First notice of loss (FNOL) documents are a mess to sort through, especially if they include handwritten reports. Your insurance adjusters need copies of the reports for claims processing, and difficult to process FNOL documents can quickly become a bottleneck
It takes time for adjusters to manually input information from the reports into your computer system, and issues inevitably come up. What if there is handwriting in the report that is hard to read? What if the report has been distorted in some way? What if a mistake is made as the information from a handwritten form is transcribed to your system?
This article explains how AI-powered unstructured data processing (UDP) can be used to quickly automate extraction of key information from FNOL documents at scale with guaranteed accuracy. By leveraging artificial intelligence (AI) to streamline FNOL document processing, insurance providers can:
The first notice of loss (FNOL), also called the first notification of loss, is typically the first step in the claims process flow. It is the initial report that is sent to insurance companies after an insured person initiates contact about loss, theft, or damage to an insured asset. As part of the FNOL process, insurers will ask for specific information such as:
With the ability to purchase and renew insurance policies online, without interacting with a person, the first notice of loss may be the first time someone makes direct contact with their insurance provider. Beyond initiating insurance claims, the FNOL process may be the first real impression of their insurance company people get. Because people are typically distraught after a loss, it is that much more critical that the FNOL experience is positive, efficient, and smooth.
Insurance adjusters rely on FNOL documents to process claims. Errors and omissions due to inaccurate, or difficult to extract, information can lead to delayed processing, increased costs, and disappointed customers. For example, a handwritten section of an incident report that says, “Hit my hand on the dashboard,” might be recorded as “Hit my head on the dashboard” because the handwriting is difficult to read, or the adjuster simply made a mistake.
For an insurance company processing a claim, heads and hands are obviously very different. Clearly this will have an impact on the claims process itself, but it also has the potential to be off-putting to a customer should they ever catch wind of the error.
Errors in FNOL documents at intake cascade further along in the claims processing flow.
Sticking with the same example, if an adjuster files a claim against the party responsible for a car accident as resulting in an injured head, when it was really an injured hand, the insurance company may deny claims for hospital bills related to treating the hand injury because they mistakenly believe it’s unrelated.
The original adjuster then has to go in and fix the problem, taking more time and costing more money. Meanwhile, your customer is frustrated because they aren’t being properly (and easily) reimbursed in the aftermath of an accident that wasn’t their fault to begin with.
Inaccurate FNOL forms and supporting documents lead to slower response times, claims processing delays, and poor customer service. According to Accenture, 41% of people who submit an insurance claim will leave their insurance provider within a year. Additionally, customers who submitted a claim in the past 24 months are nearly twice as likely to change insurance companies within 12 months compared to those who have not.
JD Power says the number one reason why people change auto insurance carriers is due to poor customer service on the part of the provider. This all underscores one simple point: to provide a positive customer experience, claims must be accurate from the start.
In addition to benefiting customer satisfaction and retention, accurate FNOLs also save on expenses. When intake forms are accurate from the beginning, adjusters don’t have to spend time identifying and correcting mistakes. Additionally, inaccuracies can lead to litigation that can be time consuming and expensive. Easily prevent
If the initial errors propagate, it becomes more expensive later on in the process because lawyers may have to litigate a case based on the false information that was initially taken down.
Not only that, the difference between a head and a hand injury can cost insurance companies tens of thousands of dollars in a settlement if the error never gets corrected.
The adjuster would also need to see initial documents from all parties involved in the loss, including reports made by investigating authorities (if any). This may include reports from police departments, fire departments, security agencies, park rangers, and other entities. Additionally, any other individuals involved in the incident will have to submit forms as well, such as contractors and witnesses.
These supporting documents might be jumbled, handwritten, or haphazard in how they present information. From the structure and formatting of the documents, to the types of data they contain (e.g., text, image, tables, etc.), there is a high amount of variability, and a lack of discernible structure, from claim to claim. Although humans are highly capable when it comes to interpreting unstructured information, they are also slow, error prone, and expensive. Unfortunately, most traditional document processing solutions such as intelligent document processing (IDP) or robotic process automation (RPA) paired with optical character recognition (OCR) struggle with semi-structured and unstructured documents.
All of these data points come in two types: structured and unstructured.
Think of structured data as if you’re filling out a form with a drop-down menu. There are certain limitations to what these data points can be, and there is a set structure to them. Unstructured data has none of these limitations, as it isn't derived from a form fill or database with set fields.
For example, a customer texting information about a loss is an example of unstructured data. This communication contains useful information, but it isn’t easily understood by most software. Can your CRM take a text message, filter it into a database, then reach conclusions about next steps in the claims process?
Police reports are another good example of unstructured data, as they do not always contain fields that have been standardized across states or even between cities within the same state. Additionally, the information within each field can be unstructured as well, such as the description of an event, or formatted variably, such as state names (CA, California, etc.) or incident codes.
With software, algorithms, AI and machine learning tools, computer programs are getting much more intelligent at determining what data should go where. A good example of this is the voice recognition technology that powers automated customer service lines. The really good systems can identify voice inflections, recognize high- and low-pitched tones, and extract keywords in a phrase that don’t necessarily match a response option.
The same thing works for written data and images in a computer system.
AI can look at a scanned document and parse the data it contains in seconds with 99% accuracy. The program can then route information to the right people or repository by understanding the data using artificial intelligence. Super.AI’s Unstructured Data Processing (UDP) Platform combines multiple AI/ML models trained on different data types to parse through texts, emails, chat logs, documents, phone calls, and any other unstructured information. This makes it possible to quickly process any FNOL or supporting documentation that may require ingestion and analysis during the claim processing flow.
Best of all, the platform integrates with the tools you’re already invested in and are familiar with. Thanks to no-code AI, you don’t need to have a machine learning degree to understand how to use it. Any business user can build, test, and deploy AI applications in a matter of days.
Super.AI’s UDP Platform can process data from any source, regardless of formatting, and give it structure so that it can be efficiently extracted, routed, analyzed, and stored.
Imagine the aftermath of an automobile accident. A claimant documents the incident by taking a video on their smartphone after determining that everyone is okay. Later, another driver involved in the accident asks for the claimant’s phone number, then sends a text message that includes relevant personal information and a photo of their insurance card. The claimant also takes still photos of the damage, and sends this, along with all the other unstructured data, to their insurance provider.
Unstructured data processing makes it possible to automatically:
How much time would it take a human to process and analyze this data? How likely are they to make mistakes when repeating this process multiple times in a row? Having all of this happen automatically speeds up the claims process and gets your customer driving again very quickly. Additionally, Super.AI’s UDP Platform has reimagined human-in-the-loop capabilities that make it possible to leverage human expertise to review data output, which in turn trains the AI model so that it performs better in the future.
Through AI-powered data processing and analysis, as well as continuous learning powered by human supervision, it is possible to process any document that may be submitted as part of a first notice of loss, or at any point during the insurance claims process.
To learn more about using artificial intelligence to automate complex processes and improve how insurance organizations operate, check out the following resources: