Method and device for supporting dynamic decision-making regarding insurance products based on artificial intelligence
The integration of LLM with insurance and medical data to calculate future disease and premium changes addresses the limitations of conventional systems, enabling accurate and actionable insurance decision-making.
Patent Information
- Application Number
- KR1020260060739
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-04-03
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-04-03
AI Technical Summary
Consumers face difficulties in making informed insurance decisions due to complex contract structures, premium fluctuations, and individual health status changes, while conventional systems fail to provide actionable recommendations and accurately reflect actual claim results.
A method utilizing a large-scale language model (LLM) integrates medical and insurance data to calculate future disease state transitions, premium fluctuations, and friction costs, generating actionable suggestions through a hybrid parameter generation structure.
Enables accurate, long-term insurance decision-making by dynamically reflecting future health and premium changes, providing practical recommendations beyond static comparisons.
Smart Images

Figure 112026040897538-PAT00002_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a method and apparatus for supporting dynamic decision-making regarding insurance products based on artificial intelligence. More specifically, it relates to a method and apparatus for dynamically supporting a user's insurance decision-making by utilizing a Large Language Model (LLM) to combine the user's medical usage history, insurance contract information, and insurance claim history, and by calculating a life-cycle net value that considers the probability of future disease state transition and premium fluctuations. Background Technology
[0002] In the medical and insurance markets, consumers find it very difficult to make insurance decisions optimized for their specific situations due to complex contract structures, diverse coverage scopes, premium fluctuations upon renewal, and differences in individual medical utilization patterns. In particular, consumers currently find it difficult to determine on their own whether their current insurance is suitable in light of their health status and future medical usage patterns.
[0003] Conventional insurance recommendation systems have primarily adopted static recommendation methods based on information from a single point in time. This approach has fundamental limitations in that it fails to reflect risk transfer due to changes in future health status, premium fluctuations over time, and realistic costs (friction costs) incurred during the actual insurance claim process.
[0004] Furthermore, conventional systems often stop at simply recommending insurance products, failing to provide specific and actionable suggestions on "what to do next" for consumers. Moreover, despite the frequent changes in national insurance systems and terms and conditions, static rule-based models struggle to respond quickly to these changes, resulting in high maintenance costs. Conversely, from the perspective of insurance agents recommending products, there is also the problem of difficulty in providing objective indicators regarding "how much annual benefit this insurance can provide to the consumer."
[0005] Furthermore, despite the fact that a significant amount of insurance benefits are not actually received due to incomplete documentation, procedural complexity, and consumer awareness during the claim process, there was no system to quantify this as a cost and reflect it in insurance decision-making.
[0006] To address these issues, there is a need to develop an AI-based decision support system capable of combining medical data and insurance contract data, dynamically calculating life-cycle net value by considering future disease state transitions and premium fluctuations over time, and generating actionable action suggestions based on this. The problem to be solved
[0007] The present invention aims to solve the problems of the aforementioned prior art by providing a method and apparatus that effectively support insurance decision-making by utilizing a large-scale language model (LLM) to integrate and analyze a user's medical usage history, insurance contract information, and insurance claim history, and dynamically calculating the probability of future disease state transition, premium fluctuation amount, expected benefits, and friction costs.
[0008] In addition, the present invention aims to resolve the discrepancy between theoretical values of policy knowledge and actual claim results and to calculate more accurate parameter values through a hybrid parameter generation structure that combines LLM / RAG-based knowledge extraction with a database of actual insurance payment results.
[0009] Furthermore, the present invention aims to support the entire decision-making process beyond simple insurance recommendations by generating a strategic signal vector based on Lifecycle Net Value (LVV) comparison, classifying it into strategic states, and generating specific action suggestions (Next Best Action) that consumers can actually execute. means of solving the problem
[0010] A method for supporting dynamic decision-making regarding an insurance product based on artificial intelligence according to an embodiment of the present invention for achieving the above objective may include: (a) acquiring user data including medical usage history, insurance contract information, and insurance claim history of a user and storing it in a database; (b) calculating the probability of future disease state transition of the user, the amount of future premium fluctuation for insurance contracts held by the user, and the expected benefits and friction costs upon subscribing to a comparable insurance, by having a large-scale language model (LLM) refer to the database; and (c) determining whether subscribing to a comparable insurance is advantageous to the user based on the calculated information, and generating and outputting a behavioral suggestion based on the result of the determination.
[0011] In one embodiment, step (b) comprises: a large-scale language model generating parameter values from unstructured knowledge data including one or more of insurance terms and conditions related to insurance contract information, insurance-related press releases, and insurance-related precedents; a step of correcting parameter values based on the difference between the generated parameter values and actual insurance payout result data stored in a database; and a step of calculating future disease state transition probability, future premium change amount, expected benefit, and friction cost using the corrected parameter values, wherein the parameter values may include at least one of payability, claim success rate, surcharge factor, and friction cost.
[0012] In one embodiment, the probability of future disease state transition may be calculated by applying a corrected parameter value to classify the user's current health state into one of a plurality of disease states, and reflecting the age, gender, and past diagnosis history included in the medical usage history to calculate the probability of transition from the current health state to the future health state over time.
[0013] In one embodiment, the future premium change amount may be calculated by applying a corrected parameter value and reflecting a group renewal factor calculated based on renewal conditions included in the insurance contract information and an individual surcharge factor calculated based on the insurance claim history, thereby calculating the premium change amount over time of the insurance contract.
[0014] In one embodiment, the expected benefit is calculated by applying a corrected parameter value to estimate the probability of medical utilization occurring at each time point based on the probability of future disease state transition, and reflecting the actual payability to the payable amount according to the coverage conditions of the comparative insurance, and the friction cost may be calculated by applying a corrected parameter value to calculate the probability of a claim success from the claim history, and calculating the cost predicted for the user to claim the insurance but not receive it based on the claim success probability.
