning based model can be trained on all historical claims to predict duration, payout, and number of reassignments for an exposure. This exposure level information then can be fed into an inference engine to determine the appropriate adjuster/group that the exposure can be assigned to.
In additions, the accuracy of these deep learning models can be significantly boosted by training them on not just the structured claims transaction data, but also unstructured notes/text and image data, which cannot easily be analyzed by rule based algorithms or manually, deep learning’s ability to find meaning from this unstructured data can provide significant new insights.
Using a deep learning based approach to enable adjuster assignment can add significant speed and accuracy to the process. The approach is dynamic and can adjust to changing nature of claims complexity and adjuster skills to learn, adjust and make increasingly better predictions and reduce reassignments.
Accurate and appropriate specialized adjuster assignment help in scrutinizing potential cases of Fraud & Subrogation, helping lower loss ratios and consequently lowering overall premiums paid by customers. Streamlining claims assignment process not only helps optimize expenses and losses, but also has a significant impact on NPS. With a third of insurance customers always shopping around for alternatives, customer experience becomes the critical factor in retention and reducing churn.