Patent Issued for Claims process assistance using models (USPTO 11972489): State Farm Mutual Automobile Insurance Company

2024 MAY 17 (NewsRx) -- By a News Reporter-Staff News Editor at Insurance Daily News -- From Alexandria, Virginia, NewsRx journalists report that a patent by the inventors Garapati, Sunitha (Normal, IL, US), Gummadi, Surendranath (Cantrall, IL, US), Husarik, Timothy John (Normal, IL, US), Swingler, Matthew Thomas (Normal, IL, US), Walder, Bradley J (Congerville, IL, US), filed on April 23, 2021, was published online on April 30, 2024.

The patent’s assignee for patent number 11972489 is State Farm Mutual Automobile Insurance Company (Bloomington, Illinois, United States).

News editors obtained the following quote from the background information supplied by the inventors: “Traditionally, insurance claims were handled by insurance adjusters walking customers through the entire claims process, including assisting with documentation gathering. However, in today’s digital world, customers embracing smart technologies may prefer to expedite portions of the process. For instance, instead of scheduling and waiting for an adjuster to travel and take photos of the damaged car, most customers have smartphones that are capable of taking high resolution photos and videos. These customers can take photos and videos of not only the vehicle damage, but also of the accident scene right after the incident, when details are fresh. With the ability to quickly send and receive data, e.g., via a smartphone, customer expectations of timeliness in claims processing have similarly increased. Accordingly, there is a need for a smart claims processing system that could assist customers and other users with documentation gathering and sharing to promote timeliness.”

As a supplement to the background information on this patent, NewsRx correspondents also obtained the inventors’ summary information for this patent: “This disclosure is directed to a claims processing system including a claims assistance component that leverages machine learning models to classify documents and determine the workflow to advance the claims process. The documents may be received as unstructured data from one or more user devices and/or other sources. Initially, the unstructured data uploaded from user devices, emails, and other sources may be received by a server (“cloud server” or “the cloud”) associated with an insurance service provider. The unstructured data may be manually classified and tagged by insurance categories. The tagged data (“labeled data” or “augmented data”) may be used as training data. In some examples, the system may use the training data to train one or more machine learning (ML) model(s) to classify input data into insurance categories. The system may determine the workflow based on the insurance categories.

“In some examples, the system may deploy the claims assistance component and/or trained model(s) to user devices to assist with document processing at the device level. The user devices may access the claims assistance component to upload documents for a particular claim to the servers. In various examples, the claims assistance component may include a user portal for a user to upload the documents and/or view documents that have been uploaded for the particular claim. Before the upload, the claims assistance component may use the trained model(s) to classify the documents by insurance categories and/or verify the data content to avoid uploading documents of questionable quality. The trained model(s) may also determine associated confidence levels for the classification. If the confidence level is low or below a threshold, the system may generate a prompt for the user to verify the classification. If the confidence level meets or satisfies a threshold, the data may be uploaded to the cloud servers for additional processing. Once uploaded, the system may perform advance processing techniques to analyze the content of the data to extract relevant information that may be used by an agent to complete the claim. For brevity, an “agent” refers to any user (“actor”) assigned by the insurance service provider to handle, review, and/or process the claim. If additional information and/or documentation for a claim is still required, the system may determine the workflow to include notifying a particular actor(s) to provide the missing documents. If the documentation for a claim is complete, the system may determine the workflow to include notifying an agent or another actor to review the documents and settle the claim.

“It is to be appreciated that, with constant improvement to camera quality, the number of cameras, and storage capacity on user devices (e.g., smartphones), the amount of data generated by photos and videos from these user devices can increase significantly. In response, the system needs to “learn” how to transfer and process data more efficiently. With the vast quantities of settled claims data, training data may be collected from samples of the settled claims data that are labeled with a severity level of a claim, the payout amount, and/or total loss claims. The system may train additional machine learning model(s) using this training data with samples of claims data labeled with severity levels. This trained model may classify data by severity levels and may increase the priority for the claim processing. Once trained, these machine learning model(s) can be transmitted to user devices to classify gathered data into the different severity levels. Based on the severity score or severity level of the data, the system components, running on user devices or on the cloud while communicating with the user devices, may determine the appropriate data transport method. For example, if the image data from a particular user device is classified as the highest level of severity, the system may determine that the data should be transmitted over the network to the cloud servers right away. In some examples, the system may compress or decrease the resolution of the image data for faster transport. However, if the data is classified as less than the highest level of severity, the system may select from any number of alternative data transport methods available to that user device, including but not limited to queueing the data to be transmitting over a Wi-Fi network when available or sending image data to another device or storage medium for transport, among other options.