[0015] In one embodiment, step (c) is performed by calculating the net value when the user maintains the insurance contract and the net value when the user subscribes to the comparable insurance, respectively, and based on the difference between the respective net values, and the net value may be calculated by subtracting the premium, deductible, and frictional costs from the expected benefit and applying a time discount rate (r).
[0016] In one embodiment, step (c) generates a plurality of strategy signals including differences between each net value, combines the plurality of strategy signals to generate a strategy signal vector, and the strategy signal vector may include one or more of a conversion signal indicating the degree of attractiveness of switching to a comparable insurance, a maintenance signal indicating the degree of attractiveness of maintaining the insurance contract, a supplementation signal indicating the degree of need to supplement coverage gaps, and an optimization signal indicating the degree of need to improve the insurance structure.
[0017] In one embodiment, step (c) classifies the user's current situation into one of a plurality of strategy states based on additional information including a strategy signal vector, the reliability of the calculated information, and the completeness of the user data, and the plurality of strategy states may include one or more of a maintenance advantage state, a switching advantage state, a state requiring supplementation, a state requiring optimization, and a state requiring additional information.
[0018] In one embodiment, step (c) further comprises: generating a behavioral suggestion corresponding to a classified strategy state; and a large-scale language model converting the behavioral suggestion into a natural language form including the basis for judgment and specific actions actionable by the user and outputting it, wherein the behavioral suggestion may include one or more of maintaining the insurance contract, switching to a comparable insurance, signing up for additional coverage to fill coverage gaps, and optimizing the insurance structure. Effects of the invention
[0019] According to the present invention, through a hybrid parameter generation structure combining a large-scale language model (LLM) and a database of actual insurance payout results, it is possible to resolve the discrepancy between theoretical figures in insurance terms and conditions and actual claim results, thereby enabling more accurate decision support.
[0020] Furthermore, according to the present invention, by calculating the Life Cycle Net Value (LNV) that dynamically reflects future disease state transition probabilities and premium fluctuations on the time axis, it is possible to support optimal insurance decision-making from a long-term perspective, going beyond static comparisons at a single point in time.
[0021] Furthermore, according to the present invention, by providing a concrete Next Best Action that consumers can actually execute through a stepwise process of generating a strategy signal vector, classifying a strategy state, and generating an action suggestion, it is possible to overcome the limitations of a simple insurance recommendation system and provide practical support for insurance decision-making.
[0022] Furthermore, the engine of the present invention is designed to allow the replacement of national insurance and medical system knowledge with an external knowledge layer (RAG / Rules), enabling universal application to insurance environments in various countries and ensuring global scalability.
[0023] The present invention can be implemented as a system that operates through network communication between a server and a terminal, but it can also operate independently at the terminal level, allowing it to be utilized without restrictions on the network environment. Brief explanation of the drawing
[0024] FIG. 1 is a conceptual diagram illustrating the concept of artificial intelligence-based dynamic insurance decision support according to one embodiment of the present invention. FIG. 2 is a diagram illustrating the configuration of a dynamic insurance decision support system according to one embodiment of the present invention. FIG. 3 is a diagram illustrating the internal module configuration of a server according to one embodiment of the present invention. FIG. 4 is a flowchart illustrating the overall process of an artificial intelligence-based dynamic insurance decision support method according to one embodiment of the present invention. FIG. 5 is an operation flowchart illustrating the detailed operation of the user data collection and storage step according to one embodiment of the present invention. FIG. 6 is an operation flowchart for explaining the detailed operation of a future information output step according to an embodiment of the present invention. FIG. 7 is a flowchart illustrating the detailed operation of the decision support and action suggestion generation step according to one embodiment of the present invention. Specific details for implementing the invention
[0025] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.
[0026] The following examples are detailed descriptions to aid in understanding the present invention and are not intended to limit the scope of the present invention. Accordingly, inventions within the same scope that perform the same function as the present invention will also fall within the scope of the present invention.
[0027] Throughout the entire disclosure, the same reference numerals refer to the same components.
[0028] This disclosure does not describe all elements of the embodiments, and general content in the art to which this disclosure pertains or content that overlaps with the embodiments is omitted. The terms “part, module, component, block” as used in the specification may be implemented in software or hardware, and depending on the embodiments, a plurality of “parts, modules, components, blocks” may be implemented as a single component, or a single “part, module, component, block” may include a plurality of components. Throughout the specification, when a part is described as being “connected” to another part, this includes not only direct connection but also indirect connection, and indirect connection includes connection via a wireless communication network.
[0029] Furthermore, when it is stated that a part "includes" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.
[0030] Throughout the specification, when it is stated that a component is located "on" another component, this includes not only cases where a component is in contact with another component, but also cases where another component exists between the two components.
[0031] The terms first, second, etc. are used to distinguish one component from another, and the components are not limited by the aforementioned terms.
[0032] Singular expressions include plural expressions unless the context clearly indicates an exception. Identification codes for each step are used for convenience of explanation and do not describe the order of the steps; furthermore, the steps may be performed differently from the specified order unless the context clearly indicates a specific order.
[0033] The operating principles and embodiments of the present disclosure will be described below with reference to the attached drawings.
[0034] Prior to the explanation, the meanings of the terms used in this specification are briefly explained. However, since the explanation of terms is intended to aid in understanding this specification, it should be noted that they are not used to limit the technical scope of this disclosure unless explicitly stated to be a limiting factor.
