Business process automation support system and method

Abstract

We provide a system that places humans at the center of business processes, allowing AI and RPA to function as an exoskeleton to support humans. [Solution] A method for supporting the automated processing of business processes that link multiple business systems, wherein an information processing device executes a process that includes the steps of: generating an AI that grasps the processing status on the operator terminal and proposes subsequent processing content; receiving the operator's decision on the proposed processing content from the operator terminal; and controlling the automated processing of the business process by RPA according to the received decision.

Description

Technical Field

[0001] The present invention relates to a business process automation support system and method.

Background Art

[0002] Conventionally, in a business environment where multiple business systems (core systems, SFA, external services, etc.) coexist, in order to realize data linkage between these systems and automation of business processes, one of the following approaches has been adopted.

[0003] The first is system integration and renovation. That is, it is a method of integrating each system into a unified platform and realizing automation through API linkage.

[0004] The second is the adoption of RPA (Robotic Process Automation). RPA is a technology for automating routine tasks by recording and playing back screen operations.

[0005] The third is the adoption of an AI agent. The AI agent performs fully autonomous business automation using generative AI.

[0006] However, in system integration and renovation, a great deal of cost and time are required for modifying existing systems. For this reason, in mission-critical systems, modification itself is often difficult or impossible.

[0007] Also, in the adoption of an AI agent, there is an issue of the reliability of AI judgment. In actual business, it is essential that "the final judgment is made by a human", and full automation is difficult to achieve.

[0008] Also, in the adoption of RPA, there are several essential issues. <Challenge 1> The first challenge in adopting RPA is the human-excluding design philosophy and the resulting time loss. Conventional RPA is based on the premise of "complete automation without human intervention," and lacks a mechanism for human intervention and decision-making during processing. As a result, the following problems occur.

[0009] (1-1) The inability to intervene during the process. Humans have no choice but to wait while RPA is running. This results in inefficient waiting times. Furthermore, it is not possible to stop and correct the process midway. Flexible process control such as "pause," "back," and "forward" is also impossible.

[0010] (1-2) Points that result in time loss and rework in the event of an error In conventional RPA systems, the following problems occur if errors or exceptions occur during processing. Specifically, if an error occurs while the RPA is running (unattended), the process stops and an alert is sent. The person in charge has to wait several minutes to several hours before noticing this alert. Once the person in charge notices the alert, they go to the site to check the situation and investigate the cause of the error. As a result, the process has to be restarted from the beginning, and all the processing time up to that point is wasted.

[0011] This problem has the following serious consequences: Firstly, it results in a loss of time, with a delay of typically several tens of minutes to several hours between the occurrence of an error and the response of the person in charge. Secondly, it involves rework, requiring the process that was partially completed to be restarted from the beginning. Thirdly, it involves wasted processing time, as all processing time up to the point of the error is lost. Fourthly, it causes business delays, resulting in serious delays for tasks with deadlines. Fifthly, the inefficiency of having to restart the entire process from the beginning, especially if an error occurs near the end of the process, is a fatal flaw of conventional RPA.

[0012] (1-3) Lack of flexibility due to human absence Because humans are not involved in the processing, the process stops even in unexpected situations or minor exceptions, resulting in repeated time losses and rework as described above.

[0013] <Challenge 2> The second challenge in adopting RPA lies in its excessive reliance on formal structure. Conventional RPA relies on "formal structure" such as the position coordinates and DOM structure of screen elements, and the column positions, item names, and file formats of data files, which leads to the following problems.

[0014] (2-1) Changes in screen structure When screen layouts change or display environments differ (e.g., different browsers, different resolutions), processing may stop or malfunction even though the work content remains exactly the same.

[0015] (2-2) Diversity of input data structures For example, if a CSV file from system A outputs data in the order "customer name, phone number, address," and a JSON file from system B has the structure {"name":"customer name", "tel":"phone number"}, a human would understand that both are customer information and process them in the same way. However, with conventional RPA, it would be necessary to create completely different logic.

[0016] (2-3) Diversity of output data structure Similarly, if the output destination system X uses field names such as "Name" and "Contact Information," and system Y uses field names such as "Client Name" and "Phone Number," then with conventional RPA, separate mapping logic must be implemented for each output destination.

[0017] (2-4) The problem of logic explosion As a result, if there are N input formats and M output formats, conventional RPA would theoretically require N × M logic patterns, leading to exponentially increasing maintenance costs. Humans can process different formats using the same rules by understanding their "meaning," but conventional RPA lacks this "meaning comprehension" ability.

[0018] A fundamental problem common to all of the above is the neglect of the "human" element. All of the conventional technologies mentioned above are based on the premise of "mechanizing and eliminating humans," and have neglected the elements of "human judgment," "human flexibility," and "human final confirmation," which are indispensable in actual work environments. However, in real-world office work, the flow of System A → Human (confirmation / judgment) → System B is essentially unavoidable, and humans are at the center of the workflow. The applicant is unaware of the existence of prior art documents relating to the above background art. [Overview of the project] [Problems that the invention aims to solve]

[0019] This invention was made in view of these circumstances, and one of its objectives is to provide a system that places humans at the center of business processes and allows AI and RPA to function as an "exoskeleton to support humans." [Means for solving the problem]

[0020] To solve the above problems, the present invention adopts the following configuration. A method for supporting the automated processing of business processes that link multiple business systems, wherein an information processing device executes a process that includes the steps of: a generating AI understanding the processing status on the operator terminal and proposing subsequent processing content; receiving the operator's decision regarding the proposed processing content from the operator terminal; and controlling the automated processing of the business process by RPA according to the received decision. [Effects of the Invention]

[0021] According to the present invention, it is possible to provide a system that places humans at the center of business processes and allows AI and RPA to function as an "exoskeleton that supports humans." Other issues, configurations, and effects not mentioned above will be clarified by the following description of the embodiments. [Brief explanation of the drawing]

[0022] [Figure 1] This is a block diagram showing a configuration example of a business process automation support system according to an embodiment of the present invention. [Figure 2] This is a flowchart showing an example of processing for business process automation according to the embodiment of FIG. 1. **[Modes for Carrying Out the Invention]**

[0023] Hereinafter, modes for carrying out the present invention will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same function or configuration are denoted by the same reference numerals, and redundant descriptions are omitted.

