Business assistance system, business assistance method, and business assistance program

Abstract

To provide a business assistance system etc. which can efficiently acquire information relating to an insurance product.SOLUTION: A business assistance system comprises: a reception unit which receives inquiry from a user; a user identification unit which identifies the user on the basis of transmission source identification information used for transmission of the inquiry; an inquiry content acquisition unit which acquires an inquiry content of the user, on the basis of the inquiry; an inquiry content recognition unit which recognizes the acquired inquiry content; an inquiry destination selection unit which selects an inquiry destination for inquiring the recognized inquiry content, from a communication destination list stored in advance; a sentence generation unit which generates a sentence for informing the selected inquiry destination of the recognized inquiry content; and a sending unit which clearly indicates that it is the inquiry from the identified user, and sends the generated sentence to the selected inquiry destination.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a business support system, a business support method, and a business support program, and more particularly to a business support system, a business support method, and a business support program suitable for supporting the business of insurance sales representatives. [Background technology]

[0002] Insurance salespeople must have basic knowledge of insurance structures and types, as well as knowledge of laws such as the Insurance Business Act and the Enforcement Order of the Insurance Business Act, contract practices, compliance, and tax treatment of insurance premiums, insurance claims, and benefits.In addition, insurance salespeople must have knowledge of the insurance products they sell and the medical underwriting guidelines that the insurance products use.

[0003] Insurance agents, who act as intermediaries between insurance companies and customers and provide various insurance services to customers, have been increasing their share of the overall sales of insurance products in recent years. In addition, the number of customers who want to compare insurance products from multiple insurance companies before purchasing is increasing, and the number of insurance agents who handle insurance products from multiple insurance companies is also increasing. Insurance sales representatives at such insurance agents need to have knowledge about insurance products from multiple insurance companies.

[0004] Insurance companies strive to promote sales of their own insurance products and expand market share by supporting the overall business of insurance agents, including their sales activities. The people who work for insurance companies and are responsible for supporting insurance agents are called solicitors. Insurance sales representatives, with the support of solicitors, sell insurance products to customers.

[0005] Since the work of insurance salespeople requires a lot of knowledge, the development of IT (Information Technology) has promoted the use of information processing terminals such as PCs (Personal Computers) and smartphones to support the work of insurance salespeople. For example, Patent Document 1 proposes a portable sales activity support device that supports sales activities of insurance products conducted by insurance salespeople to customers. The technology disclosed in Patent Document 1 is said to effectively stimulate customers' desire to purchase insurance products, thereby enabling effective sales activities by insurance salespeople.

[0006] However, the technique disclosed in Patent Document 1 does not take into consideration how insurance salespeople can efficiently obtain information about insurance products. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Patent Publication No. 2007-80257 Summary of the Invention [Problem to be solved by the invention]

[0008] Therefore, an object of the present invention is to provide a business support system, a business support method, and a business support program that can efficiently obtain information about insurance products, thereby reducing the work burden on insurance sales representatives. [Means for solving the problem]

[0009] That is, the business support system of the first aspect is a business support system that supports a user's business, and is characterized by comprising: a reception unit that receives inquiries from a user; a user identification unit that identifies the user based on caller identification information used to send the inquiry received by the reception unit; an inquiry content acquisition unit that acquires the content of the user's inquiry based on the inquiry received by the reception unit; an inquiry content recognition unit that recognizes the inquiry content acquired by the inquiry content acquisition unit; an inquiry destination selection unit that selects an inquiry destination to inquire about the inquiry content recognized by the inquiry content recognition unit from a pre-stored contact list; a sentence generation unit that generates a sentence to inform the inquiry destination selected by the inquiry destination selection unit of the inquiry content recognized by the inquiry content recognition unit; and a sending unit that sends the sentence generated by the sentence generation unit to the inquiry destination selected by the inquiry destination selection unit, clearly indicating that the inquiry is from the user identified by the user identification unit.

[0010] In a second aspect, in the business support system according to the first aspect, an inquiry from a user is made by telephone, and the user identification unit identifies the user based on the telephone number used to make the inquiry call, and the inquiry content acquisition unit includes a voice acquisition unit that acquires voice related to the user's inquiry from a telephone answering device that responds to the inquiry call accepted by the reception unit, and a voice recognition unit that performs voice recognition on the voice acquired by the voice acquisition unit to acquire the inquiry content as text data, and the inquiry content recognition unit may recognize the inquiry content by analyzing the text data acquired by the voice recognition unit using natural language processing.

[0011] In a third aspect, in the business support system according to the first aspect, an inquiry from a user is made by email, the user identification unit identifies the user based on the email address used to send the inquiry, the inquiry content acquisition unit acquires text data included in the email, and the inquiry content recognition unit recognizes the inquiry content by analyzing the text data acquired by the inquiry content acquisition unit using natural language processing.

[0012] A fourth aspect is a business support system according to the first aspect, wherein the business support system includes a website for users to access, and inquiries from users are made via the website, the user identification unit identifies the user based on login information used by the user to log in to the website, the inquiry content acquisition unit acquires text data indicating the inquiry content entered by the user on the website, and the inquiry content recognition unit recognizes the inquiry content by analyzing the text data acquired by the inquiry content acquisition unit using natural language processing.

[0013] In a fifth aspect, in the business support system according to the first aspect, the sentence generation unit may generate sentences using a large-scale language model that is fine-tuned using text data relating to the insurance industry.

[0014] As for the sixth aspect, in the business support system according to the fifth aspect, the text data may include a data set of an exchange between the user and a person in charge at the insurance company and an annotation indicating an evaluation of the exchange.

[0015] In a seventh aspect, in the business support system of the first aspect, a first learning model that has learned information related to a user's business is prepared in advance, and the system may include an answer feasibility determination unit that determines whether the inquiry content recognized by the inquiry content recognition unit can be answered by the business support system, a first answer generation unit that uses the first learning model to generate an answer to the inquiry content recognized by the inquiry content recognition unit when the answer feasibility determination unit determines that an answer can be provided by the business support system, and an escalation unit that connects an inquiry from a user related to the inquiry content to a manned answer desk when the answer feasibility determination unit determines that an answer cannot be provided by the business support system.

[0016] An eighth aspect is a business support system according to the seventh aspect, in which a first specific document set comprising first documents is prepared in advance, and the system further includes a first document search unit that searches for and extracts first documents related to the query content from the first specific document set, and the first answer generation unit obtains the first documents searched by the first document search unit as the context of a prompt, and generates an answer to the query content.

[0017] A ninth aspect is a business support system according to the eighth aspect, in which a first document learning model that learns the correspondence between a query content and a first document related to the query content is prepared in advance, and the first document search unit uses the first document learning model to search for and extract a first document related to the query content from the first specified document set.

[0018] In a tenth aspect, in the business support system relating to the first aspect, a selection reception unit is further provided that receives an insurance company to be contacted selected by a user, and the contact selection unit may select the contact of the insurance company accepted by the selection reception unit from a contact list.

[0019] In an eleventh aspect, in the business support system related to the tenth aspect, a second learning model that has learned information about product information of an insurance product and medical underwriting guidelines adopted by the insurance product is prepared in advance for each insurance company, and the system further includes a second answer generation unit that generates an answer to the inquiry content recognized by the inquiry content recognition unit using the second learning model related to the insurance company of the inquiry destination accepted by the selection acceptance unit, and when the selection acceptance unit accepts the insurance company of the inquiry destination selected by the user, the second answer generation unit may generate an answer to the inquiry content recognized by the inquiry content recognition unit using the second learning model related to the insurance company of the inquiry destination accepted by the selection acceptance unit.

[0020] A twelfth aspect is a business support system according to the eleventh aspect, in which a second specific document set comprising second documents is prepared in advance for each insurance company, and the system further includes a second document search unit that searches for and extracts second documents related to the inquiry content from the second specific document set relating to the insurance company that is the subject of the inquiry and accepted by the selection acceptance unit, and the second answer generation unit obtains the second document searched by the second document search unit as the context of the prompt and generates an answer to the inquiry content.

[0021] In a thirteenth aspect, in the business support system relating to the twelfth aspect, a second document learning model that learns the correspondence between the inquiry content and the second document related to the inquiry content is prepared in advance for each insurance company, and the second document search unit uses the second document learning model related to the insurance company that is the destination of the inquiry accepted by the selection acceptance unit to search for and extract second documents related to the inquiry content from the second specific document set related to the insurance company that is the destination of the inquiry accepted by the selection acceptance unit.

[0022] A fourteenth aspect is a business support system according to the thirteenth aspect, in which, when a user selects multiple insurance companies as destinations for an inquiry, the selection reception unit accepts the multiple destination insurance companies selected by the user, and the second document search unit uses a second document learning model prepared in advance for each of the multiple insurance companies accepted by the selection reception unit to search for and extract, for each of the multiple insurance companies, a second document related to the inquiry content recognized by the inquiry content recognition unit from a second specific document set prepared in advance for each of the multiple insurance companies accepted by the selection reception unit. In a fifteenth aspect, in the business support system according to the first aspect, the user may be an insurance salesperson.