[0035] In this specification, the term "device" includes all various devices capable of performing computational processing and providing results to a user. For example, a device may include a computer, a server device, and a portable terminal, or be any one of these forms.
[0036] Here, the computer may include, for example, a notebook, desktop, laptop, tablet PC, slate PC, etc. equipped with a web browser.
[0037] The above server device is a server that processes information by communicating with an external device, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, and a web server.
[0038] The above portable terminal may include, for example, all types of handheld-based wireless communication devices such as PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (WCode Division Multiple Access), WiBro (Wireless Broadband Internet) terminals, smartphones, etc., as well as wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head-mounted devices (HMDs).
[0039] Functions related to artificial intelligence according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.
[0040] The predefined rules of operation or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a predefined rules of operation or artificial intelligence models configured to perform a desired characteristic (or objective) are created by a basic artificial intelligence model being trained using a number of training data by a learning algorithm. Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.
[0041] An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values and performs neural network operations through operations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights may be updated so that the loss value or cost value obtained from the artificial intelligence model during the learning process is reduced or minimized. The artificial neural network may include a Deep Neural Network (DNN), such as a Convolutional Neural Network (CNN), Deep Neural Network (DNN), Recurrent Neural Network (RNN), Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), Bidirectional Recurrent Deep Neural Network (BRDNN), or Deep Q-Networks, but is not limited to the examples mentioned above. The processor can create a neural network, train (or learn) the neural network, perform operations based on received input data, generate an information signal based on the results of the operation, or retrain the neural network.
[0042] Neural networks include CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), perceptron, multilayer perceptron, FF (Feed Forward), RBF (Radial Basis Network), DFF (Deep Feed Forward), LSTM (Long Short Term Memory), GRU (Gated Recurrent Unit), AE (Auto Encoder), VAE (Variational Auto) Encoder), DAE (Denoising Auto Encoder), SAE (Sparse Auto Encoder), MC (Markov Chain), HN (Hopfield Network), BM (Boltzmann Machine), RBM (Restricted Boltzmann Machine), DBN (Depp Belief Network), DCN (Deep Convolutional Network), DN (Deconvolutional Network), DCIGN (Deep Convolutional Inverse Graphics Network), GAN (Generative Adversarial Network), LSM (Liquid State Machine), ELM (Extreme Learning Machine), ESN (Echo It will be understood by a person skilled in the art that any neural network may be included, but is not limited to, State Network, Deep Residual Network, Differential Neural Computer, Neural Turning Machine, Capsule Network, Kohonen Network, and Attention Network.According to an exemplary embodiment of the present disclosure, the processor comprises a Convolutional Neural Network (CNN) such as GoogleNet, AlexNet, VGG Network, Region with Convolutional Neural Network (R-CNN), Region Proposal Network (RPN), Recurrent Neural Network (RNN), Stacking-based Deep Neural Network (S-DNN), State Space Dynamic Neural Network (S-SDNN), Deconvolution Network, Deep Belief Network (DBN), Restructured Boltzmann Machine (RBM), Fully Convolutional Network, Long Short-Term Memory (LSTM) Network, Classification Network, Generative Modeling, eXplainable AI, Continual AI, Representation Learning, AI for Material Design, BERT, SP-BERT, MRC / QA, Text Analysis, Dialog System, GPT-3, GPT-4 for Natural Language Processing, Visual Analytics, Visual Understanding, Video Synthesis for Vision Processing, Anomaly Detection, Prediction, Time-Series Forecasting, Optimization, Recommendation for ResNet Data Intelligence, Various artificial intelligence structures and algorithms, such as data creation, may be used, but are not limited thereto.
[0043] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings.
[0044] [1. Concept of the Invention and Configuration of the System]
[0045] FIG. 1 is a conceptual diagram illustrating the concept of artificial intelligence-based dynamic insurance decision support according to one embodiment of the present invention.
[0046] Referring to FIG. 1, the user can input their current health status and input a query regarding insurance decision-making through a user terminal (200). Subsequently, the server (100) can provide the user with an optimal insurance decision-making result by utilizing a large-scale language model (LLM) and a database (DB).
[0047] For example, if a user asks, "I am not sure if the insurance I am currently subscribed to is appropriate. Considering my current health condition, please tell me if it is okay to maintain my current insurance," the system of the present invention provides specific and quantified behavioral suggestions in natural language, such as, "Maintaining the current 1st generation will cost 120 million won in premiums over the next 20 years. Switching to the 4th generation will cost 38 million won. Upon switching, the benefit per instance decreases by 40,000 won, but there is a total savings effect of 82 million won. It is recommended to switch to the 4th generation."
[0048] In other words, in the realm of insurance, very limited information is disclosed to users, often making it difficult for them to easily make decisions regarding the subscription, cancellation, or maintenance of insurance. In this context, the present invention, based on an LLM and a database storing insurance and health information, can provide answers to the decisions users need to make very quickly and simply. In particular, it can provide answers while clearly presenting the rationale compared to subscribing to a new insurance policy.
[0049] Meanwhile, in the present invention, the user may be a subscriber who intends to use insurance, or an insurance agent who sells or proposes insurance. Depending on the type of user, the large-scale language model may construct different answers for the same data and provide them to the user terminal (200).
[0050] FIG. 2 is a drawing illustrating the overall configuration of a system (1) according to one embodiment of the present invention.
[0051] Referring to FIG. 2, the system (1) includes a server (100), a user terminal (200), and a communication network (300). The server (100) and the user terminal (200) are interconnected through the communication network (300) to transmit and receive data.
