Property insurance claims have a great variance in type and complexity. Whether it is for residential or commercial property, the characteristics of the structure, its contents, and the areas surrounding that structure make each individual property unique in its exposure profile. As in other types of claims, there are many claims that would be classified as simple; more common claims that can (or should) be processed quickly and efficiently. However, there are many claims that require more in-depth analysis and restoration/remediation actions. Even low severity claims that, on the surface, appear simple yet may spawn more complex elements that sometimes lead to unexpectedly large reserves. For example, there may be fraud involved, or the claimant may be dissatisfied with the carrier’s response and engage an attorney. Furthermore, many property claims may be as a result of negligence or liability by a third party, creating subrogation opportunities that enable the carrier to recover part or all of the claim payment(s). Overall, it is imperative that expert knowledge and experience is required for the investigation and estimation of claims. Since there is a significant exodus of seasoned professionals underway, the new opportunity is to turn to data and analytics to help address the skill gap.
Virtually every claim – especially those with the complexities enumerated here – has a range of unstructured data that has been submitted or collected to aid the claims adjusting process. This provides an excellent opportunity to apply solutions based on AI technologies. The unstructured data may be in the form of an email, imaged form, adjuster notes in a claim administration system, spreadsheet, voice transcript from a recorded statement or many other sources. It may represent loss descriptions, details of incident circumstances, cost estimates, expert reports, repair offers, claimant requests and many other aspects of the claim. The great leap forward in natural language processing (NLP) technology and machine learning (ML) now enables purpose-built solutions for property claims that provide significant business results in a variety of specific use cases. For many of these use cases, the great potential is in identifying opportunities to take action earlier in the lifecycle – even during the initial stages of first notice of loss (FNOL).
The initial reporting of a claim will, typically, include the collection of both structured and unstructured data. Often, the call to a claims center, reporting via a website, or submission through an agent portal will include a focus on the capture of the basic information in structured data fields. However, that is usually followed by the collection of other information to support the claim, such as pictures of the damage, recorded statements, or other supporting documentation. Most of this additional information tends to be in various unstructured data formats. Converting the unstructured data into structured formats, and combining it with other data, provides a rich set of information for analysis and provides more context for the claim. Some illustrative examples follow:
- FNOL triage: One of the most important early applications for AI and analytics is for triage.. The first triage pass determines if the claim can be handled in subsequent steps via automation (in straight-through-processing mode) or needs to be referred to an adjuster. If its complexity requires human expertise, the next triage task is to determine the nature of the claim and then match it to the adjuster with the appropriate skill and available workload. Accurately identifying the best adjuster specialist at the beginning requires a granular understanding of the claim. Today, most claims triage for property claims is handled in manual mode with a claim supervisor assessing claims to determine where to route them to for next steps and actions.
- Subrogation insights: Property accidents or incidents may have occurred based on product defects or malfunctions, or improper installation by contractors or repair professionals. The insurers first responsibility is to fulfill the insurance contract and make the policyholder “whole” again – either through repair/replacement of damaged property or financial reimbursement for the property. However, there are often opportunities to recover damages from an at-fault party. Analysis of the wealth of data collected during FNOL, and subsequent investigation, enables carriers to identify these opportunities and begin actions for financial recovery from the third party. The determination of liability is often complicated as in the example of water damage in an apartment building. Determining the origin of the damage and the party at fault often creates subrogation opportunities but the determination is not always straightforward. Once again, analyzing unstructured data to support earlier identification enables insurers to better prepare, and build a case in pursuing a successful recovery.
- Investigation: Claim adjusters often have heavy caseloads, mounting pressure to settle claims more rapidly and major backlogs of claims. It is well known that the speed of settling a claim is directly correlated to the ultimate settlement amount – faster resolution means lower claim costs and happier customers. Complex property claims such as significant fire or hurricane damage to a residential property, or the major flooding of a commercial property, may require months to fully investigate and ultimately settle. In complex cases – and even simpler claims – it is very useful to analyze all the data related to claims to assist in workload prioritization, as well as providing new information and insights that aid in estimation.
Insurers are already reaping the benefits of NLP and ML-based solutions for claims for every line of business – including both personal and commercial lines property claims. Every P&C insurance company that writes property insurance should be evaluating and implementing advanced AI solutions to harness new insights from unstructured data in the property claims process.