[0024] [Features of the Embodiment] The present embodiment relates to a business process automation support system that enables cooperative operation between human judgment / operation and machine automatic processing without modifying an existing business system. In particular, it relates to a retrofit-type business support technology in which AI (Artificial Intelligence), RPA (Robotic Process Automation), and humans are organically integrated.

[0025] The present embodiment is characterized by the following. (Feature 1) Retrofit implementation without modifying the existing system Without making any modifications to existing business systems A, B, C, etc., these systems are treated integrally to achieve data linkage between systems and automation support for business processes.

[0026] (Feature 2) Automation of the interactive control interface type (Human in the Process) A mechanism is provided that places humans at the center of the business process and functions as "an exoskeleton that supports humans" for AI and RPA. Specifically, the following are realized.

[0027] (Feature 2-1) Human intervention at any point in the process Humans can intervene at any timing of the automatic processing to enable the following operations. • Pause processing • Go back to the previous processing step (back) • Skip to the next processing step (proceed) • Confirmation and correction of processing details • Restart processing

[0028] This resolves the following problems that occurred with conventional RPA. • The inefficiency of "having humans wait until the RPA process is complete." • Time lost when an error occurs (waiting time until a technician arrives) • Rework due to restarting the entire process.

[0029] (Feature 2-2) Human-led decision-making All automated processes are presented to humans as "suggestions," and the final decision on their implementation is always made by a human. The AI ​​does not force a decision, but rather supports human decision-making.

[0030] (Feature 3) Realization of general-purpose processing through understanding the meaning of data and screens This solves the fundamental problem of "dependence on formal structure" inherent in conventional RPA, and achieves the following:

[0031] (Feature 3-1) Understanding the meaning of input data Even if the input data format (CSV, JSON, XML, fixed-length text, etc.), field names ("Customer Name", "name", "Name", "Business Partner Name", etc.) and field order differ, the generating AI (LLM) understands that "this data represents the customer's name," enabling processing with a single logic.

[0032] (Feature 3-2) Understanding the meaning of the output destination and mapping Even if the field names, data formats, and input methods (Web forms, API requests, file output, etc.) required by the destination system are different, the generating AI will determine the semantic correspondence, such as "the customer name should be placed in this field," and map it appropriately.

[0033] (Feature 3-3) Understanding the meaning of the screen structure Even if the screen layout or display environment changes, the generating AI understands that "this screen element is an input field for the customer's name," allowing it to continue processing regardless of coordinates or DOM structure.

[0034] (Feature 3-4) Dramatic reduction in the number of logic blocks As a result of the above, even when there are N input formats and M output formats, it becomes possible to handle everything with "one rule (e.g., transferring customer information)" instead of the N x M logic required in conventional RPA.

[0035] Specifically, with conventional RPA, if there are two types of input formats (CSV, JSON) and two types of output formats (Web form, API), a total of 2 x 2 = 4 logic blocks are required. - System A (CSV format) → System X (Web form): Logic 1 - System A (CSV format) → System Y (API): Logic 2 - System B (JSON format) → System X (Web form): Logic 3 - System B (JSON format) → System Y (API): Logic 4

[0036] In contrast, this embodiment employs a system where the logic does not increase even if the number of formats increases, by using a single rule such as "transcribe customer information (name, phone number, address)" and a function in which the generating AI understands the input format and maps it appropriately to the output destination.

[0037] [Configuration of the Embodiment] Figure 1 is a block diagram showing an example configuration of a business process automation support system according to this embodiment. The business process automation support system is configured by connecting an operator terminal 20, a support server 30, and a group of business systems 50 via a computer network 60.

[0038] The operator terminal 10 is equipped with a client application 20. The client application 20 has the function of displaying the screens of each business system included in the business system group 50 in an integrated manner, the function of visually presenting information provided by the support server 30, and the function of accepting input operations and processing control operations by the operator. The client application 20 is equipped with an integrated display function 21, a suggestion display function 22, and a process operation function 23. In addition, the client application 20 has an integrated authentication function 24 and a business system linkage function 25.

[0039] The integrated display function 21 is a function that enables the integrated display of web screens, form screens, information source screens, or destination screens provided by multiple business systems in a browser within the client application. This function switches between or displays multiple business system screens in parallel on the operator terminal 10, and also displays each screen based on screen information or display control information provided by the support server 30.

[0040] The proposal display function 22 is a function that allows the operator to visually present the proposal content, candidate information, auxiliary information, or related information generated by the support server 30. This function has a configuration that displays the proposal information in a dedicated area on the screen and is equipped with a user interface that accepts input operations such as approval, modification, and selection by the operator.

[0041] The process operation function 23 is a function for receiving operation inputs from an operator regarding business processing. This function has an operation UI that accepts instruction inputs such as pausing processing, going back to the previous step, proceeding to the next step, making corrections, or resuming processing, and has a configuration for sending the input operation instructions to the support server 30.

[0042] The integrated authentication function 24 is a function that allows an operator to input authentication information for logging into this system and to transmit said authentication information to the support server 30. This function is equipped with means for inputting various authentication information (ID, password, token, etc.) necessary for operator authentication, and is configured to control the usage status of the functional parts within the client application 20 based on the authentication result.