[0052] However, the dynamic insurance decision support function of the present invention can operate in a system environment where the server (100) and the user terminal (200) are separated, and can also operate independently at the user terminal (200) level. That is, when the user terminal (200) is equipped with the necessary computational resources and models, the server (100) is excluded, and the core function of the present invention can be performed without a network connection. In addition, in the following description, the processor may be the processor of the server (100) or the processor of the user terminal (200).
[0053] The server (100) collects and stores user data, and utilizes a large-scale language model (160) and a database (150) to calculate future information and then generates action suggestions and provides them to the user terminal (200).
[0054] The user terminal (200) is a terminal device directly operated by a user, such as a smartphone, tablet, or PC, and receives a query from the user and displays action suggestions received from the server (100). The communication network (300) is a network that mediates data communication between the server (100) and the user terminal (200), and includes the Internet and mobile communication networks. Additionally, the user terminal (200) may include a terminal of a customer who wishes to obtain knowledge about insurance, or a terminal of an insurance agent who sells or recommends insurance.
[0055] [2. Server Configuration]
[0056] FIG. 3 is a diagram illustrating the internal module configuration of a server (100) according to an embodiment of the present invention. Referring to FIG. 3, the server (100) includes a data collection module (110), a parameter generation module (120), an analysis engine module (130), a behavior suggestion module (140), a database (150), and an LLM (160).
[0057] First, the data collection module (110) performs the role of collecting and standardizing user data, including medical usage history, insurance contract information, and insurance claim history for the user, and storing it in the database (150).
[0058] The collected data consists of five categories.
[0059] First, as medical utilization data, it is a past medical usage history including a user's medical consultation, prescription, treatment, costs, and visit types (inpatient and outpatient, etc.), and is standardized into a MedicalEvent schema consisting of time point, disease or condition, cost, type of use, and medication and treatment information.
[0060] Second, as insurance contract and coverage data, it includes contract information and coverage details for Actual Loss and Flat Rate insurance, and is standardized into an Insurance Contract schema consisting of premiums, deductibles, copayment ratios, coverage limit renewal rules, and surcharge rules.
[0061] Third, as payment history data, it is standardized into a ClaimOutcome schema that includes the actual insurance payment results and consists of whether the claim was successful or unsuccessful, the payment amount, the reason for failure, and the probability.
[0062] Fourth, as data on missed insurance benefits and unclaimed details (Hidden Benefit), it includes information such as claim failure, claim abandonment, or non-coverage structure, and has label characteristics such as exclusion, non-coverage, duplication, deduction, and document issues, and is used as training data to estimate claim success rates and payability.
[0063] Fifth, as a User Capacity indicator, it includes the consumer's economic constraints, such as income, expenditure, and available cash flow.
[0064] Next, the parameter generation module (120) performs the role of generating core parameters of the engine by utilizing the LLM (160) and database (150), and is composed of three sub-modules as a hybrid knowledge layer.
[0065] The first module, the LLM / RAG-based knowledge extraction module, takes unstructured data as input—such as insurance policy terms, press releases, case law, and notices regarding the latest system changes—and extracts rules and knowledge. Specifically, LLM performs the role of verifying coverage status within policy texts, extracting premium surcharge range rules specific to each insurer, identifying the types of documents required for claims, and determining requirements such as symptom improvement. Furthermore, it searches for the latest policy terms and system changes in real-time to maintain the knowledge layer in an up-to-date state.
[0066] The second module, the DB-based execution data probability extraction module, takes structured execution data as input—such as actual payment history, hidden benefits, and rejected claims—and statistically calculates the rate of actual payments for specific insurance products, the rate of abandoned claims due to incomplete documentation, and the average point of entry for premium surcharges. This module does not involve LLM and consists purely of statistical and mathematical operations on structured data.
[0067] The third module, the weighted combination and confidence calculation module, calculates the final parameter by weighted combining the theoretical value extracted by the first module and the actual probability extracted by the second module, and simultaneously calculates the confidence score of the parameter. For example, if the LLM (160) in the first module extracts the policy knowledge that "the limit for manual therapy in 4th generation indemnity insurance is 50 times per year," the second module checks the execution data from the database (150) that "the actual payment rejection rate is 35% when exceeding 10 times," and the third module operates by combining these two values to correct the probability of payment to 0.85.
[0068] Next, the analysis engine module (130) calculates the probability of future disease state transition, premium change amount, expected benefit, and friction cost using the parameter values generated by the parameter generation module (120), calculates the net life value (LNV) based on this, generates a strategy signal vector, and classifies the strategy state.
[0069] Next, the action suggestion module (140) generates action suggestions corresponding to the strategy states classified by the analysis engine module (130) and converts them into natural language form through the LLM (160) to output them.
[0070] Next, the database (150) performs the role of storing user data collected from the data collection module (110), parameter values generated from the parameter generation module (120), and output results from the analysis engine module (130).
[0071] Finally, LLM (160) performs the role of extracting rules and knowledge from unstructured knowledge data and converting action suggestions into natural language form.
[0072] At this time, although the LLM (160) is shown in FIG. 3 as being included in the server (100), it may be stored in an external server or external object depending on the case, and may be implemented through communication with the server (100) or user terminal (200).
[0073] [3. Operation Flow]
[0074] FIG. 4 is a flowchart illustrating the overall operation flow of an artificial intelligence-based dynamic insurance decision support method according to one embodiment of the present invention.
[0075] Referring to FIG. 4, the server (100) obtains user data including medical usage history, insurance contract information, and insurance claim history for the user and stores it in a database (S100).
[0076] In step (S100), the data collection module (110) standardizes the acquired user data into five common schemas: Medical Event, Insurance Contract, Claim Outcome, Hidden Benefit, and User Capacity, and then stores them in the database (150).