[0023] A business support method according to a sixteenth aspect is a business support method for supporting a user's business, characterized in that a computer executes the following steps: a reception step for receiving an inquiry from a user; a user identification step for identifying the user based on source identification information used to send the inquiry received in the reception step; a query content acquisition step for acquiring the content of the user's inquiry based on the inquiry received in the reception step; a query content recognition step for recognizing the query content acquired in the query content acquisition step; a query destination selection step for selecting from a pre-stored contact list a query destination to inquire about the query content recognized in the query content recognition step; a sentence generation step for generating a sentence to inform the query destination selected in the query destination selection step of the query content recognized in the query content recognition step; and a sending step for sending the sentence generated in the sentence generation step to the query destination selected in the query destination selection step, clearly indicating that the query is from the user identified in the user identification step.

[0024] A business support program according to a seventeenth aspect is a business support program that supports a user's business, and is characterized in that it causes a computer to perform the following functions: a reception function that receives inquiries from a user; a user identification function that identifies a user based on caller identification information used to send the inquiry received by the reception function; an inquiry content acquisition function that acquires the content of the user's inquiry based on the inquiry received by the reception function; an inquiry content recognition function that recognizes the inquiry content acquired by the inquiry content acquisition function; an inquiry destination selection function that selects an inquiry destination to inquire about the inquiry content recognized by the inquiry content recognition function from a pre-stored contact list; a sentence generation function that generates a sentence to inform the inquiry destination selected by the inquiry destination selection function of the inquiry content recognized by the inquiry content recognition function; and a sending function that sends the sentence generated by the sentence generation function to the inquiry destination selected by the inquiry destination selection function, clearly indicating that the inquiry is from the user identified by the user identification function. Effect of the Invention

[0025] The business support system etc. of the present invention is a business support system that supports a user's business, and includes a reception unit that receives inquiries from a user, a user identification unit that identifies a user based on source identification information used to send the inquiry received by the reception unit, an inquiry content acquisition unit that acquires the content of the user's inquiry based on the inquiry received by the reception unit, an inquiry content recognition unit that recognizes the inquiry content acquired by the inquiry content acquisition unit, an inquiry destination selection unit that selects an inquiry destination to inquire about the inquiry content recognized by the inquiry content recognition unit from a pre-stored contact list, a text generation unit that generates a text to inform the inquiry destination selected by the inquiry destination selection unit of the inquiry content recognized by the inquiry content recognition unit, and a sending unit that sends the text generated by the text generation unit to the inquiry destination selected by the inquiry destination selection unit, clearly indicating that the inquiry is from the user identified by the user identification unit.Therefore, information regarding insurance products can be acquired efficiently, and ultimately the user's workload can be reduced. [Brief description of the drawings]

[0026] [Figure 1] FIG. 1 is a diagram showing an example of an infrastructure configuration of a business support system according to the present embodiment. [Diagram 2] FIG. 2 is a block diagram showing an example of a hardware configuration of the business support system according to this embodiment. [Diagram 3] FIG. 3 is a block diagram showing an example of the functional configuration of the business support system according to the present embodiment. [Figure 4] FIG. 4 is a diagram showing an example of a display screen of the monitor of the user terminal according to this embodiment. [Diagram 5] FIG. 5 is a diagram showing an example of a display screen on the monitor of a user terminal according to this embodiment. [Figure 6] FIG. 6 is a diagram showing an example of a display screen on the monitor of a user terminal according to this embodiment. [Figure 7]FIG. 7 is a diagram showing an example of a display screen of the monitor of the user terminal according to this embodiment. [Figure 8] FIG. 8 is a diagram showing an example of a display screen on the monitor of the user terminal according to this embodiment. [Figure 9] FIG. 9 is a flowchart showing an example of the business support program according to this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] (Overview of Business Support System 10) A task support system 10 according to an embodiment of the present disclosure will be described with reference to FIGS. First, an example of an infrastructure configuration that is a usage environment of a task support system 10 according to the present embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of an infrastructure configuration of the task support system 10 according to the present embodiment.

[0028] The business support system 10 is connected via the Internet 31 to a telephone answering device 45, a generation AI server 46, a user terminal 49, a web server 50, an insurance company server 60, and the like so as to enable two-way communication. The business support system 10 is a type of information processing device represented by a server, a personal computer (hereinafter, PC), a notebook PC, etc., and provides various information to a user terminal 49 used by a user by running a business support program described below.

[0029] The user is a sales representative of insurance products, and may include, but is not limited to, those directly or indirectly involved in the sale of insurance products, such as insurance solicitors and those who assist the insurance solicitors. By using the business support system 10, the user can obtain information about insurance products and receive support for the user's business from the business support system 10. The user is not limited in terms of affiliation, and may be, for example, an employee of an insurance company, an employee of an insurance agency, an employee of an insurance consultation desk, or a sole proprietor. An insurance agent is a person who introduces, proposes, and sells insurance products at insurance companies, insurance agencies, and insurance consultation centers. Insurance agents are divided into non-life insurance agents and life insurance agents depending on the type of insurance product they handle.

[0030] The user may be a general consumer who is considering purchasing an insurance product. A user who is an ordinary consumer can inquire about the contents of insurance products across multiple insurance companies, regardless of the insurance company, to the business support system 10, and a user who is considering purchasing an insurance product can obtain useful information from the business support system 10. For example, a user can request the business support system 10 to provide information on insurance products that meet the conditions (insurance premiums, benefit amounts, special clauses, etc.) he or she expects regarding medical insurance. In this case, the business support system 10 may provide information on insurance products that are suitable for the user by ranking them.

[0031] The user terminal 49 is an information processing terminal used by a user who is to be supported by the business support system 10, and is, for example, a PC, a notebook PC, a tablet PC, a smartphone, etc. The user terminal 49 connects to the business support system 10 via the Internet 31 and receives support such as various information provided by the business support system 10.

[0032] The telephone response device 45 is an IVR (Interactive Voice Response: telephone automatic response system), is connected to the business support system 10 via the Internet 31, and is used for responding to calls from the user telephone terminal 48, and the like. The telephone answering device 45 may be a PBX (Private Branch eXchange) or may be a manned telephone help desk that connects a user's call to an internal line and answers the call by a man. In this case, the PBX is connected to the business support system 10 via the Internet 31, and the business support system 10 acquires the voice of the manned telephone help desk answering the user's call. The telephone response device 45 may be provided as one functional unit of the business support system 10 by modularizing its functions and incorporating it into the business support system 10 .

[0033] The user telephone terminal 48 is a so-called telephone set, and is used for calling the telephone answering device 45 via the public telephone line network 47 and for voice input and predetermined operations by the user prompted by the voice guidance of the telephone answering device 45. The user terminal 49 and the user telephone terminal 48 may be a portable information terminal, such as a smartphone, which is an integrated device that combines information processing functions and telephone functions.

[0034] The generating AI server 46 is equipped with a generating AI (generative artificial intelligence) and provides an API (Application Programming Interface) of the generating AI to an application connected from outside. Generative AI is an artificial intelligence system that can generate various content, and can generate content (media) such as text, images, and audio in response to input (instructions / commands: prompts). A prompt is a command statement that gives instructions to the generative AI. There are various types of generation AI, such as ChatGPT (registered trademark) 3.5, ChatGPT 4.0, and Bard (registered trademark), and they are built based on a large language model (LLM).

[0035] ChatGPT is a conversational AI system (chatbot) based on GPT (Generative Pre-trained Transformer), and Bard is a conversational AI system based on PaLM (Pathways Language Model). These conversational AI systems receive instructions and commands (prompts) in text data as input and output answers in text data.

[0036] Instead of accessing the generation AI server 46 to use the generation AI, the business support system 10 may incorporate the generation AI as one of its functional units. The large-scale language model constituting the generation AI can be made lighter by using, for example, a technique called LoRA (low-rank adaptation), which can reduce the parameters by 1 / 10,000 and reduce the required memory capacity of the GPU (Graphics Processing Unit) by 1 / 3. For example, Alpaca-LoRA is a large-scale language model that has been made lighter by using LoRA from Alpaca, and can be run on a notebook PC. Alpaca is a large-scale language model called LLaMA (Large Language Model Meta AI) released by Facebook (registered trademark) that has been fine-tuned by Stanford University.

[0037] The web server 50 is connected to the business support system 10 via the Internet 31, and is used for storing web server software used to operate a website (described later) that is accessible by the user terminal 49, and for storing configuration files of the website. The web server 50 may be built into the business support system 10 as one of its functional parts.