[0043] The business system integration function 25 receives login status or session information for multiple business systems established by the support server 30, and enables the integrated display function 21 to display each business system screen in a logged-in state. This function has a configuration in which it maintains session information corresponding to different authentication methods for each of the multiple business systems, and assigns this session information to the integrated display function 21 to display the screen.

[0044] Meanwhile, the support server 30 includes a backend application 40 for integrally acquiring screen information and data provided from multiple business systems, and for performing semantic understanding processing, automatic processing, and authentication linkage. The backend application 40 includes a screen information acquisition function 48, a generation AI processing function 41, an automatic processing function 42, and a processing management function 43. The backend application 40 also has a knowledge management function 44, an integrated authentication management function 45, an authentication linkage function 46, and an authentication information management function 47.

[0045] The screen information acquisition function 48 is a function that enables the non-invasive acquisition of screen information, structural information, item information, or API responses from user interfaces provided by multiple business systems. This function enables the acquisition of structural information (DOM, HTML, labels, field information, etc.) of the web screen displayed by the operator terminal 10, and provides foundational information for realizing semantic understanding processing by the generation AI processing function 41.

[0046] The generation AI processing function 41 is a function that analyzes the screen structure and input data content provided by the screen information acquisition function 48 using generation AI (LLM) and performs natural language processing and semantic comprehension processing to understand their meaning. This function has a configuration for extracting the meaning of screen items, determining semantic correspondences between related items, and generating semantic information necessary for data structure transformation.

[0047] The automated processing function 42 is a function that enables automated processing such as data transfer processing, report output processing, and API call processing required for multiple business systems. This automated processing function 42 includes an automated operation engine (so-called RPA function) that performs operations such as automatic input, automatic transitions, and button clicks on the screens of business systems, and has a configuration for executing necessary automated processing (automatic input, automatic transitions, report output, file generation (CSV, PDF, etc.), API calls, etc.) based on information provided by the screen information acquisition function 48 and the generation AI processing function 41.

[0048] The processing control unit 43 is a function for maintaining the status and managing the progress of various automated processes provided by the backend application 40. This function maintains status information of processing steps, error status, intermediate stop status, or information necessary for resuming processing, and has a configuration that enables integrated cooperation between each functional unit, such as the generation AI processing function 41, automated processing function 42, screen information acquisition function 48, and integrated authentication management function 45. The processing control unit 43 generates execution instructions to the automated processing function 42 based on the operator's judgment results in the approval phase, and the automated processing function 42 executes automated operations on the business system (automatic input, automatic transitions, report output, etc.) according to these instructions, thereby realizing unified progress management of the entire business process.

[0049] The knowledge management function 44 is a function for holding business knowledge, item correspondences, master data, or semantic information referenced by the generation AI processing function 41, related to tasks such as reception work. This function has a configuration for managing item structures and judgment criteria that should be standardized across multiple business systems.

[0050] The integrated authentication management function 45 authenticates operators based on authentication information transmitted from the operator terminal 10 and manages login sessions throughout the entire system. This function has a configuration that maintains the operator's usage rights and valid session information.

[0051] The authentication linkage function 46 is a function for performing authentication linkage with multiple business systems based on the authentication results provided by the integrated authentication management function 45. This function has a configuration that can support various authentication methods (password authentication, token authentication, SSO method, etc.) provided by each business system.

[0052] The authentication information management function 47 is a function that securely stores authentication information required for multiple business systems and provides that authentication information to the authentication linkage function 46 as needed. This function has a configuration that performs encrypted storage, reference management, and update management of credentials.

[0053] [Operation of the Embodiment] Next, the operation of this embodiment will be explained with reference to the flowchart in Figure 2.

[0054] The system first executes the login step S1. When an operator starts processing a task, the client application 20 is launched on the operator terminal 10. The client application 20 displays a login screen using the integrated authentication function 24, and the operator enters their authentication information on the login screen. The entered authentication information is sent to the support server 30, where the integrated authentication management function 45 performs operator authentication processing based on the authentication information.

[0055] Once operator authentication is successfully completed, the support server 30 uses the authentication linkage function 46 and the authentication information management function 47 to automatically execute login processes for multiple business systems available to the operator. This automated login process includes the automatic execution of routine login operations by RPA, token exchange processing with systems that employ API authentication methods, and processing to obtain cookie information or session information.

[0056] Login session information for each business system established by the support server 30 is linked to the business system integration function 25 within the client application 20. This allows the operator to immediately use multiple business systems on the client application 20 without performing any additional operations.

[0057] Next, the system executes the business start step S2. Once integrated authentication is complete and login sessions for multiple business systems are established, the operator begins processing business operations on client application 20. To begin processing, the operator selects the target customer, case, application, transaction, contract, or other processing target (hereinafter referred to as "processing target") on the starting screen. This selection of the processing target is performed using the business system's search function, selection from a list screen, or direct input of a specific ID.

[0058] The support server 30 and RPA may automatically display relevant screens necessary for the process and automatically perform routine operations corresponding to the initial steps of the business flow, based on the processing target selected by the operator. Furthermore, if an external data source related to the processing target exists, the client application 20 can import the external data source into the support server 30 based on instructions from the operator or integration from an external system.

[0059] External data sources may include any structured or semi-structured data, such as CSV, JSON, XML, Excel, message payloads, API responses, log data, and file uploads. Such data sources are treated as input data for the generation AI processing function 41 in the support server 30 and are integrated with screen-derived data in subsequent steps for semantic analysis.

[0060] Next, the system executes the screen information acquisition step S3. Once the target of processing is determined and access to the business systems 50 necessary to proceed with the processing becomes available, the support server 30 and RPA automatically execute standard screen transitions corresponding to the target of processing, and automatically display the necessary starting screen and related detail screens (information source screens). The screen transitions executed here include a series of standard operations in the business systems, such as menu selection, input of search conditions, selection of target records, and transition to detail screens.