[0077] Through this standardization, the same engine structure can be applied even if national insurance and medical systems differ, and global scalability is secured by externalizing country-specific knowledge into sets of insurance contract rules and injecting them into the engine in the form of RAGs or Rules.
[0078] Here, FIG. 5 is a flowchart illustrating the detailed operation of user data collection and storage corresponding to step (S100). Referring to FIG. 5, the server (100) acquires user data (S110), standardizes the acquired data, and stores it in a database (150) (S120).
[0079] Next, the server (100) receives a request to query the optimal insurance decision based on the user's current health condition (S200).
[0080] The request received in step (S200) is input from the user terminal (200) and includes a query regarding insurance decision-making along with the user's current health status information. If the server (100) receives the request, it initiates a subsequent future information calculation step (S300).
[0081] Next, a large-scale language model (LLM (160)) refers to a database (150) to calculate the probability of a user’s future disease state transition, the amount of future premium fluctuation for the insurance contract, the expected benefits and friction costs when subscribing to a comparable insurance (S300). Here, the detailed operation of step (S300) is illustrated in FIG. 6.
[0082] Referring to FIG. 6, the LLM (160) generates parameter values by referring to the database (150) (S310), calculates the probability of future disease state transition through the generated parameter values (S320), calculates the future premium change amount (S330), and calculates the expected benefits and friction costs by selecting a comparable insurance (S340). At this time, steps (S320) to (S340) may be performed simultaneously in parallel as needed.
[0083] In step (310), the LLM (160) generates parameter values from unstructured knowledge data including one or more of insurance policy terms, insurance-related press releases, and insurance-related case precedents. The generated parameter values are compared with actual insurance payout result data stored in the database (150), and corrections are performed based on the difference. The corrected parameter values include at least one of payability, claim success rate, surcharge factor, and friction cost.
[0084] (1)-① Calculation of disease state transition probability
[0085] In step (S320), the analysis engine module (130) applies the corrected parameter value to classify the user's current health status into one of a plurality of disease states, such as Healthy, Mild, Chronic, and Severe.
[0086] Next, the probability of transition from the current health state to the future health state over time is calculated by reflecting the age, gender, and past diagnosis history included in the medical use history, and based on this, the intensity of medical use at each point in time is estimated as a function of age, health state, and medical use history.
[0087] In this step (320), the LLM (160) performs an auxiliary role of searching for the latest clinical guidelines for the user's diagnosis, extracting information on the standard number of treatments and prognosis of the disease, and providing for the calculation of metastasis probability. For example, if a 48-year-old male shows early symptoms of cervical and lumbar disc herniation and uses manual therapy 15 times a year, the intensity of medical use after 10 years can be estimated to be about 16.8 times per year by applying an aging factor (3% per year) and a manual therapy fixation factor (2.5 times).
[0088] (1)-② Calculation of insurance premium fluctuation amount
[0089] In step (S330), the analysis engine module (130) applies corrected parameter values to calculate the amount of premium change over time of the insurance contract by reflecting the group renewal factor calculated based on the renewal conditions included in the insurance contract information and the individual surcharge factor calculated based on the insurance claim history.
[0090] The group renewal factor is a group premium adjustment factor at the national or product group level, and the individual surcharge factor is a surcharge coefficient applied according to an individual's claim frequency or usage pattern, and both factors are calculated in the parameter generation module (120). For example, the first generation medical indemnity insurance is calculated to have a total premium of approximately 120 million won over 20 years from the current monthly premium of 120,000 won by applying a group renewal factor of 15% per year, and the fourth generation medical indemnity insurance can be calculated to have a total premium of approximately 38 million won over 20 years by applying a 3-stage individual surcharge factor (maximum 300%) to a basic renewal factor of 3% per year.
[0091] (1)-③ Calculation of Expected Benefits and Friction Costs
[0092] In step (340), the analysis engine module (130) selects a comparable insurance and calculates the expected benefits and friction costs. The expected benefits are calculated by reflecting the actual payability of the payable amount according to the coverage conditions of the comparable insurance and the probability of medical utilization occurring at each time point, estimated based on the probability of future disease state transition.
[0093] The actual payability is a parameter that reflects realistic factors such as non-coverage, exclusion, duplication, and deduction, and is calculated by correcting the theoretical value extracted from the terms and conditions by the LLM (160) with the execution data of the database (150). For example, when the LLM (160) extracts knowledge of the terms and conditions that "in 4th generation indemnity insurance, confirmation of symptom improvement is required when manual therapy is performed more than 10 times," it checks the execution data in the database (150) that "the rate at which actual users give up on claims due to insufficient documentation or stricter screening is 15%" and corrects the payability to 0.85.
[0094] Friction costs quantify the invisible costs incurred during the insurance claim process. They calculate the predicted costs associated with a user claiming insurance benefits but failing to receive them, based on the probability of success derived from the claim history. The probability of success comprehensively reflects factors such as the completeness of documentation, procedural difficulty, consumer awareness levels, claim abandonment rates, and rejection rates, with unclaimed data (Hidden Benefit) utilized as key label data. For example, the friction costs for 4th generation medical indemnity insurance can be calculated as 310,000 won per year and 6 million won over 20 years, including documentation complexity (15% claim abandonment) and proof costs.
[0095] To explain the entire process of the present invention again through FIG. 4, the server (100) determines whether it is advantageous for the user to subscribe to a comparable insurance based on the calculated information, and supports decision-making by generating and outputting a suggestion for action based on the result of the determination (S400). Here, the detailed operation of step (S400) is illustrated in FIG. 7.