[0038] The insurance company server 60 is a server operated by an insurance company, and each insurance company has its own server and operates it independently. The insurance company server 60 may have a web server function used to publish a homepage, etc., and a mail server function when used for sending and receiving mail, etc. The insurance company server 60 is also used to disclose information about the insurance products of the insurance company, and users can access the insurance company server 60 to obtain materials about the insurance products of the insurance company in file formats such as PDF (registered trademark).

[0039] In addition, for convenience, Figure 1 shows one each of the telephone answering device 45, generation AI server 46, user terminal 49, web server 50, and insurance company server 60, but this is not limited to this, and multiple telephone answering devices 45, generation AI servers 46, user terminals 49, web servers 50, and insurance company servers 60 may be connected to the business support system 10 via the Internet 31.

[0040] (Hardware configuration of Business Support System 10) Next, the hardware configuration of the task support system 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a block diagram showing an example of the hardware configuration of the task support system 10.

[0041] The business support system 10 includes a communication interface 10a, a ROM (Read Only Memory) 10b, a RAM (Random Access Memory) 10c, a storage unit 10d, a CPU (Central Processing Unit) 10e, and an input / output interface 10f. Furthermore, a dedicated computing element (processor) specialized for the generation AI may be implemented in place of the CPU 10e or together with the CPU 10e.

[0042] The communication interface 10a has a function of transmitting and receiving data of the business support system 10 to and from other devices via the Internet 31. The data of the business support system 10 may include voice at the time of inquiry when the user makes an inquiry by telephone, email used for the inquiry when the user makes an inquiry by email, and input information entered on the website when the user makes an inquiry via the website of an insurance company.

[0043] A user's inquiry refers to an inquiry made to the business support system 10 by a user about matters that the user wishes to know about in the course of business, instead of asking an insurance company or the like, and refers to a request from the user to the business support system 10 to receive support for the user's business. The business support system 10 supports the user's business by answering the user's inquiry or by notifying the contact point of the user of the content of the user's inquiry. Specifically, a user's inquiry refers to a user's question or confirmation of information about an insurance product, etc., made to the business support system 10, similar to inquiries made to solicitors by users in the past. For example, a user's inquiry refers to a question or confirmation made to the business support system 10 about information about an insurance product, products associated with the insurance product, and medical underwriting guidelines adopted by the insurance product. Further, a user inquiry means a request for a proposal from the business support system 10. Specifically, a user inquiry means a request for a proposal regarding insurance products and products associated with insurance products from the business support system 10, and for example, when a user's customer is considering purchasing education insurance, the user's inquiry means a request to the business support system 10 for a proposal regarding which insurance company's education insurance should be recommended to the customer, regarding education insurance suitable for the customer. Moreover, a user inquiry means a request from the business support system 10 for information that is necessary for the user to carry out business, regardless of the insurance company.

[0044] In addition, a user's inquiry refers to a question to the business support system 10, such as, for example, whether there is an age limit for enrolling in cancer insurance, up to what age one can enroll in advanced medical insurance, or how much the annual premium will be if the amount of fire insurance coverage is 30 million yen. Another example of a user inquiry is an inquiry about the merits and demerits of each insurance company's automobile insurance compared to other companies. Instead of making such an inquiry to each insurance company, the user can make the inquiry to the business support system 10 and ask the questions to the persons in charge of each insurance company at the same time from the business support system 10.

[0045] In addition, a user's inquiry refers to a general consumer inquiring of the business support system 10 about questions that arise when the general consumer considers purchasing an insurance product and requesting an answer to the question. Specifically, the user does not know which insurance product to select from among many insurance companies' insurance products, and requests the business support system 10 to propose an insurance product suitable for the user. For example, when a user considers purchasing medical insurance and wants to know which insurance product from which insurance company is suitable for the user, the user inputs personal information such as the user's age, sex, and health condition, as well as information such as the conditions for purchasing insurance that the user can accept (insurance premium, benefit amount, coverage period, special provisions, etc.) into the business support system 10, whereby the user can receive an answer about medical insurance suitable for the user from the business support system 10. The answer from the business support system 10 may be an orderly answer of medical insurance from multiple insurance companies in order of suitability for the user.

[0046] When consumers apply for online insurance, they can only receive a response regarding whether or not they can apply for insurance from the online insurance company they applied for, even if they are able to apply for other online insurance. Online insurance refers to insurance where procedures such as premium estimates and application for insurance can be completed online. In addition, when a general consumer consults an insurance agency, insurance consultation center, etc. about purchasing an insurance product, the general consumer may be limited to selecting an insurance product that suits him / herself from among the insurance products handled by the insurance agency, insurance consultation center, etc. As described above, conventionally, general consumers have had the fear that the insurance products they can purchase depend on where they go for insurance consultation or application. The business support system 10 learns information about insurance products of all insurance companies that can be obtained, and is therefore able to respond to inquiries from general consumers based on information about insurance products of all insurance companies.

[0047] In addition, a user's inquiry means a request to the business support system 10 to review the contents of an insurance contract when a general consumer subscribes to an insurance product. A contract review means verifying from a legal perspective whether the contents of the agreement in the contract are clear and whether the contract poses any unexpected hidden risks to the user. When reviewing the contract, the business support system 10 grasps the contract contents of the insurance contract and judges whether the contract contains the risk of the general consumer, who is the user. If the business support system 10 judges that the insurance contract contains the user's risk, it creates a proposal to amend the insurance contract and judges whether the proposed amendment contains the user's risk.

[0048] Understanding the contents of an insurance contract means recognizing the type of insurance and its contents, as well as the premiums, benefit amounts, coverage period, and whether or not there are special clauses. The risks include insurance premiums not being set appropriately for the insurance content, clauses being included that are disadvantageous to the user, and clauses that are insufficient and do not cover all the clauses that the user needs. The creation of a revised insurance contract includes correcting typos, correcting unclear sentences to make them clear, and deleting or amending clauses that are disadvantageous to the user.

[0049] In addition to insurance companies, users may contact government agencies related to the insurance industry (such as the Financial Services Agency, police, and fire departments) and local governments. A solicitor is a person who conducts sales activities for insurance agents of insurance companies. Solicitors provide support to people working at insurance agents, such as by providing information about their company's insurance products, and strive to promote sales of their company's insurance products. Conventionally, insurance product salespeople such as insurance solicitors had to confirm the details of insurance products with different solicitors for each insurance company. Furthermore, insurance products handled by different solicitors may differ, and insurance product salespeople such as insurance solicitors had to confirm the details of insurance products with different solicitors for each insurance product. The present disclosure proposes a business support system 10, as well as a business support method and a business support program described below, in order to eliminate the hassle faced by insurance product sales representatives such as insurance solicitors, and users can obtain the necessary information about insurance products from each insurance company simply by inquiring about the business support system 10. As for the form of a user inquiry, the user notifies the business support system 10 of the content of the inquiry by telephone, e-mail, or using a website.

[0050] The storage unit 10d can be used as a storage device for the business support system 10, and is composed of, for example, a hard disk drive, a solid state drive, and a flash memory, and the storage unit 10d can also be composed of cloud storage. The storage unit 10d may also be composed of a database. By configuring the storage unit 10d as a database, it is possible to efficiently manage huge amounts of data such as a contact list described below, and it is possible to quickly refer to required information with excellent searchability and accessibility.

[0051] The contact list is a list of email addresses and telephone numbers of people to be contacted regarding the contents of the user's inquiry, and at least one of them is registered. The people to be contacted refer to people who belong to the insurance company, such as employees of the insurance company, and it is preferable that they have sufficient knowledge of the insurance product to be inquired about, such as people who belong to the department that handles the insurance product. The telephone numbers in the contact list are used when the inquiry contents are sent to the person to be visited using SMS (Short Message Service). The telephone numbers in the contact list are also used when the sending unit 26, which will be described later, conveys the inquiry contents to the person to be inquired about by voice.

[0052] The storage unit 10d of the business support system 10 stores a business support program (to be described later), an OS (Operating System) required for the business support system 10 to operate, various applications, and various data used by the applications. The OS is a type of software built into the business support system 10 and has a function of managing basic control of the business support system 10.

[0053] The business support system 10 stores a business support program, which will be described later, in the ROM 10b or the storage unit 10d, and loads the business support program into a main memory, which is constituted by the RAM 10c, etc. The CPU 10e accesses the main memory into which the business support program has been loaded, and executes the business support program.

[0054] The input / output interface 10f transmits and receives data to and from devices external to the business support system 10. The external devices are an input device 10g and an output device 10h that input and output data to and from the business support system 10. The input device 10g is, for example, a keyboard and a mouse, and the output device 10h is, for example, a monitor, a printer, and a speaker.

[0055] (Functional configuration of business support system 10) The functional configuration of the task support system 10 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing an example of the functional configuration of the task support system 10.