[0061] It is also possible for operators to manually operate the business system to perform screen transitions. In this preparation step, the system establishes the screen display state necessary for subsequent steps without distinguishing between screens displayed by automated operation (RPA) and manual operation (operator).

[0062] When the client application 20 detects that the business system screen has been updated, it sends screen update information to the support server 30 indicating that an update has occurred. The screen update information indicates that a screen transition or redraw has occurred, but does not include the screen content itself.

[0063] When the support server 30 receives the screen update information, it uses the screen information acquisition function 48 to acquire the updated screen information. The updated screen information here refers to the entire content of one screen displayed on the operator terminal 10, and if the screen includes multiple web display areas, it includes all the DOM structure, HTML elements, label strings, value display fields, input fields, links, table structures, etc., corresponding to each display area.

[0064] Next, the system performs analysis step S4. In the business start step S2, once the processing target is determined and the necessary external data sources (structured data and semi-structured data) are imported into the support server 30, the updated screen information acquired in the screen information acquisition step S3 and the external data sources acquired in the business start step S2 are treated as input data for the generation AI processing function 41.

[0065] The generation AI processing function 41 utilizes generation AI (LLM) to comprehensively analyze updated screen information (DOM structure, HTML elements, multiple web display areas) and external data sources (arbitrary structured or semi-structured data such as CSV, JSON, XML, Excel, log data, message payload, and API responses). In the analysis process, features such as item names, values, hierarchical structure, list structure, and key-value relationships obtained from both sources are extracted and expressed uniformly as a "semantic structure".

[0066] Regarding screen information, the generation AI processing function 41 analyzes the DOM / HTML structure of the entire screen and identifies multiple display areas. For each identified display area, it extracts structural elements such as headings, labels, value display fields, input fields, tabular data, and buttons, and organizes the information contained in each display area as a semantic structure.

[0067] For multiple web display areas identified from screen information, the generation AI processing function 41 compares them with business knowledge (such as screen types specific to the business system, item structure, processing flow, screen transition rules, and correspondence between label names and item names) stored in the knowledge management function 44, and classifies the display areas into screen types such as starting screen, detail screen (information source screen), input screen (transfer destination screen), and intermediate confirmation screen.

[0068] The generation AI processing function 41 can continue processing even when the screen structure of the business system is changed, by analyzing the meaning of each element on the screen, even in situations where conventional RPA would stop. For example, consider the case where the screen layout and DOM structure are changed as follows.

[0069] In the original screen, the "Customer Name" input field was located at coordinates (100, 200) and its DOM element was defined as input#customer_name. On the other hand, in the modified screen, the same input field may have been changed to a different label, "Customer's Full Name," relocated to coordinates (300, 150), and its DOM element may have been changed to input#client_fullname.

[0070] Conventional RPA systems are highly dependent on changes in coordinate positions and DOM IDs, and often fail to recognize the target element when such changes occur, causing processing to stop. In contrast, the generation AI processing function 41 according to this embodiment semantically understands that both "Customer Name" and "Customer Name" are fields for entering the customer's name, and can treat both as the same input item.

[0071] Therefore, even if the screen layout changes or the DOM element names change, the generation AI processing function 41 can correctly recognize the meaning of the screen element and appropriately identify the input field, thereby enabling automated operations by RPA to continue.

[0072] For external data sources, the generation AI processing function 41 analyzes, for example, CSV column name and value pairs, JSON key and value pairs, XML hierarchical structures, Excel cell structures, etc., and converts each into the same internal "semantic structure" as the screen information. This enables a configuration where, regardless of the format in which the data is input, subsequent processing can apply the same processing based on a single abstract rule.

[0073] In particular, for reference items common to both screen information and external data sources (such as customer information, transaction information, and contract information), the generation AI processing function 41 compares their semantic structure, determines the consistency between the external data source and the screen information, and comprehensively extracts the information necessary for subsequent processing. If there is an area on the screen side that contains reference data, that area is identified as the information source screen, and the data in that area is extracted as detailed information.

[0074] For example, from System A ----- Customer name, phone number, address Taro Tanaka, 03-1234-5678, Tokyo... ----- A CSV file like the one shown may be provided.

[0075] Also, from System B ----- { "name": "Taro Tanaka", "tel": "03-1234-5678", "address": "Tokyo..." } ----- Sometimes, data in JSON format like this is provided.

[0076] Furthermore, in System C, the customer information is displayed on the web screen. ----- Name: Taro Tanaka Contact number: 03-1234-5678 Location: Tokyo... ----- It may be presented in a format like this.

[0077] The generation AI processing function 41 analyzes these disparate representation formats such as CSV, JSON, and screen display, and recognizes that each item represents the same real-world concept. Specifically, it integrates strings such as "customer name," "name," and "name" represented by each input data as information belonging to the same semantic category, and generates semantic structure data as follows. ----- Semantic structure = { "Customer's name": "Taro Tanaka", "Customer's phone number": "03-1234-5678" "Customer's address": "Tokyo..." } -----

[0078] The semantic structure data generated by extracting information source screens and integrating with external data sources is formatted by the processing control function 43 and stored in an internal data format that can be input into the next proposal generation step.

[0079] Next, the system executes the proposal generation step S5. In analysis step S4, once the information source screen is identified and reference data is extracted from the information source screen, the support server 30 proceeds to generate various input assistance information, consistency check results, matching results, transcription candidates, selection candidates, etc. (hereinafter referred to as "suggested content") necessary for subsequent processing, based on the extracted reference data.