[0096] Referring to FIG. 7, the analysis engine module (130) generates a plurality of strategy signal vectors through the difference between net values (S410), classifies the strategy state through the reliability of the strategy signal vectors and the calculated information and the completeness of the user data (S420), the action suggestion module (140) generates an action suggestion corresponding to the classified strategy state (S430), generates output data for the action suggestion through the LLM (160) (S440), and provides the output data and information of the comparable insurance to the insurance subscriber terminal or the insurance agent terminal (S450).
[0097] (2) Generation of strategy signal vectors
[0098] In step (S410), the analysis engine module (130) calculates the net value when the user maintains the insurance contract they hold and the net value when they subscribe to the comparable insurance, respectively, generates multiple strategy signals based on the difference, and then combines them to generate a strategy signal vector.
[0099] In this case, the net value is the Lifecycle Net Value (LVV), which is calculated by summing the present value of the net benefits at each point in time—obtained by subtracting premiums, out-of-pocket expenses (OOP), and friction costs from expected benefits over the consideration period—using the time discount rate. Here, the Lifecycle Net Value for insurance contract j is given by [Equation 1] below.
[0100]
[0101] Here, T is the consideration period (e.g., N years / lifetime), r is the discount rate, E[Benefit] is the expected benefit at time t, Premium is the change in premium at time t, OOP is the out-of-pocket cost (stability) at time t, and RiskAdj is the frictional cost at time t.
[0102] For example, for a high-frequency user of manual therapy, if the lifetime net value of 1st generation indemnity insurance is calculated as -52 million won and the lifetime net value of 4th generation indemnity insurance is calculated as -2 million won, the difference in net value becomes +50 million won.
[0103] In addition, the strategy signal vector includes one or more of a switching signal indicating the degree of attractiveness of switching to a comparable insurance, a maintenance signal indicating the degree of attractiveness of maintaining the insurance contract, a supplementation signal indicating the degree of need to supplement coverage gaps, and an optimization signal indicating the degree of need to improve the insurance structure.
[0104] Here, the conversion signal is the value obtained by normalizing the aforementioned net value difference through a non-linear transformation function, and the maintenance signal is defined as the complementary value of the conversion signal. The complementary signal is the value obtained by normalizing the increase in net value when specific coverage is added to an existing contract, and the optimization signal is a function value calculated by combining friction costs, the self-pay structure, and coverage gaps.
[0105] (3) Strategic State Classification
[0106] In step (420), the analysis engine module (130) classifies the user's current situation into one of a plurality of strategy states based on additional information including a strategy signal vector, the reliability of the calculated information, and the completeness of the user data.
[0107] In this context, reliability indicates model uncertainty and data scarcity, data completeness indicates the level of missing data in contractual and medical records, and risk trends represent an increasing risk trend based on the probability of disease metastasis.
[0108] There are five classified strategic states. The "Maintain Advantage" state is one where maintaining the current contract is the best course of action. The "Switch Advantage" state is one where switching to an alternative contract is advantageous. The "Supplement Needed" state is one where the current contract needs to be maintained, but coverage gaps need to be addressed. The "Optimization Needed" state is one where the insurance structure needs to be maintained, but structural improvements regarding claims methods, coverage structure, and deductibles are required. The "Additional Information Needed" state is one where judgment is withheld due to insufficient data, and further information collection is required. For example, in the case of the aforementioned high-frequency manual therapy user, the lifetime net value difference is +50 million KRW and the data reliability reaches 98%, so the switching signal is calculated as 0.95, classifying it as the "Switch Advantage" state.
[0109] (4) Generating action suggestions
[0110] In step (S430), the action suggestion module (140) generates an action suggestion corresponding to the classified strategy state. The generation of the action suggestion is performed by receiving as input constraints such as the strategy state, the difference in life cycle net value, major drivers that worsen or improve the life cycle net value, and the consumer's income, cash flow, and risk avoidance preference.
[0111] The direction and framework of the action proposal are determined by the mathematical output of the engine, and the LLM (160) is limited to the role of concretizing this. Examples of action proposals according to each strategy state are as follows.
[0112] In a state of retention advantage, verification of current contract maintenance, guidance on cost-saving points, and support for the recovery of unclaimed insurance benefits may be proposed. In a state of switching advantage, the generation of alternative contract candidates, provision of comparative reports on 5-, 10-, and 15-year life cycle net value scenarios, and comparisons by chronic disease transition scenario (conservative / neutral / aggressive) may be proposed. In a state requiring supplementation, identification of coverage gaps and presentation of additional coverage candidates, as well as analysis of the life cycle net value improvement effect relative to additional costs, may be proposed. In a state requiring optimization, guidance on improving claims methods, redesign of the copayment structure, and measures to reduce friction costs (such as document automation) may be proposed. In a state requiring additional information, the generation of essential questions to minimize data gaps and the presentation of highly sensitive variables (frequency of non-covered expenses, income fluctuations, etc.) may be proposed.
[0113] (5) Natural language conversion and output
[0114] In step (S440), the LLM (160) converts the action suggestion into a natural language form containing the basis for judgment and specific actions that the user can take, and outputs it. The LLM (160) retrieves the latest terms and conditions and system change information through the RAG and reflects it in the basis text, thereby ensuring that the output content remains up to date.
[0115] In step (S450), the server (100) provides output data and information on the comparable insurance to the insurance subscriber's terminal or to the insurance agent's terminal. When providing action suggestions to the insurance agent's terminal, contract information and coverage scope information regarding the comparable insurance are provided together. Meanwhile, when operating independently at the terminal level, the user terminal (200) can independently perform all functions of the present invention without a network connection and display results to the user.
[0116] [4. Example: Simulation of High-Frequency Manual Therapy Users]
[0117] The operation of the present invention will be described in detail below through specific embodiments. The user profile is a 48-year-old male with a BMI of 27 and early symptoms of cervical / lumbar disc herniation, who uses manual therapy 15 times a year (cost of 150,000 won per session) and has medical usage history data for 3 years.