[0056] By executing the business support program, the business support system 10 provides the CPU 10e with functional units such as a selection reception unit 11, a reception unit 12, a user identification unit 13, an inquiry content acquisition unit 14, a voice acquisition unit 15, a voice recognition unit 16, an inquiry content recognition unit 17, an inquiry destination selection unit 18, an answer feasibility determination unit 19, a first answer generation unit 20, a first document search unit 21, an escalation unit 22, a second answer generation unit 23, a second document search unit 24, a sentence generation unit 25, and a sending unit 26.

[0057] The selection receiving unit 11 receives the insurance company to which the inquiry is directed selected by the user. When the user selects a plurality of insurance companies as the destinations of the inquiry, the selection receiving unit 11 receives the plurality of destination insurance companies selected by the user. The monitor 32 of the user terminal 49 displays a page for the user to select an insurance company to inquire about. The business support system 10 provides a display screen configured with HTML (Hyper Text Markup Language) or the like to the monitor 32 of the user terminal 49 connected via the Internet 31 (see Figs. 4 to 8).

[0058] Note that the types of insurance that a user can handle may be limited. For example, if a user is a life insurance agent but not a non-life insurance agent, the user can handle life insurance but not non-life insurance. Similarly, if a user is a non-life insurance agent but not a life insurance agent, the user can handle non-life insurance but not life insurance. Furthermore, if a user is both a life insurance agent and a non-life insurance agent, the user can handle both life insurance and non-life insurance. Only insurance products that the user can handle are displayed on the monitor 32 of the user terminal 49, and the user can select from among the insurance products displayed on the monitor 32.

[0059] (Regarding the display screen of the monitor 32 of the user terminal 49) The display screen of the monitor 32 of the user terminal 49 will be described with reference to Fig. 4 to Fig. 8. Fig. 4 to Fig. 8 are diagrams showing an example of the display screen of the monitor 32 of the user terminal 49.

[0060] 4 and 5 show the display screen of the monitor 32 of the user terminal 49 when the user makes an inquiry to the insurance company. A display switching button 33 is provided at the top of the display screen displayed on the monitor 32. The display switching button 33 includes a medical underwriting guideline selection button 34, a pamphlet selection button 35, and an inquiry selection button 36.

[0061] When the user wishes to make an inquiry to an insurance company, the user presses the inquiry selection button 36 by mouse or touch operation. A touch operation refers to a pointing operation performed by touching the monitor 32 with a finger or a pen. A touch operation can be performed by directly pointing at an icon or button on the monitor 32 with a finger. The monitor 32 displays an inquiry content entry field 44 and an insurance company selection screen display button 43 above it. The inquiry content entry field 44 is where the user enters the content of the inquiry he or she wishes to make as text. The input method is typing on a keyboard, character input by touch operation, or input by voice recognition. The inquiry content acquisition unit 14 described below acquires the text entered in the inquiry content entry field 44 as the user's inquiry content.

[0062] When insurance company selection screen display button 43 is pressed with a mouse or touch operation, insurance company list 37 is displayed on monitor 32 (see FIG. 5). When the user checks check box 38 of the insurance company for which he or she wants a response to his or her inquiry, inquiry destination selection unit 18, which will be described later, selects the inquiry destination of the insurance company for which check box 38 is checked from a pre-stored contact list. The user can select multiple insurance companies.

[0063] FIG. 6 shows the display screen of the monitor 32 of the user terminal 49 when the medical underwriting guidelines of the insurance company that the user wants to view are displayed on the monitor 32. The medical underwriting guidelines refer to business information provided by the insurance company to the insurance agency as a guideline for determining whether or not to enroll in insurance. When the user presses the medical underwriting guidelines selection button 34 by mouse or touch operation, the insurance company list 37 is displayed on the monitor 32. The user checks the check box 38 of the insurance company of the medical underwriting guidelines that the user wants to view. When the user presses the confirm button (not shown) on the display screen by mouse or touch operation, the medical underwriting guidelines of the selected insurance company are displayed on the monitor 32 and can be viewed. Furthermore, when the user presses the confirm button by mouse or touch operation, the download of a PDF file of the medical underwriting guidelines may be started from the insurance company server 60 of the selected insurance company. The user can select multiple insurance companies.

[0064] 7 and 8 show the display screen of the monitor 32 of the user terminal 49 when the insurance brochure of the insurance company that the user wants to view is displayed on the monitor 32. When the user presses the brochure selection button 35 by mouse or touch operation, an insurance company selection screen display button 39 and an insurance type selection screen display button 40 are displayed on the monitor 32.

[0065] When the user presses an insurance company selection screen display button 39 by mouse or touch operation, an insurance company list 37 is displayed on the monitor 32. The user checks the check box 38 of the insurance company whose pamphlet the user wishes to view. Next, the user presses an insurance type selection screen display button 40 by mouse or touch operation, whereby an insurance type list 41 is displayed on the monitor 32. The user checks a check box 42 for the type of insurance in the pamphlet that the user wishes to view. The user can select multiple insurance companies.

[0066] By pressing a Confirm button (not shown) on the display screen with a mouse or touch operation, the user can display and view a pamphlet of the selected insurance company and the selected type of insurance on the monitor 32. Furthermore, by pressing the Confirm button with a mouse or touch operation, the user may start downloading a PDF file of the pamphlet of the selected insurance company and the selected type of insurance. In addition, by pressing the Confirm button with a mouse or touch operation, the user may be able to request the insurance company to mail the pamphlet of the selected insurance company and the selected type of insurance to the insurance company via the insurance company server 60.

[0067] The reception unit 12 shown in FIG. 3 receives inquiries from users. If the user's inquiry is made by telephone, the reception unit 12 acquires call information from the telephone answering device 45 indicating that the user has made an inquiry call. The telephone may be an analog telephone or digital telephone using a public telephone line network, or an IP telephone using an Internet line. An analog telephone is a calling method in which voice is transmitted directly over telephone lines. A digital telephone is a calling method in which voice is converted into a digital signal and transmitted over telephone lines. An IP telephone is a calling method that uses VoIP (Voice over Internet Protocol) technology. When the inquiry from the user is made by e-mail, the reception unit 12 acquires mail reception information indicating that the inquiry e-mail has been received from the user. When the user's inquiry is made via the website of the business support system 10, the reception unit 12 acquires forwarding information indicating that the user's inquiry has been forwarded from the website.

[0068] The user identification unit 13 identifies the user based on the sender identification information used to send the inquiry from the user that was accepted by the acceptance unit 12 . When the user makes an inquiry by telephone, the caller identification information refers to the telephone number used to make the user's inquiry, and the user identification unit 13 identifies the user based on the telephone number used to make the user's inquiry. A telephone number list in which the telephone numbers of all users of the business support system 10 are registered is prepared in advance, and the user identification unit 13 refers to this telephone number list to identify the user based on the telephone number used to make the call. The telephone number list is stored in the memory unit 10d. When a user makes an inquiry by email, the sender identification information refers to the email address used for the user's inquiry, and the user identification unit 13 identifies the user based on the email address used for sending the user's inquiry. Note that an email address list in which the email addresses of all users of the business support system 10 are registered is prepared in advance, and the user identification unit 13 refers to this email address list to identify the user based on the email address used for the inquiry. The email address list is stored in the memory unit 10d. When an inquiry from a user is made via the website of the business support system 10, the caller identification information refers to login information used by the user when logging in to the website, and the user identification unit 13 identifies the user based on the user's login information. Login information refers to information that can distinguish a user from other users, such as a user name, email address, telephone number, or employee number used as the user's ID (identification). A login information list in which the login information of all users of the business support system 10 is registered is prepared in advance, and the user identification unit 13 refers to this login information list to identify the user based on the login information used for the login. The login information list is stored in the memory unit 10d.

[0069] The inquiry content acquisition unit 14 acquires the content of the user's inquiry based on the inquiry from the user accepted by the acceptance unit 12 . The inquiry content acquisition unit 14 includes a voice acquisition unit 15 and a voice recognition unit 16. When an inquiry from a user is made by telephone, the inquiry content acquisition unit 14 acquires the contents of the user's inquiry by the voice acquisition unit 15 and the voice recognition unit 16.

[0070] The voice acquisition unit 15 acquires voice related to a user's inquiry from the telephone response device 45 that has responded to a call from the user that has been accepted by the acceptance unit 12 as an inquiry. When the telephone answering device 45 is an IVR, the voice acquiring unit 15 acquires the voice when the call is answered by an automatic telephone answering device. When the telephone answering device 45 is a PBX, the voice acquiring unit 15 acquires the voice when the call is answered by a manned telephone help desk.

[0071] Next, the voice recognition unit 16 performs voice recognition on the voice acquired by the voice acquisition unit 15 and acquires the inquiry content as text data. The voice recognition unit 16 converts the voice acquired by the voice acquisition unit 15 into text data by automatic speech recognition (ASR) and acquires the text data.

[0072] When the user inquiry is made by e-mail, the inquiry content acquisition unit 14 acquires text data included in the e-mail related to the user inquiry.