[0080] The generation AI processing function 41 semantically compares the reference data with the business knowledge held in the knowledge management function 44 (meaning of input items in each business system, correspondence between reference data and input items, business rules, check items, input constraints, exception handling rules, etc.) and performs context-appropriate inference to estimate the judgment content required in subsequent steps of the business process. As a result, the support server 30 generates a proposal that includes the following:

[0081] • Candidate mappings between reference data and input fields • Results of comparison with existing data • Suggested input values ​​(proposed transcription, estimated value, candidate value) • Results of consistency checks based on business rules • Confirmation items required for subsequent processing • Identifying areas where inconsistencies between business systems are suspected. • Selection options (dropdown menus or master selection options)

[0082] The multiple decision candidates and input candidates generated by the generation AI processing function 41 are integrated by the processing control function 43 and formatted into a data format that can be presented by the suggestion display function 22 of the client application 20. The formatted suggestion content is organized by suggestion point and converted into a structure that is easy for the operator to refer to.

[0083] For example, the following proposal content can be structured: ----- [suggestion] Data obtained from System A: XXX Content to be transferred to System B: YYY [Action Selection] □ Approve and implement the proposal □ Modify the content and execute. □ Skip this process □ Return to previous process -----

[0084] The support server 30 sends the formatted proposal to the operator terminal 10 and displays it on the client application 20. The operator reviews the proposal and selects an action such as approve, modify, hold, or skip for the next process to be performed.

[0085] Next, the system performs the approval step S6. When the proposal generated in proposal generation step S5 is displayed by the proposal display function 22 of the client application 20, the operator reviews the proposal and selects an action for each proposal, such as approve, modify, hold, or skip. These actions are performed via the input device of the operator terminal 10, and the results are sent to the support server 30 as approval result information.

[0086] The generation AI processing function 41 automatically interprets input data and determines the meaning of screen elements, but if there is uncertainty in the AI's judgment or if an important business decision is required, a final confirmation by an operator can be intervened.

[0087] For example, even if the screen layout of a business system is changed and the display position or DOM element name changes, the generation AI processing function 41 may still estimate that both the original input field and the modified input field represent the "customer's name." Furthermore, even if the item name in the input data is changed from "name" to "customer_name," the AI ​​can determine that both represent the same type of information.

[0088] However, if such AI decisions involve a certain degree of uncertainty, or if human verification is required for critical business processes, the system can present the operator with a confirmation screen like the one shown below.

[0089] "Although the screen layout has been changed, we have determined that the business content remains the same. Also, although the names of the input data fields have been changed, we have determined that they all represent the customer's name." "Do you wish to continue processing as is?"

[0090] The operator, □ "Yes (Continue processing)" □ "No (Stop processing and confirm)" By selecting one of the following options, processing can be continued or interrupted. In this embodiment, by combining automated judgment by generating AI with final confirmation by a human, both automation efficiency and work quality can be ensured.

[0091] When the support server 30 receives approval result information, the processing control function 43 interprets the operator's selection results for each proposed item and determines the content to be passed on to the subsequent automated processing step. Specifically, it distinguishes between transcription candidates approved by the operator, input values ​​that have been corrected and confirmed, verification processes that have been instructed to be executed, and items for which automatic input has been selected not to be performed, and determines the execution policy corresponding to each item.

[0092] If the operator chooses to modify the proposed content, the client application 20 displays editable input fields via the proposal display function 22, allowing the operator to add or correct necessary items while referring to the proposed values. The modified values ​​are then finalized through another approval operation and sent to the support server 30 as approval result information. This enables collaborative input decision-making, combining AI-generated suggestions with human judgment and correction.

[0093] On the other hand, if the operator selects to skip or postpone a specific suggestion item, the support server 30 treats that item as excluded from automated processing and logs it as necessary. This allows the operator to flexibly control the scope of automated processing.

[0094] This approval step S6 is the step in which a human operator makes the final decision on the proposals generated in the proposal generation step S5, embodying the "Human in the Process" concept in which automated processing by AI and RPA is performed under the supervision of an operator.

[0095] Next, the system executes the automated processing step S7. In the approval step S6, once the operator has determined an execution policy such as approval, modification, or skipping the proposed content, the support server 30 determines the specific automated processing content to be executed based on the approval result and the semantic structure data obtained in the analysis step S4.

[0096] The generation AI processing function 41 can implement abstract processing rules that are independent of the format of the input data and the output system by performing the semantic understanding processing described above. In other words, in this embodiment, for example, it is not necessary to prepare individual logic such as "processing exclusively for CSV input," "processing exclusively for JSON input," "processing exclusively for screen input," or "processing exclusively for API transmission," which were previously required. By defining a single abstract rule, it is possible to perform unified processing for various input and output formats.

[0097] For example, for a process called "transferring customer information," you only need to define one abstract rule like the following: ----- • Process name: Customer information transfer • Input: Customer's name, phone number, address • Output: Customer information input screen of the target system, or API endpoint The processing based on this rule will be carried out according to the following procedure. 1. Obtain customer information from the input source (CSV, JSON, XML, screen display, API response, etc.). 2. The generation AI processing function 41 extracts the semantic content of the input data and converts it into a unified semantic structure. 3. Based on the extracted semantic structure, appropriately map it to the items in the output destination (Web form, API, file generation, etc.). 4. Ask the operator to confirm the mapping details. 5. After approval, the automatic data transfer process will be executed for the target system. -----

[0098] Thus, the abstract rules according to this embodiment operate with the same processing logic regardless of whether the input source is CSV, JSON, or a screen display, or whether the output destination is a web form, an API, or a file generation. This eliminates the need to create and maintain separate business logic for each input / output format, as was the case conventionally, resulting in a significant improvement in the efficiency and maintainability of business processes.

[0099] The generation AI processing function 41 analyzes both the semantically structured reference data and the structure of the output interface (business system screen, API, external integration module, file generation process, etc.), and automatically derives the correspondence between items in both. This eliminates the need to prepare multiple rules, such as "screen-specific logic if the input destination is a screen, and API-specific logic if the input destination is an API," as was done in the past. It enables automated processing based on a single abstract rule, regardless of whether the input originates from data, a screen, or an API.