[0118] The comparison contracts are (A) 1st generation indemnity insurance: signed in 2008, 100% coverage for inpatient / outpatient treatment for injury / illness, deductible of 5,000 won, current monthly premium of 120,000 won, and (B) 4th generation indemnity insurance: as of 2024, deductible of 20% for covered expenses / 30% for non-covered expenses, annual limit of 50 manual therapy sessions / 3.5 million won, requirement for verification of symptom improvement every 10 sessions, current monthly premium of 30,000 won.
[0119] When the user data is input into the system of the present invention, the parameter generation module (120) calculates key parameters through a hybrid knowledge layer. The analysis engine module (130) predicts the continuation of non-covered treatment approximately 16.8 times per year after 10 years by reflecting the aging (3% per year) and manual therapy fixation (2.5 times) factors as the probability of disease state transition. As for premium fluctuations, the first generation is increased by an average of 15% per year, resulting in a total premium of approximately 120 million won over 20 years, while the fourth generation is increased by a basic annual increase of 3% with a 3-stage surcharge applied, resulting in a total premium of approximately 38 million won over 20 years. As for expected benefits and friction costs, the first generation is calculated to have a total benefit of 68 million won, while the fourth generation is calculated to have a total benefit of 42 million won and friction costs of 6 million won when applying a claim success rate of 85%. Based on this, when comparing the net value of the life cycle, the 1st generation is -52 million won and the 4th generation is -2 million won, and the difference in net value of the life cycle is calculated to be +50 million won.
[0120] Accordingly, the conversion signal is calculated as 0.95 (strong conversion advantage), the data reliability as 98%, and the strategy status is classified as a conversion advantage.
[0121] The action suggestion module (140) and LLM (160) generate the following action suggestions in natural language. The process is completed by providing a visualization report stating, "If you maintain the current 1st generation, the insurance premium for the next 20 years will be 120 million won. If you switch to the 4th generation, it will cost 38 million won. When switching, the benefit per session decreases by 40,000 won, but there is a total savings effect of 82 million won," a friction management support statement stating, "When switching to the 4th generation, you can set the app to automatically request a symptom improvement confirmation certificate required every 10 manual therapy sessions," and an immediate execution item stating, "You can immediately claim 520,000 won for the last 3 years of unclaimed history from the 1st generation insurance prior to the switch."
[0122] Meanwhile, in the above description, the steps described may be further divided into additional steps or combined into fewer steps according to an embodiment of the present disclosure.
[0123] Additionally, some steps may be omitted as necessary, and the order of steps may be changed.
[0124] Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium that stores instructions executable by a computer. The instructions may be stored in the form of program code and, when executed by a processor, may generate a program module to perform the operation of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.
[0125] Computer-readable recording media include all types of recording media that store instructions that can be decoded by a computer. Examples include ROM (Read Only Memory), RAM (Random Access Memory), magnetic tape, magnetic disk, flash memory, optical data storage devices, etc.
[0126] As described above, the disclosed embodiments have been explained with reference to the attached drawings. Those skilled in the art will understand that the present disclosure may be practiced in forms different from the disclosed embodiments without changing the technical spirit or essential features of the present disclosure. The disclosed embodiments are illustrative and should not be interpreted restrictively.
[0127] One embodiment of the present invention may also be implemented in the form of a recording medium comprising computer-executable instructions, such as program modules executed by a computer. A computer-readable medium may be any available medium accessible by a computer and includes both volatile and non-volatile media, and both removable and non-removable media. Additionally, a computer-readable medium may include all computer storage media. A computer storage medium includes both volatile and non-volatile, removable and non-removable media implemented by any method or technique for storing information, such as computer-readable instructions, data structures, program modules, or other data.
[0128] Although the method and system of the present invention have been described in relation to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.
[0129] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.
[0130] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols
[0131] 1: System 100: Server 200: User terminal
Claims
Claim 1 A method for supporting dynamic decision-making regarding insurance products based on artificial intelligence, performed by a processor, comprising: (a) acquiring user data including medical usage history, insurance contract information, and insurance claim history of the user and storing it in a database; (b) calculating the probability of future disease state transition of the user, the amount of future premium fluctuation for insurance contracts held by the user, the expected benefits and friction costs upon subscribing to comparable insurance, by a large-scale language model (LLM) referencing the database; The method comprises the steps of: (c) determining whether it is advantageous for the user to subscribe to the comparative insurance based on the information calculated above, and generating and outputting a behavioral suggestion based on the result of the determination; wherein step (c) is performed by calculating the net value when the user maintains the insurance contract held by the user and the net value when subscribing to the comparative insurance, respectively, and based on the difference between each net value, wherein the net value is calculated by subtracting the premium, deductible, and frictional cost from the expected benefit and dividing by the time discount rate, and the frictional cost is the cost predicted for the user to claim the insurance benefit but not receive it based on the probability of success of the insurance claim, and generating the behavioral suggestion by deriving one or more of a conversion signal indicating the degree of advantage of switching to the comparative insurance, a maintenance signal indicating the degree of advantage of maintaining the insurance contract, a supplementation signal indicating the degree of need to supplement coverage gaps, and an optimization signal indicating the degree of need to improve the insurance structure based on the difference between the net values. Claim 2 A method for supporting dynamic decision-making for an insurance product based on artificial intelligence, wherein, in claim 1, step (b) comprises: a step in which the large-scale language model generates parameter values from unstructured knowledge data including one or more of insurance terms and conditions related to insurance contract information, insurance-related press releases, and insurance-related precedents; a step in which the parameter values are corrected based on the difference between the generated parameter values and actual insurance payout result data stored in the database; and