[0073] When an inquiry from a user is made via the website of the business support system 10, the inquiry content acquisition unit 14 acquires text data indicating the inquiry content entered by the user on the website of the business support system 10. Specifically, the inquiry content acquisition unit 14 acquires text data entered by the user in the inquiry content entry field 44 displayed on the monitor 32 of the user terminal 49 (see FIG. 4).

[0074] The inquiry content recognition unit 17 recognizes the inquiry content acquired by the inquiry content acquisition unit 14 . When a user makes an inquiry by telephone, the inquiry content recognition unit 17 recognizes the content of the inquiry by using natural language processing to analyze the text data acquired by the voice recognition unit 16. Recognition here refers to a computer understanding human language (natural language) in order to respond to or generate a response to the human language. Natural language processing (NLP) is a technology that uses computers to process and analyze the language used by humans (natural language).

[0075] When a user inquiry is made by e-mail or via the website of the business support system 10, the inquiry content recognition unit 17 recognizes the content of the user inquiry by analyzing the text data acquired by the inquiry content acquisition unit 14 using natural language processing.

[0076] The inquiry destination selection unit 18 selects an inquiry destination for making an inquiry about the inquiry content recognized by the inquiry content recognition unit 17 from a contact list stored in advance. The inquiry destination selection unit 18 selects, from a pre-stored contact list, the inquiry destination of the insurance company accepted by the selection acceptance unit 11. The contact list is pre-stored in the storage unit 10d. Note that the inquiry destination selection unit 18 may select multiple inquiry destinations, and if there are multiple insurance companies from which the user wishes to receive a response, multiple inquiry destinations can be selected.

[0077] When the user inquiry is made by telephone, the selection receiving unit 11 receives the user's desired inquiry destination from the inquiry content recognition unit 17 which recognizes the text data related to the user's inquiry acquired by the voice acquisition unit 15 and the voice recognition unit 16.

[0078] When the inquiry from the user is made by e-mail, the selection receiving unit 11 receives the inquiry destination desired by the user from the inquiry contents recognized by the inquiry content recognition unit 17.

[0079] If an inquiry from a user is made via the website of the business support system 10, the selection acceptance unit 11 accepts the contact information of the insurance company that the user has checked in the check box 38 of the insurance company list 37 displayed on the monitor 32 of the user terminal 49 as the insurance company selected by the user to contact (see Figure 5).

[0080] The answer possibility determination unit 19 determines whether the inquiry content recognized by the inquiry content recognition unit 17 can be answered by the business support system 10 or not. The answer possibility determination unit 19 uses a possibility learning model that has been machine-learned in advance to determine the correspondence between the inquiry content and what the business support system 10 can or cannot answer, to determine whether the inquiry content recognized by the inquiry content recognition unit 17 can be answered by the business support system 10. The possibility learning model is stored in the storage unit 10d. A dataset of supervised learning data used for machine learning of the possible / possible learning model is created as follows. When the inquiry content recognized by the inquiry content recognition unit 17 is something that the business support system 10 can answer, an annotation is performed to add metadata that the business support system 10 can answer the inquiry content, thereby creating a set of datasets. When the inquiry content recognized by the inquiry content recognition unit 17 is something that the business support system 10 cannot answer, an annotation is performed to add metadata that the business support system 10 cannot answer the inquiry content, thereby creating a set of datasets. In this way, two types of datasets obtained by two types of annotation are prepared in large numbers, and machine learning is performed on the possible / possible learning model. Alternatively, a list of cases in which the business support system 10 cannot respond may be created in advance, and the answer availability determination unit 19 may refer to the list and determine whether or not the inquiry content recognized by the inquiry content recognition unit 17 corresponds to an inquiry in which the business support system 10 cannot respond. The list is stored in the storage unit 10d. For example, business support system 10 cannot respond to inquiries whose response includes personal information, inquiries whose response would violate laws such as the Insurance Business Act and the Enforcement Order of the Insurance Business Act, and inquiries whose content can only be judged by an insurance agent.

[0081] When the answer feasibility determination unit 19 determines that an answer is possible using the business support system 10, the first answer generation unit 20 uses the first learning model to generate an answer to the inquiry content recognized by the inquiry content recognition unit 17. A first learning model that has learned information about the user's work is prepared in advance. The information about the user's business learned by the first learning model is general information about the user's business, and refers to information that is necessary to carry out business regardless of the insurance company.

[0082] The first learning model is a generative AI that is constructed by providing training data to a pre-trained large-scale language model and fine-tuning it so that it can be applied to the tasks of the first learning model. Fine-tuning refers to making fine adjustments to make a neural network model suitable for a specific task. Specifically, it refers to fixing the lower layers (low-level features on the input side) of a pre-trained neural network model using a dataset at hand to apply the model to a specific task, and re-training some or all of the upper layers (high-level features on the output side), thereby fine-tuning the parameters of the neural network model and generating a neural network model suitable for a specific task.

[0083] A task refers to the work (processing) required to achieve a specific goal, and the task in the first learning model is to answer questions about general information about the user's business. General information about the user's business refers to information about how to use the computer system used in the business, laws related to the business such as the Insurance Business Act and the Enforcement Order of the Insurance Business Act, contract practices, compliance, and laws related to insurance premiums, insurance claims, and benefits. Computer systems used in the business refer to, for example, customer management systems, exam application and qualification information systems, and educational content provision systems. In other words, the task of the first learning model is to answer questions about information that is necessary to carry out work regardless of the insurance company. Therefore, the first learning model is fine-tuned for a pre-trained large-scale language model based on information data on how to use the computer systems used in business, laws related to the business such as the Insurance Business Act and the Insurance Business Act Enforcement Ordinance, contractual practices, compliance, and laws related to insurance premiums, insurance claims, and benefits.

[0084] Moreover, the task of the first learning model is to answer inquiries from users who are general consumers who are considering purchasing an insurance product. In this case, the first learning model is fine-tuned based on data such as product information for all available insurance companies' insurance products (such as product brochures and information posted on the insurance company's website) in order to respond to inquiries about insurance products from general consumers.

[0085] In addition, the task of the first learning model is to conduct a contract review of the contents of the insurance contract when general consumers sign up for an insurance product. In this case, the first learning model is fine-tuned based on the data of the contract contents, legal information related to the conclusion of the contract, specifically, the wording and clauses of the contract, and related legal provisions and precedents, etc., for the pre-trained large-scale language model. This information is used by the first learning model to accurately analyze the contract and extract legal issues. In addition, the first learning model constantly learns new information, and the latest legal information is reflected. By utilizing the contract review function of the business support system 10, a user can extract and identify in advance any content of the insurance contract that is disadvantageous to the user and any risks of disputes before entering into a contract with an insurance company.

[0086] In addition, measures to prevent hallucination are necessary when using generation AI. Hallucination occurs when a generative AI using large-scale language models to process natural language such as text generation and language translation generates answers that are inconsistent with the facts in response to prompts. Hallucination can occur due to incorrect training of data, incorrect interpretation of training data, combination of different training data, and guesswork due to insufficient training data. In order to prevent or suppress business negligence and losses caused by a user performing business based on an answer that is different from the facts, the first answer generation unit 20 may indicate the information source of the generated answer to the user together with the generated answer. The user can confirm the facts of the answer of the first answer generation unit 20 by checking the information source. In addition, improving the quality and increasing the quantity of training data can suppress the occurrence of hallucination itself.

[0087] It is also known that the quality of the answers generated by the generation AI varies greatly depending on how the instructions / commands (prompts) are given. Therefore, more appropriate instructions / commands (prompts) are required for the first answer generation unit 20. The instructions / commands (prompts) for the first answer generation unit 20 are generated based on the inquiry content recognized by the inquiry content recognition unit 17, but the contents of the prompts are required to be specific and clear, and further, it is desirable for the prompts to include four elements: instructions / commands, background / context, input data, and output format.

[0088] The four elements of a prompt are explained below. An instruction is a command or task that a generating AI should carry out. Background / context refers to the background and external information that you want the generative AI to take into account in order to improve the quality of the output (answer). In addition, since conversations (chat) with the generative AI are stateless, the prompts must include the history of past conversations each time. Stateless refers to a method in which the system does not retain data that represents the current state, and the output is determined only by the contents of the input. Therefore, the output (answer) of the generative AI will be different between a prompt that includes the history of past conversations and a prompt that does not include the history of past conversations. The input data refers to the question that the generation AI is desired to answer, and refers to the inquiry content recognized by the inquiry content recognition unit 17. The output indicator refers to the format in which you want the output to be generated; specifically, it refers to the items, format, number, range, etc. that you want the generation AI to output.