[0100] For example, the generating AI analyzes the output system (Web form, API, file, etc.) and performs the following:

[0101] <Example Output Destination 1: Web Form on System X> ┌─────────────┐ │Client Name:[ ]│ │Telephone:[ ] │ │Address:[ ]│ └─────────────┘

[0102] <Output destination example 2: API specifications for system Y> POST / customer { "customer_name": "", "apple_number": "", "location": "" }

[0103] The generation AI associates the semantic information of the input data with the requirements of the output destination.

[0104] Mapping = { "Customer's name" → "Business partner name" field in System X "Customer's name" → The "customer_name" parameter in System Y "Customer's phone number" → "Phone" field in System X "Customer's phone number" → System Y's "phone_number" parameter "Customer's address" → "Address" field in System X "Customer's address" → System Y's "location" parameter }

[0105] The key point here is that this mapping is not hardcoded beforehand, but rather dynamically determined by the generating AI based on the data and screen / API specifications.

[0106] The determined automated processing content is classified into input operations, selection operations, matching operations, screen transition operations, file generation, API transmission, etc., and the processing control function 43 organizes them into the optimal execution order. The processing control function 43 issues specific execution instructions to the RPA or appropriate output module according to the type of element to be executed (screen items, API endpoints, data generation modules, etc.).

[0107] For items requiring on-screen input, the RPA receives execution instructions and performs routine actions such as automatic input into input fields in the business system, checkbox selection, dropdown selection, and button presses. In this process, even if the DOM structure or coordinate arrangement on which the RPA depends is changed, the generation AI processing function 41 performs semantic analysis of the screen elements to correct for fluctuations in the actual element positions and structure, thereby assisting the RPA to continue operations without stopping. This enables automated processing with high resilience to DOM changes and screen rearrangements, which previously often caused RPA to stop.

[0108] For items that can be sent via API, the processing control function 43 generates a parameter set based on semantic structure data for the API output module and sends it to the external system in an appropriate request format (JSON, XML, Form-Data, etc.). Since the API output is independent of the screen structure, automated processing can be performed quickly and stably.

[0109] In addition, if external file generation is required (CSV / Excel / PDF, etc.), the support server 30 can generate a file formatted in a specified format based on its internal semantic structure data and perform integration processing with business systems or external systems.

[0110] If an error screen, warning display, input constraint violation, communication error, etc., occurs during the execution of automated processing, the RPA detects the abnormal state and notifies the support server 30. Based on semantic analysis by the generation AI processing function 41, the support server 30 estimates the cause of the error, generates a re-input suggestion, presents an alternative processing suggestion, and requests re-approval or intervention from the operator as necessary. This achieves the "Human in the Process" principle, which maintains an automated processing flow that "continues without stopping" even during automated execution, while retaining the operator's supervisory authority.

[0111] Finally, once all automated processes are successfully completed, the support server 30 saves the execution details as a log and sends a completion notification to the client application 20. The operator can then check the processing results and continue executing the workflow for the next processing target.

[0112] The business process flow according to this embodiment consists of multiple steps: login step S1, business start step S2, screen information acquisition step S3, analysis step S4, proposal generation step S5, approval step S6, and automated processing step S7. However, these steps are not limited to those that are completed in a single, continuous loop.

[0113] During the execution of business processes, certain steps may be repeated for the following reasons:

[0114] 1. The need for re-analysis due to screen transitions and updates to displayed content. If the screen content changes due to input execution in the automated processing step S7 or manual operation by an operator, the support server 30 receives the screen update information again and re-executes the screen information acquisition step S3 and analysis step S4 to perform subsequent processing based on the latest screen information.

[0115] 2. Appearance of new input candidates and reference data. If new reference data is displayed due to page transitions, tab switching, or disclosure of detail screens, the analysis step S4 and proposal generation step S5 are executed again, and the system returns to the approval step S6 if necessary.

[0116] 3. Correction instructions and rework in approval step S6. If the operator makes modifications to the proposed content, or gives instructions such as "go back" or "hold," the support server 30 returns the processing content to the appropriate step and executes the analysis step S4 or the proposal generation step S5 again.

[0117] 4. Errors or warnings occur during automated processing. If an error screen or warning is displayed during RPA execution, the generation AI processing function 41 analyzes the error and performs appropriate recovery processing, such as returning to the proposal generation step S5 or transitioning to the approval step S6.

[0118] Thus, the business process flow of this embodiment is not a fixed linear flow, but rather has a loop structure in which each step is flexibly repeated according to the situation, demonstrating high adaptability to complex business flows and exception handling in practical applications.

[0119] At this point, the system is capable of executing process control steps asynchronously with each of the steps described so far.

[0120] The system of this embodiment includes a flexible process control mechanism that allows an operator to intervene in the processing flow at any time, even while automated processing is in progress. This process control step can be activated asynchronously during any of the analysis phase S4, proposal generation phase S5, approval phase S6, or automated processing phase S7, and the operator can instruct control operations of the process via the client application 20, regardless of the progress of the processing.

[0121] The client application 20 is provided with a dedicated operation interface that works in conjunction with the processing control function unit 43. This interface visually displays the current processing step, past processing history, pending suggestions, and potential error locations, and presents options such as "pause," "rewind," "forward," "correct," and "resume" as actions that can be performed at that time. The operator can arbitrarily activate these operations using UI components such as buttons, selection menus, and confirmation dialogs.

[0122] When the support server 30 receives instructions from the operator via the operation interface, the processing control function unit 43 updates the internal processing status (step management information) and immediately switches the progress of automatic processing based on the instructions. If a pause operation is given, the current processing step is held in a suspended state and does not proceed to the next step until a resume instruction is given. If a rollback operation is selected, the internal state is rolled back, allowing the operator to return to the step corresponding to the part that needs to be checked or corrected. If a forward operation is selected, processing is skipped to the specified subsequent step and automatic processing resumes from there. In addition, with a correction operation, the operator can directly add to or correct the generated input draft or the value to be transcribed, and processing continues after the corrected value is reflected.