a step in which the probability of future disease state transition, the future premium change amount, the expected benefit, and the friction cost are calculated using the corrected parameter values, wherein the parameter values include at least one of payability, claim success rate, surcharge factor, and friction cost. Claim 3 A method for supporting dynamic decision-making for an insurance product based on artificial intelligence, wherein the probability of future disease state transition is calculated by applying the corrected parameter value to classify the user's current health state into one of a plurality of disease states, and reflecting the age, gender, and past diagnosis history included in the medical usage history to calculate the probability of transition from the current health state to the future health state over time. Claim 4 A method for supporting dynamic decision-making for an insurance product based on artificial intelligence, wherein, in paragraph 2, the future premium fluctuation amount is calculated by applying the corrected parameter value and reflecting a group renewal factor calculated based on the renewal conditions included in the insurance contract information and an individual surcharge factor calculated based on the insurance claim history to calculate the premium fluctuation amount over time of the insurance contract. Claim 5 In paragraph 2, the above-mentioned expected benefit is calculated by applying the above-mentioned corrected parameter value to estimate the probability of medical utilization occurring at each time point based on the probability of future disease state transition, and reflecting the actual payability probability in the payable amount according to the coverage conditions of the above-mentioned comparative insurance, and the above-mentioned friction cost is calculated by applying the above-mentioned corrected parameter value to calculate the probability of a successful insurance claim from the above-mentioned claim history, and calculating the cost predicted for the user claiming the insurance but not receiving it based on the above-mentioned probability of a successful insurance claim. This is a method for supporting dynamic decision-making regarding an insurance product based on artificial intelligence. Claim 6 delete Claim 7 A method for supporting dynamic decision-making for an insurance product based on artificial intelligence, wherein step (c) comprises generating a plurality of strategy signals including differences between each of the net values, combining the plurality of strategy signals to generate a strategy signal vector, and the strategy signal vector comprises one or more of a conversion signal indicating the degree of attractiveness of switching to the comparable insurance, a maintenance signal indicating the degree of attractiveness of maintaining the insurance contract, a supplementation signal indicating the degree of need to supplement coverage gaps, and an optimization signal indicating the degree of need to improve the insurance structure. Claim 8 In claim 7, the above step (c) classifies the user's current situation into one of a plurality of strategic states based on additional information including the strategy signal vector, the reliability of the calculated information, and the completeness of the user data, wherein the plurality of strategic states includes one or more of a maintenance advantage state, a switching advantage state, a state requiring supplementation, a state requiring optimization, and a state requiring additional information, a method for supporting dynamic decision-making for insurance products based on artificial intelligence. Claim 9 In claim 8, the above step (c) further comprises: a step of generating the action suggestion corresponding to the classified strategy state; and a step in which the large-scale language model converts the action suggestion into a natural language form including the basis for the judgment and specific actions actionable by the user and outputs it, wherein the action suggestion includes one or more of maintaining the insurance contract, switching to the comparable insurance, additional subscription to supplement coverage gaps, and optimizing the insurance structure, a method for supporting dynamic decision-making for an insurance product based on artificial intelligence. Claim 10 A method for supporting dynamic decision-making regarding an insurance product based on artificial intelligence according to claim 1, further comprising: (a-1) a step of receiving a request to query for an optimal insurance decision in the user's current health condition prior to step (b); and (d) a step of providing the action suggestion to the insured's terminal or providing the action suggestion to the insurance agent's terminal after step (c); wherein, when providing the action suggestion to the insurance agent's terminal, contract information and coverage scope information regarding the comparable insurance are provided together. Claim 11 A device for supporting dynamic decision-making regarding insurance products based on artificial intelligence comprises a memory storing a program that provides a method for supporting dynamic decision-making regarding insurance products based on artificial intelligence, and a processor that executes the program stored in the memory. The processor, by executing the program, acquires user data including medical usage history, insurance contract information, and insurance claim history of a user and stores it in a database. A large-scale language model (LLM) refers to the database to calculate the probability of future disease state transition of the user, the amount of future premium fluctuation for the insurance contract held by the user, and the expected benefits and friction costs upon subscribing to a comparable insurance. Based on the calculated information, the processor determines whether subscribing to the comparable insurance is advantageous to the user and executes a step of generating and outputting a behavioral suggestion based on the result of the determination. The process of generating and outputting a behavioral suggestion based on the result of the determination is performed by calculating the net value when the user maintains the insurance contract held by the user and the net value when subscribing to the comparable insurance, respectively, and based on the difference between each net value. The net value is calculated by subtracting the premium, deductible, and friction costs from the expected benefits and dividing by a time discount rate. The above friction cost is the cost predicted based on the probability of success of the insurance claim that the user has claimed the insurance but will not receive it, and the device for supporting dynamic decision-making for an insurance product based on artificial intelligence generates the above action suggestion by deriving one or more of a conversion signal indicating the degree of attractiveness of switching to the above comparative insurance based on the difference between the above net values, a maintenance signal indicating the degree of attractiveness of maintaining the above insurance contract, a supplementation signal indicating the degree of need to supplement a coverage gap, and an optimization signal indicating the degree of need to improve the insurance structure.
Citation Information
Patent Citations
Method and apparatus for recommanding customized insurance
KR1020220004543A
Method and device for calculating insurance premium
KR102830990B1
Method for forecasting vehicle insurance
KR1020190135399A
Method and apparatus of service for fitting insurance coverage based on integrated analysis of customer information
KR1020230001435A
Insurance platform service method, and system thereof
KR1020250118794A