“Implementations of the techniques and systems described herein can improve existing technologies (e.g., claims processing technologies), and can improve the functioning of the systems with respect to conventional systems. In particular, the implementations described herein allow a claims processing system to: assist users in documentation gathering; automatically analyze the content; and/or automatically determine a workflow to expedite the claims process. Additionally, the claims processing system may train ML models to classify input data and verify the classification of collected data at the user device level before uploading the data. By classifying and verifying data at the user device level, the system improves the document collection process by timely identifying documents of questionable quality and notifying the user about the documents. For instance, if the user is trying to upload a blurry image of car damage or an image of illegible repair estimates, giving the user timely notice to correct documentation may save the user additional trips to the shop. In additional examples, by classifying and verifying data at the user device level, the system may refrain from transmitting data for documents of questionable quality. Implementations of this disclosure can also conserve resources, such as processing resources, for example, by refraining from performing data analysis and/or transmitting data for review in when the data is incorrect or insufficient (e.g., in cases where the data is not matching what is needed to advance the claim). Additionally, refraining from transmitting claims data to servers reduces overall network usage. These techniques will be described in more detail below.

“It should be appreciated that the subject matter presented herein can be implemented as a computer process, a computer-controlled apparatus, a computing system, or an article of manufacture, such as a computer-readable storage medium. While the subject matter described herein is presented in the general context of program components that execute on one or more computing devices, those skilled in the art will recognize that other implementations can be performed in combination with other types of program components. Generally, program components include routines, programs, components, data structures, and other types of structures that perform particular tasks or implement particular abstract data types.”

The claims supplied by the inventors are:

“1. A system comprising: one or more processors; and computer-readable media storing first computer-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising: receiving a plurality of electronic files containing information associated with previously settled claims; determining, for individual electronic files of the plurality of electronic files and based on the information contained in the individual electronic files, insurance categories; generating training data comprising samples of images in the plurality of electronic files, wherein generating the training data includes adding labeled metadata to individual samples of images, the labeled metadata identifying the insurance categories; training a machine learning (ML) model with the training data to: extract text information from image input data, classify the image input data with image input data classifications and based on the labeled metadata, determine image quality data for the image input data, and determine confidence scores for the image input data classifications, wherein the image input data may be in one or more file formats; receiving image data associated with insurance claims documents for an insurance claim; executing the ML model using the image data as first input to generate first structured information comprising text data, a first insurance category, an image quality of the image data, and a first confidence score for the first insurance category as first output; determining that the first confidence score meets or exceeds a first threshold; based on the image quality, and on determining that the first confidence score meets or exceeds the first threshold: generating metadata indicating the first structured information and the first insurance category; and generating, based on the first insurance category, a first workflow for the metadata, the first workflow including performing advanced data analysis associated with the first insurance category on the image data; determining, based on performing the advanced data analysis on the image data, a type of structured information to indicate in the metadata; generating a user interface; in response to receiving unstructured data via the user interface, executing the trained ML model using the unstructured data as second input to generate second structured information of the type of structured information and a second insurance category of the insurance categories corresponding to the unstructured data as second output; determining, based on the second insurance category, a second confidence score for the second structured information; determining that the second confidence score is less than a second threshold; and generating, based on determining that the second confidence score is less than the second threshold, a second workflow for the metadata, the second workflow including generating a notification to request additional documentation for the insurance claim.

“2. The system of claim 1, wherein: the operations further comprise, further based on the image quality, and on determining that the first confidence score meets or exceeds the first threshold and the image quality: augmenting the training data with the image data, the first structured information, and the first insurance category; and retraining the trained ML model using the augmented training data to generate a retrained ML model; and executing the trained ML model using the unstructured data as the second input to generate the second structured information and the second insurance category as the second output comprises executing the retrained ML model using the unstructured data as the second input to generate the second structured information and the second insurance category as the second output.