[0089] A first specific document set that is a collection of first documents is prepared in advance. The primary document is an electronic document related to general information about the user's business, specifically, an electronic document related to information that the user needs to carry out the business, regardless of the insurance company. For example, the primary document is an electronic document related to the method of using the computer system used in the business, laws related to the business such as the Insurance Business Act and the Enforcement Order of the Insurance Business Act, contract practices, compliance, and laws related to insurance premiums, insurance claims, and benefits. An electronic document refers to a document created as digital information, and further refers to a document that was not originally an electronic document but has been converted into an electronic document. The first specified document set is a collection of the above-mentioned first documents, and is stored in advance in the storage unit 10d.

[0090] The first document is an electronic document that can be used as a reference by general consumers who are considering purchasing an insurance product. Specifically, the first document includes practical books, business books, self-help books, etc. related to the insurance industry and insurance products. The first document is an electronic book that is useful for insurance reviews of insurance contracts when general consumers apply for insurance products. Specifically, the first document includes electronic documents written by experts (people with specialized knowledge) regarding insurance contracts.

[0091] The first document search unit 21 searches for and extracts first documents related to the inquiry content recognized by the inquiry content recognition unit 17 from the first specified document set. A first document learning model that learns the correspondence between the inquiry content recognized by the inquiry content recognition unit 17 and the first document related to the inquiry content is prepared in advance. The first document search unit 21 uses the first document learning model to search for and extract first documents related to the inquiry content recognized by the inquiry content recognition unit 17 from the first specified document set. The first document learning model is a learning model that searches for and extracts documents related to a given text. Specifically, it is a learning model that has previously learned the correspondence between the query content recognized by query content recognition unit 17 and the first document related to the query content, and searches for and extracts the first document related to the query content recognized by given query content recognition unit 17 from the first specified document set.

[0092] The first document related to the inquiry content recognized by the inquiry content recognition unit 17 may be the first document having the highest similarity to the inquiry content recognized by the inquiry content recognition unit 17. In addition, the first document related to the inquiry content recognized by the inquiry content recognition unit 17 is the first document having the highest similarity to the inquiry content recognized by the inquiry content recognition unit 17, and may be more than one. The first document learning model is generated by machine learning using a data set consisting of a large number of combinations of the query content recognized by the query content recognition unit 17 and the first document having the highest similarity to the query content as learning data.

[0093] . The first document learning model may also calculate the similarity of each of the first documents included in the first specified document set to the query content recognized by the query content recognition unit 17, and search and extract the first document with the highest similarity from the first specified document set. In this case, as an example of calculating the similarity between two documents, the first document learning model may convert each of the two documents into a numerical vector and calculate the similarity between the two documents using a measure of COS similarity. The COS similarity takes a value in the range of -1 to 1, and the larger it is, the more similar the two documents are. For example, BoW (Bag of Words) and TF-IDF (Term Frequency-Inverse Document Frequency) are used to convert a text into a numerical vector. BoW is a method of counting the number of times a word appears in a text and creating a vector with the number of times as an element. TF-IDF achieves more accurate vectorization by weighting the word frequency in addition to BoW.

[0094] In addition, as the first document learning model, it is possible to use search engine services such as Amazon Kendra (registered trademark), Google Cloud Search (registered trademark), IBM Watson Discovery (registered trademark), Microsoft Azure Cognitive Search (registered trademark), etc., which use natural language processing and advanced machine learning algorithms. These search engine services analyze the query content recognized by the query content recognition unit 17, and search for and extract the first document having the highest similarity from the first specified document set.

[0095] The first answer generating unit 20 acquires the first document searched by the first document searching unit 21 as the context of the prompt, and generates an answer to the inquiry recognized by the inquiry recognition unit 17. A prompt is a command statement for issuing an instruction to the generation AI. A context is a background or external information that is included in a command statement (prompt) for issuing an instruction to the generation AI and that the generation AI should take into consideration in order to improve the quality of the output (answer) of the generation AI. The generation AI can suppress the occurrence of hallucination by considering the context. A method of suppressing the occurrence of hallucination by providing a document related to the user's input data as a context to a prompt, which is a command statement to the generation AI, is called Retrieval Augmented Generation (RAG). Since the maximum number of tokens in the prompt of the large-scale language model that constitutes the generation AI is limited, including the first document that is most similar to the query content recognized by the query content recognition unit 17 extracted from the first specified document set as a context in the prompt is effective in suppressing hallucination of the generation AI.

[0096] When the reply availability determination unit 19 determines that the business support system 10 is unable to reply, the escalation unit 22 connects the user's inquiry related to the inquiry content to a manned reply desk. Escalation refers to the process of requesting a higher-level person to handle a task that the business support system 10 cannot handle. Specifically, when an inquiry cannot be answered by the business support system 10, the escalation department connects the inquiry to a manned response desk, and the manned response desk handles the inquiry. A manned response desk refers to a telephone number and email address that connects to a person who responds to inquiries from users. When a user makes an inquiry by telephone, the escalation unit 22 transfers the call to a manned response desk. When a user makes an inquiry by e-mail, the escalation unit 22 transfers the e-mail to a manned response desk. When an inquiry from a user is made via the website of the business support system 10, the escalation unit 22 transfers the information entered by the user on the website to a manned response desk.

[0097] The second answer generation unit 23 uses the second learning model related to the insurance company that is the inquiry destination accepted by the selection acceptance unit 11 to generate an answer to the inquiry content recognized by the inquiry content recognition unit 17. A second learning model is prepared in advance for each insurance company, which has learned information regarding the product information of the insurance product and the medical underwriting guidelines adopted by the insurance product.

[0098] When the selection receiving unit 11 receives the insurance company to be inquired about selected by the user, the second answer generation unit 23 generates an answer to the inquiry content recognized by the inquiry content recognition unit 17 using a second learning model related to the insurance company to be inquired about received by the selection receiving unit 11.

[0099] The second learning model is a generative AI that is constructed by providing training data to a pre-trained large-scale language model and fine-tuning it so that it can be applied to the tasks of the second learning model. A task refers to the work (processing) required to achieve a specific goal, and the task of the second learning model is to answer inquiries from users about insurance products that differ from insurance company to insurance company and the medical underwriting guidelines adopted by the insurance products. Therefore, the second learning model is generated for each insurance company and fine-tuned based on data such as information posted on the insurance product brochures and homepages, as well as information on the medical underwriting guidelines of the insurance products. In addition, measures to prevent hallucination and measures for instructions and commands (prompts) are necessary for the second learning model as well as for the first learning model. The second learning model is not limited to generative AI, and may be a machine learning model other than generative AI. A machine learning model other than generative AI extracts patterns from given learning data and judges or predicts unknown data based on the extracted patterns, and is different from generative AI, which learns the patterns based on input data and generates new data. The second learning model may be a machine learning model other than a generative AI, and may use natural language processing by the second learning model to classify inquiries from users or extract information from the inquiries from users, and respond to inquiries from users by searching for related documents from accumulated information based on the classified inquiries or the extracted information.

[0100] A second specific document set, which is a collection of second documents, is prepared in advance for each insurance company. The second document is an electronic document for each insurance company regarding product information of the insurance product and the medical underwriting guidelines adopted by the insurance product. The second specific document set is a collection of the above-mentioned second documents for each insurance company, and is prepared for each insurance company and stored in advance in the storage unit 10d.

[0101] The second document search unit 24 searches for and extracts second documents related to the inquiry content recognized by the inquiry content recognition unit 17 from the second specific document set related to the insurance company that is the subject of the inquiry and accepted by the selection acceptance unit 11. A second document learning model that learns the correspondence between the inquiry content recognized by the inquiry content recognition unit 17 and the second document related to the inquiry content is prepared in advance for each insurance company. The second document search unit 24 uses a second document learning model related to the insurance company of the inquiry received by the selection receiving unit 11 to search for and extract second documents related to the inquiry content recognized by the inquiry content recognition unit 17 from a second specific document set related to the insurance company of the inquiry received by the selection receiving unit 11. If the user selects multiple insurance companies as the destinations of the inquiry, the selection receiving unit 11 accepts the multiple destination insurance companies selected by the user. In this case, the second document search unit 24 uses a second document learning model prepared in advance for each of the multiple insurance companies accepted by the selection receiving unit 11 to search for and extract, for each of the multiple insurance companies, a second document related to the inquiry content recognized by the inquiry content recognition unit 17 from a second specific document set prepared in advance for each of the multiple insurance companies accepted by the selection receiving unit 11. The second document learning model is a learning model that searches for and extracts documents related to a given text. Specifically, it is a learning model that has previously learned the correspondence between the query content recognized by query content recognition unit 17 and the second document related to the query content, and searches for and extracts the second document related to the query content recognized by given query content recognition unit 17 from the second specified document set.