[0123] This process control phase eliminates the need to restart the entire process from the beginning when an error occurs, as is the case with conventional RPA. Instead, it allows for rewinding and correcting only the problematic section before continuing the process. Operators can flexibly participate not merely as monitors of automated processes, but as "collaborative executors" who can intervene and adjust the process flow as needed. This minimizes time loss and rework during errors, while significantly improving the reliability and efficiency of the overall operation.

[0124] [Regarding the use of hardware] The operator terminal 10 is a general-purpose information processing device that runs the client application 20 and performs screen operations of the business system, confirmation of proposal content, and input operations for approval and modification. The operator terminal 10 includes a display device for displaying the user interface, an input device for receiving operation input, a CPU, main memory, and storage necessary for the operation of the client application, and a network interface for communication with the support server 30.

[0125] The client application 20 includes a built-in browser for displaying the web screen of the business system, and this browser communicates directly with the business system's web server via HTTP / HTTPS. Furthermore, the client application 20 has bidirectional communication capabilities via REST API, HTTPS, or WebSocket to communicate with the support server 30, and transmits and receives data to and from the support server 30 via an independent path separate from the business system.

[0126] The support server 30 is an information processing device for centrally executing various server-side processes, such as the generation AI processing function 41, RPA control function 42, processing management function 43, knowledge management function 44, and screen information acquisition unit 48 in this embodiment. To execute these processes, the support server 30 is equipped with a multi-core CPU, large-capacity memory, and persistent storage for storing business knowledge databases and log information.

[0127] The support server 30 receives screen update information and screen information transmitted from the operator terminal 10, formats it into an analyzable format by the screen information acquisition unit 48, and then the generation AI processing function 41 comprehensively analyzes the screen information and external data sources. A GPU may be provided as a computing resource for performing this analysis, if necessary.

[0128] Furthermore, the support server 30 is equipped with a network interface for bidirectional communication with the operator terminal 10, and, when coordinating with external AI services or external systems, it is equipped with network connectivity to communicate with those server groups. From the perspective of business continuity, it is desirable to have a redundant configuration and backup devices.

[0129] One of the distinctive features of this embodiment, as described above, is that it organically integrates the three elements of AI (artificial intelligence), RPA (robotic process automation), and human operators not as opposing forces, but as complementary elements, thereby providing advanced business execution capabilities that cannot be achieved by any single element alone.

[0130] First, AI is primarily responsible for the following advanced decision-making and reasoning processes: interpreting error screens and warning messages in business systems and suggesting appropriate solutions; understanding the meaning and content of screens and data even when their format changes, and determining their identity; extracting important information from input data and mapping it to output destinations; predicting the next action in the business flow and suggesting it to the operator; and detecting data inconsistencies and generating corrective solutions.

[0131] Next, RPA handles mechanical and repetitive processing, automatically executing routine screen transitions, automatically acquiring and transferring data items, automatically logging into multiple business systems, and performing fast and accurate repetitive processing. This automates repetitive tasks that humans would not normally need to perform, significantly improving work efficiency.

[0132] Furthermore, human operators oversee automated processes using AI and RPA, handling critical business decisions and final approvals, responding to exceptional situations, and ensuring quality based on business knowledge. This allows for the appropriate handling of complex exceptions and decisions that would be difficult to address through full automation alone.

[0133] In this way, by having AI, RPA, and humans work collaboratively as a trinity, this embodiment achieves a work execution style that is intermediate yet most practical, and cannot be described by the dichotomy of "fully automated" and "fully manual." In other words, the reasoning power of AI, the mechanical processing capabilities of RPA, and the judgment of humans work together complementaryly to simultaneously improve the efficiency, flexibility, and quality of work processing.

[0134] [Effects of the Embodiment] This embodiment provides the following wide range of remarkable effects. (Effect 1) Can be implemented in stages without any modification to the existing system. This embodiment employs a configuration that can be applied to existing business systems without modification, enabling significant reductions in implementation costs, shorter implementation periods, and low-risk implementation without impacting existing operations. In particular, it allows for safe and phased implementation even for mission-critical core systems.

[0135] (Effect 2) Extends human capabilities and significantly improves work efficiency. This embodiment does not eliminate humans, but rather enhances human capabilities by placing their judgment at the core while delegating routine tasks to AI and RPA. This simultaneously achieves resource optimization through the automation of simple tasks, allows humans to focus on tasks that require judgment, enables high-speed processing of parts that can be automated, and provides a sense of security for humans (retaining final decision-making authority), resulting in an improvement in operational efficiency of approximately 30-50%.

[0136] (Effect 3) Minimizing time loss and rework In conventional RPA, if an error occurred, the entire process had to be restarted from the beginning, which presented a serious problem: if an error occurred towards the end of a long-running process, the entire processing time would be wasted. In this embodiment, the AI ​​immediately interprets the error content and corrects only the necessary parts, allowing the process to continue. This eliminates time loss and minimizes rework. The effect is particularly significant in long-running processes, dramatically improving business continuity compared to conventional methods.

[0137] (Effect 4) Achieves both high flexibility and reliability This embodiment allows for flexible intervention and modification even during automated processing. Even if screen or data specifications change, the AI ​​can determine whether to continue the task based on its understanding of the meaning, thus providing high flexibility. Simultaneously, because it enables decisions based on the essence of the task, it avoids the black-boxing often seen in conventional RPA, ensuring transparency and reliability of the process.