“3. The system of claim 1, wherein the second workflow further includes augmenting the metadata with the first structured information.

“4. The system of claim 1, wherein determining the first workflow includes: determining a first native format of the image data is different from a preferred file format associated with the first insurance category, wherein the first insurance category is associated with an insurance policy or an insurance claim type; and determining to reformat the image data to be stored as the first preferred file format.

“5. The system of claim 1, wherein the image data includes an image of at least one of repair estimates, emails, police reports, or medical bills.

“6. The system of claim 1, wherein the metadata indicates a medical bill and executing the trained ML model to generate the second structured information as output comprises: executing the trained ML model to generate the second structured information including one or more of a cost of medical service or a service provider name.

“7. The system of claim 1, wherein the metadata indicates a repair estimate and executing the trained ML model to generate the second structured information as output comprises: executing the trained ML model to generate the second structured information including one or more of an estimated cost to repair or an auto shop name.

“8. The system of claim 1, the operations further comprising: creating new training data by identifying sample data from stored data including the metadata; and generating one or more new ML models with the new training data to classify input by the insurance categories.

“9. The system of claim 1, the operations further comprising: generating a second user interface to present the insurance claims documents for viewing; receiving a request to view a first document of the insurance claims documents, the first document associated with the image data; retrieving the image data in a native format; and presenting the image data for viewing.

“10. A method, comprising: receiving a trained model trained to extract text information from image input data classify the image input data with image input data classifications and based on payout amount, determine image quality data for the image input data, and determine confidence scores for the image input data classifications, wherein the trained model is trained using training data comprising samples of images in claims data and respective labeled metadata indicating payout amounts corresponding to the claims data; generating a user portal to receive insurance claims documents; receiving, from a computing device via the user portal, image data associated with an insurance claim from a user account; executing the trained model using the image data as first input to generate first structured information comprising text data, a first insurance category, an image quality of the image data, and a first confidence score for the first insurance category as first output; determining that the first confidence score meets or exceeds a first threshold; based on the image quality, and on determining that the first confidence score meets or exceeds the first threshold: generating metadata indicating the first structured information and the first insurance category; and determining, based on the first insurance category, a first workflow for the metadata, the first workflow including performing advanced data analysis on the image data; determining, based on performing advanced data analysis on the image data, a type of structured information to indicate in the metadata; generating a user interface; in response to receiving unstructured data via the user interface, executing the trained model using the unstructured data as second input to generate second structured information of the type of structured information and a second insurance category of the insurance categories corresponding to the unstructured data as second output; determining, based on the second insurance category, a second confidence score for the second structured information; determining that the second confidence score is less than a second threshold; generating, based on determining that the second confidence score is less than the second threshold, a second workflow for the metadata, the second workflow including generating a notification for the unstructured data; and causing to present, via the user portal, the notification for the user account.

“11. The method of claim 10, where performing advanced data analysis on the image data comprises: determining the image data includes a vehicle photo and a license plate number; and determining to indicate the license plate number in the metadata.

“12. The method of claim 10, further comprising: creating the training data by identifying sample data from settled claims, individual data of the sample data including first labels with claims categories; generating a trained machine learning (ML) model with the training data to classify input by the claims categories and to determine associated confidence scores; receiving, from the computing device, the claims data, wherein a portion of the claims data includes confidence scores below a third threshold; creating second training data by identifying incorrectly classified data in the portion of the claims data; labeling the second training data by correct insurance categories; and generating a second trained ML model with the second training data.

“13. The method of claim 10, wherein: the method further comprises, further based on determining that the first confidence score meets or exceeds the first threshold and the image quality: augmenting the training data with the image data, the first structured information, and the first insurance category; and retraining the trained model using the augmented training data to generate a retrained model; and executing the trained model using the unstructured data as the second input to generate the second structured information and the second insurance as the second output comprises executing the retrained model.”

There are additional claims. Please visit full patent to read further.

For additional information on this patent, see: Garapati, Sunitha. Claims process assistance using models. U.S. Patent Number 11972489, filed April 23, 2021, and published online on April 30, 2024. Patent URL (for desktop use only): https://ppubs.uspto.gov/pubwebapp/external.html?q=(11972489)&db=USPAT&type=ids

(Our reports deliver fact-based news of research and discoveries from around the world.)