[0102] The second document related to the inquiry content recognized by the inquiry content recognition unit 17 may be the second document having the highest similarity to the inquiry content recognized by the inquiry content recognition unit 17. In addition, the second document related to the inquiry content recognized by the inquiry content recognition unit 17 is the second document having the highest similarity to the inquiry content recognized by the inquiry content recognition unit 17, and may be more than one. The second document learning model is prepared for each insurance company, and like the first document learning model, is generated by machine learning using a data set consisting of a large number of combinations of the inquiry content recognized by the inquiry content recognition unit 17 and the second document that is most similar to the inquiry content as learning data.

[0103] In addition, like the first document learning model, the second document learning model may calculate the similarity of each second document included in the second specified document set to the query content recognized by query content recognition unit 17, and search for and extract the second document with the highest similarity from the second specified document set.

[0104] In addition, as the second document learning model, similar to the first document learning model, it is possible to use search engine services such as Amazon Kendra (registered trademark), Google Cloud Search (registered trademark), IBM Watson Discovery (registered trademark), and Microsoft Azure Cognitive Search (registered trademark), which use natural language processing and advanced machine learning algorithms. Amazon Kendra can perform a semantic search on the query content recognized by the query content recognition unit 17, and search and extract highly relevant second documents from the second specific document collection. Semantic search refers to a technology in which a search engine understands the meaning of a query statement and provides search results that are in line with that meaning.

[0105] The second answer generating unit 23 acquires the second document searched by the second document searching unit 24 as the context of the prompt, and generates an answer to the inquiry recognized by the inquiry recognition unit 17. When the user selects multiple insurance companies as destinations for an inquiry, the selection receiving unit 11 receives the multiple destination insurance companies selected by the user, and the second document search unit 24 searches for and extracts a second document for each of the multiple destination insurance companies. Then, the second answer generation unit 23 obtains the second document extracted by the second document search unit 24 for each of the multiple destination insurance companies received by the selection receiving unit 11 as a prompt context, and generates an answer to the inquiry content recognized by the inquiry content recognition unit 17 for each of the multiple insurance companies. Specifically, for example, the user selects Insurance Company A and Insurance Company B as the insurance companies to be inquired about, and the selection receiving unit 11 accepts Insurance Company A and Insurance Company B. Then, the second document search unit 24 extracts the second document A as a second document related to the inquiry content for Insurance Company A, and extracts the second document B as a second document related to the inquiry content for Insurance Company B. The second answer generation unit 23 acquires the second document A as the context of the prompt when generating an answer to the inquiry content for Insurance Company A, and acquires the second document B as the context of the prompt when generating an answer to the inquiry content for Insurance Company B.

[0106] The sentence generating unit 25 generates sentences for informing the inquiry destination selected by the inquiry destination selecting unit 18 of the inquiry content recognized by the inquiry content recognizing unit 17. The sentence generation unit 25 generates sentences using a large-scale language model that has been fine-tuned using text data related to the insurance industry.

[0107] The sentence generation unit 25 connects to the generation AI server 46 and provides instructions and commands (prompts) to the generation AI of the generation AI server 46, with the query content recognized by the query content recognition unit 17 as input data (Input Data), in accordance with the API provided by the generation AI server 46. The sentence generation unit 25 obtains the output (answer) to the instructions and commands (prompts) of the generation AI server 46 in accordance with the API provided by the generation AI server 46.

[0108] Text data relating to the insurance industry refers to text data used in communications between insurance product sales representatives, such as insurance solicitors, and insurance company personnel, as well as the contents of insurance product brochures and homepages. The text data related to the insurance industry includes a data set of interactions between users and insurance company personnel and annotations indicating evaluations of the interactions. This data set of text data and annotations is used as supervised learning data for retraining the large-scale language model that constitutes the generation AI server 46. The evaluation may be a multi-level evaluation such as "good, bad" or a three- or five-level evaluation by the user and the insurance company staff. An insurance company representative refers to a person who belongs to an insurance company and is responsible for supporting insurance product sales representatives, insurance agents, and insurance consultation desks, and may be called a solicitor, for example.

[0109] For example, fine tuning of ChatGPT can be performed by connecting to the ChatGPT API via API communication and uploading learning data in an environment using Google Colaboratory and Python. Note that an API key issued by OpenAI is required to connect to the ChatGPT API.

[0110] The sending unit 26 sends the text generated by the text generating unit 25 to the inquiry destination selected by the inquiry destination selecting unit 18, clearly indicating that the inquiry is from the user identified by the user identifying unit 13. If the inquiry destination selection unit 18 selects a plurality of inquiry destinations, the sending unit 26 sends the text generated by the text generation unit 25 to the plurality of inquiry destinations. The sending unit 26 sends the text to the contact by email, but this is not limited to this. The text may be sent by voice call to the contact's telephone number. The above-mentioned explicit indication by the sending unit 26 means that, when an email is used to send the text generated by the text generation unit 25 to an inquiry destination, information that can identify the user identified by the user identification unit 13, such as the name, ID, or email address of the user, is included in the sender field, title, or body of the email. Furthermore, the above-mentioned explicit indication by the sending unit 26 refers to, when a voice call by telephone is used to send the text generated by the text generating unit 25 to the contact, reading out information capable of identifying the user, such as the name, ID, or email address of the user identified by the user identifying unit 13, at the beginning or end of the voice call. Furthermore, the information capable of identifying the user, such as the name, ID, or email address of the user identified by the user identifying unit 13, may be sent by a message using SMS addressed to the telephone number of the telephone or by email addressed to the email address of the user.

[0111] (Business support methods and business support programs) A business support program according to an embodiment of the present invention will be described together with a business support method with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the business support program according to the present embodiment. The business support method is executed by the CPU 10e of the business support system 10 based on the business support program.

[0112] The business support program enables the CPU 10e of the business support system 10 to realize and perform a selection reception function, a reception function, a user identification function, an inquiry content acquisition function, an inquiry content recognition function, an inquiry destination selection function, an answer availability determination function, a first answer generation function, a first document search function, an escalation function, a second answer generation function, a second document search function, a text generation function, and a sending function. The inquiry content acquisition function includes a voice acquisition function and a voice recognition function.

[0113] These functions are executed in the order shown in the flowchart in Fig. 9, but the order can be changed as appropriate. Note that each function overlaps with the description of the various functional units of the business support system 10 described above, so detailed description thereof will be omitted.

[0114] The selection reception function receives the insurance company to which the user makes an inquiry (S11: selection reception step). The reception function receives an inquiry from a user (S12: reception step). The user identification function identifies the user based on the sender identification information used to send the inquiry from the user that was accepted by the reception function (S13: user identification step).

[0115] The inquiry content acquisition function acquires the contents of the user's inquiry based on the inquiry from the user accepted by the acceptance function (S14: inquiry content acquisition step). The inquiry content recognition function recognizes the inquiry content acquired by the inquiry content acquisition function (S17: inquiry content recognition step). The inquiry destination selection function selects an inquiry destination for making an inquiry about the inquiry content recognized by the inquiry content recognition function from a pre-stored contact list (S18: inquiry destination selection step).

[0116] The answer possibility determination function determines whether or not the inquiry content recognized by the inquiry content recognition function can be answered by the business support system 10 (S19: answer possibility determination step). When the answer feasibility determination function determines that an answer is possible by the business support system 10, the first answer generation function uses the first learning model to generate an answer to the inquiry content recognized by the inquiry content recognition function (S20: first answer generation function step). The first document search function searches for and extracts, from the first specified document collection, a first document related to the inquiry content recognized by the inquiry content recognition function (S21: first document search step). The escalation function connects a user's contact related to the inquiry to a manned response desk when the business support system 10 determines in the answer availability determination function that an answer is not possible (S22: escalation step).

[0117] The second answer generation function generates an answer to the inquiry content recognized by the inquiry content recognition function using a second learning model related to the insurance company of the inquiry destination accepted by the selection reception function (S23: second answer generation step). The second document search function searches for and extracts a second document related to the inquiry content recognized by the inquiry content recognition function from a second specific document set related to the insurance company that is the subject of the inquiry and accepted by the selection reception function (S24: second document search step). The sentence generation function generates sentences for informing the inquiry destination selected by the inquiry destination selection function of the inquiry content recognized by the inquiry content recognition function (S25: sentence generation step). The sending function clearly indicates that the inquiry is from the user identified by the user identification function, and sends the text generated by the text generation function to the inquiry destination selected by the inquiry destination selection function (S26: sending step).

[0118] According to the business support system 10 of the embodiment described above, when making an inquiry to an insurance company, a user can make inquiries to multiple insurance companies by entering one inquiry, and can efficiently obtain information from multiple insurance companies. Furthermore, according to the business support system 10 of the embodiment described above, the generation AI generates the text of the inquiry that the user makes to the insurance company based on the inquiry content created by the user, so that the person to whom the inquiry is made can fully understand the content of the inquiry that the user wants to make, and therefore a better response can be obtained from the insurance company.

[0119] Furthermore, if the business support system 10 according to the embodiment described above determines that it cannot answer the user's inquiry, it transfers the user's inquiry to a manned inquiry desk, so that the user can be sure to receive an answer to his or her inquiry.

[0120] The present invention is not limited to the business support system 10, the business support method, and the business support program according to the above-described embodiments, and can be implemented by various other modified examples or application examples without departing from the gist of the present invention described in the claims. [Explanation of symbols]

[0121] 10 Business Support System 10a Communication Interface 10b ROM 10c RAM 10d Recording section 10e CPU 10f Input / Output Interface 10g Input Device 10h output device 11 Selection reception section 12 Reception 13 User Identification Part 14 Inquiry content acquisition section 15 Voice Acquisition Unit 16 Voice Recognition Unit 17 Query content recognition unit 18 Contact Selection Section 19 Response determination section 20 1st answer generation part 21 First Document Search Unit 22 Escalation Department 23 Second answer generation part 24 Second Document Search Section 25 Sentence generation section 26 Sending Department 25 Telephone answering machine 26 Generative AI Server 27 Public telephone network 29 User telephone terminal 30 User terminals 31 Internet 32 Monitors 33 Display switching button 34 Medical Underwriting Selection Button 35 Brochure selection button 36 Inquiry selection button 37 List of Insurance Companies 38 Checkbox 39 Insurance company selection screen display button 40 Insurance type selection screen display button 41 List of insurance types 42 Checkbox 43 Insurance company selection screen display button 44 Enquiry details entry field 45 Telephone answering machine 46 Generative AI Server 47 Public telephone network 48 User telephone terminal 49 User terminals 50 Web Server 60 Insurance Company Server

Claims

1. A business support system that supports a user's business, A reception unit that receives inquiries from the user; a user identification unit that identifies the user based on sender identification information used to send the inquiry received by the reception unit; an inquiry content acquisition unit that acquires the content of the user's inquiry based on the inquiry accepted by the acceptance unit; an inquiry content recognition unit that recognizes the inquiry content acquired by the inquiry content acquisition unit; a contact selection unit that selects a contact to inquire about the inquiry content recognized by the inquiry content recognition unit from a pre-stored contact list; a sentence generating unit that generates a sentence for notifying the inquiry destination selected by the inquiry destination selecting unit of the inquiry content recognized by the inquiry content recognizing unit; a sending unit that sends the text generated by the text generating unit to the inquiry destination selected by the inquiry destination selecting unit, while clearly indicating that the inquiry is from the user identified by the user identifying unit; A business support system comprising:

2. The inquiry from the user is made by telephone, The user identification unit identifies the user based on a telephone number used to make the inquiry call, The inquiry content acquisition unit, a voice acquisition unit that acquires a voice related to the user's inquiry from a telephone answering device that has responded to the inquiry call accepted by the acceptance unit; a voice recognition unit that recognizes the voice acquired by the voice acquisition unit and acquires the inquiry content as text data; Equipped with 2. The business support system according to claim 1, wherein the inquiry content recognition unit recognizes the inquiry content by analyzing the text data acquired by the voice recognition unit using natural language processing.

3. the inquiry from the user is made by email, The user identification unit identifies the user based on an email address used to send the inquiry, The inquiry content acquisition unit acquires text data included in the email, 2. The business support system according to claim 1, wherein the inquiry content recognition unit recognizes the inquiry content by analyzing the text data acquired by the inquiry content acquisition unit using natural language processing.

4. the business support system includes a website for the user to access; The inquiry from the user is made via the website, The user identification unit identifies the user based on login information used by the user when logging in to the website; the inquiry content acquisition unit acquires text data indicating the inquiry content input by the user to the website; 2. The business support system according to claim 1, wherein the inquiry content recognition unit recognizes the inquiry content by analyzing the text data acquired by the inquiry content acquisition unit using natural language processing.

5. The business support system according to claim 1 , wherein the sentence generation unit generates the sentences using a large-scale language model that is fine-tuned using text data related to the insurance industry.

6. 6. The business support system according to claim 5, wherein the text data includes a data set of an exchange between the user and an insurance company representative and an annotation indicating an evaluation of the exchange.

7. A first learning model that has learned information about the user's work is prepared in advance; an answer possibility determination unit that determines whether the inquiry content recognized by the inquiry content recognition unit can be answered by the business support system; a first answer generation unit that generates an answer to the inquiry content recognized by the inquiry content recognition unit by using the first learning model when the answer feasibility determination unit determines that an answer is possible by the business support system; an escalation unit that connects the inquiry from the user related to the content of the inquiry to a manned response desk when the reply possibility determination unit determines that a reply is impossible by the business support system; The business support system according to claim 1, further comprising:

8. A first specific document set including first documents is prepared in advance, a first document search unit that searches for and extracts the first document related to the query content from the first specific document set; 8. The business support system according to claim 7, wherein the first answer generation unit acquires the first document searched by the first document search unit as a context of a prompt and generates an answer to the inquiry content.

9. a first document learning model that learns a correspondence relationship between the query content and a first document related to the query content is prepared in advance; The business support system according to claim 8, wherein the first document search unit uses the first document learning model to search for and extract first documents related to the query content from the first specific document set.

10. A selection receiving unit is further provided for receiving the insurance company to which the inquiry is to be made selected by the user, 2. The business support system according to claim 1, wherein the inquiry destination selection unit selects the inquiry destination of the insurance company accepted by the selection acceptance unit from the contact list.

11. A second learning model that has learned product information of an insurance product and information regarding medical underwriting guidelines adopted by the insurance product is prepared in advance for each insurance company; A second answer generation unit generates an answer to the inquiry content recognized by the inquiry content recognition unit by using the second learning model related to the inquiry destination insurance company accepted by the selection acceptance unit, The business support system of claim 10, characterized in that when the selection acceptance unit accepts the insurance company to be inquired about selected by the user, the second answer generation unit generates an answer to the inquiry content recognized by the inquiry content recognition unit using the second learning model related to the insurance company to be inquired about accepted by the selection acceptance unit.

12. A second specific document set that is a collection of second documents is prepared for each insurance company in advance, a second document search unit that searches for and extracts the second document related to the inquiry content from the second specific document set related to the inquiry destination insurance company accepted by the selection acceptance unit, The business support system according to claim 11, wherein the second answer generation unit acquires the second document searched by the second document search unit as a context of a prompt and generates an answer to the inquiry content.

13. A second document learning model that learns a correspondence relationship between the inquiry content and a second document related to the inquiry content is prepared in advance for each insurance company; The business support system described in claim 12, characterized in that the second document search unit uses the second document learning model related to the insurance company of the inquiry accepted by the selection accepting unit to search for and extract a second document related to the inquiry content from the second specific document set related to the insurance company of the inquiry accepted by the selection accepting unit.

14. When the user selects a plurality of insurance companies as the contact points, the selection receiving unit receives the insurance companies as the contact points selected by the user, The business support system described in claim 13, characterized in that the second document search unit uses the second document learning model prepared in advance for each of the multiple insurance companies accepted by the selection reception unit to search and extract, for each of the multiple insurance companies, the second documents related to the inquiry content recognized by the inquiry content recognition unit from the second specific document set prepared in advance for each of the multiple insurance companies accepted by the selection reception unit.

15. 2. The business support system according to claim 1, wherein the user is an insurance agent.

16. A business support method for supporting a user's business, comprising: The computer A receiving step of receiving an inquiry from the user; a user identification step of identifying the user based on sender identification information used to send the inquiry received in the receiving step; an inquiry content acquisition step of acquiring the content of the user's inquiry based on the inquiry accepted in the acceptance step; a query content recognition step of recognizing the query content acquired in the query content acquisition step; a contact selection step of selecting a contact to inquire about the inquiry content recognized in the inquiry content recognition step from a contact list stored in advance; a sentence generating step of generating a sentence for notifying the inquiry destination selected in the inquiry destination selecting step of the inquiry content recognized in the inquiry content recognizing step; a sending step of sending the text generated in the text generating step to the inquiry destination selected in the inquiry destination selecting step, while clearly indicating that the inquiry is from the user identified in the user identifying step; A business support method comprising the steps of:

17. A business support program for supporting a user's business, On the computer, A reception function for receiving inquiries from the user; a user identification function that identifies the user based on sender identification information used to send the inquiry received by the reception function; an inquiry content acquisition function that acquires the content of the user's inquiry based on the inquiry accepted by the acceptance function; an inquiry content recognition function that recognizes the inquiry content acquired by the inquiry content acquisition function; a contact selection function for selecting a contact to inquire about the inquiry content recognized by the inquiry content recognition function from a pre-stored contact list; a sentence generation function for generating a sentence for informing the inquiry destination selected by the inquiry destination selection function of the inquiry content recognized by the inquiry content recognition function; a sending function for sending the text generated by the text generating function to the inquiry destination selected by the inquiry destination selecting function, while clearly indicating that the inquiry is from the user specified by the user specifying function; A business support program characterized by exerting the above-mentioned functions.

Images