[0138] (Effect 5) Covers areas that were previously considered "difficult to automate". In this embodiment, since automation is performed while human judgment is involved, the following tasks, which were difficult to automate with conventional technology, can now be supported by automation. • Tasks that involve human judgment • Complex business workflows spanning multiple systems • Tasks where exception handling occurs frequently • Tasks where screen specifications are frequently changed. This significantly expands the areas that can be automated.

[0139] (Effect 6) Technical originality This embodiment has distinct originality not found in the prior art in the following respects. • A new design philosophy: "Human-centered interactive control interface type automation" • A retrofitting method called "exosuit-type reinforcement," which reinforces existing systems from the outside. • A "trinity-integrated architecture" where AI, RPA, and humans work closely together. • An AI semantic understanding mechanism that determines essential identity even when screen representation or data format changes. • A novel approach to general-purpose processing based on semantic understanding As a result, this embodiment forms an automation technology system that is fundamentally different from the prior art.

[0140] (Effect 7) Dramatic reduction in the number of logic blocks According to this embodiment, the enormous number of logics required in conventional RPA can be significantly reduced. In conventional RPA, if there are N input formats and M output formats, it is necessary to create N × M individual logics for each combination of formats, and the number of logics increases exponentially as the number of input and output formats increases.

[0141] In contrast, this embodiment uses semantic understanding by generative AI to execute processing using abstract rules that are independent of the format of the input data and output destination. Therefore, it is not necessary to prepare individual logic for each format that was previously required, and the necessary logic is consolidated into a single rule.

[0142] As an example of quantitative effects, considering a scenario with 5 input formats and 5 output formats, conventional RPA would require 25 logic rules, whereas this embodiment can handle it with just one rule, resulting in a logic reduction of 96%.

[0143] Furthermore, it offers significant advantages in terms of maintenance costs. When a new input format is added, conventional RPA requires the implementation of new logic for each output format. However, in this embodiment, existing abstract rules can be applied as is, eliminating the need for additional implementation. Even if an existing input format changes, conventional RPA required the modification of all related individual logic. In contrast, in this embodiment, the generating AI automatically absorbs the format change, eliminating the need for modification or requiring only minimal modification.

[0144] (Effect 8) High adaptability to denormalized data In actual business environments, data is managed in different formats by each department, and the pre-processing required to standardize the format was a significant burden. In this embodiment, the AI ​​can understand the meaning of various data and process them uniformly, resulting in the following benefits. • Integration of data managed in a unique format by each department. • Automatically handles data format differences with external systems. • It is possible to mechanically integrate data that humans can determine to be "the same". This eliminates the traditional assumption that "automation is not possible unless data is standardized in advance," significantly improving its applicability to practical applications.

[0145] Next, we will compare the effects of this embodiment with those of the prior art. This embodiment has fundamentally different characteristics from conventional RPA technology and conventional AI agents in all aspects, including its philosophy, processing structure, error tolerance, operational adaptability, and data understanding capabilities. Below, we will explain how this embodiment differs from the prior art from each key perspective.

[0146] (1) Differences in design philosophy Conventional RPA and AI agents were designed to eliminate humans and aim for "complete automation," limiting human involvement to specific roles such as exception handling and monitoring. In contrast, this embodiment adopts a completely new design philosophy: a "human-centered, interactive control interface." Its key feature is that humans are at the center of business processing, and AI / RPA functions like an exoskeleton (Exosuit) that complements and extends their capabilities.

[0147] (2) The role of humans Traditionally, the role of humans in RPA has been limited to "passive roles" such as handling exceptions and recovering from failures. Furthermore, conventional AI agents only monitored the overall process and did not deeply participate in actual business decisions. In this embodiment, humans actively participate in the work as "collaborative executors," supervising the AI's decisions and making the final decisions. This integrates human judgment with the automated processing capabilities of AI and RPA.

[0148] (3) Flexibility of process control Conventional RPA systems are unable to deviate from defined flows, making flexible process control, such as stopping, reversing, or restarting processes, impossible. Conventional AI agents also exhibit similar limitations, resulting in low flexibility in process control. In this embodiment, flexible control, such as pausing, reversing, and continuing processes, is possible, and the business flow is dynamically reconfigured in response to human judgment and changing circumstances.

[0149] (4) How to handle errors Conventional RPA required a complete restart of the entire process if screen changes or input value exceptions occurred. Even with conventional AI agents, error behavior could not be clearly controlled, and the continuity of processing was not guaranteed. In contrast, this embodiment allows the semantic understanding mechanism of the generated AI to partially correct only the erroneous parts and continue processing. This prevents critical time losses during long-running processes.

[0150] (5) Impact on time loss Traditional RPA systems sometimes lost processing time by tens of minutes to several hours after an error occurred because they had to wait for a human to respond. Similarly, traditional AI agents also suffered time loss due to process interruptions. In this embodiment, however, a human can intervene immediately and partially correct the error, allowing the process to continue without interruption, thus minimizing time loss.

[0151] (6) Criteria for determining continuity Conventional RPA systems were extremely vulnerable to screen changes because they determined whether processing could continue based on "formal elements" such as screen coordinates and DOM structure. Even conventional AI agents lacked established criteria for this determination. In contrast, this embodiment determines continuity based on the "essential identity of the business content." In other words, even if the screen's presentation changes, processing can continue as long as the business meaning remains the same.

[0152] (7) Mechanisms for understanding data Traditional RPA systems were heavily reliant on formal structures such as column positions and item names. Even conventional AI agents lacked universal semantic understanding capabilities. This embodiment utilizes higher-order semantic understanding through generative AI, absorbing differences in data formats and enabling unified handling of various formats such as CSV, JSON, and screen displays.

[0153] (8) Differences in the number of logics Conventional RPA required N × M individual logic blocks when there were N input formats and M output formats. Conventional AI agents have unclear logic structures. In this embodiment, all patterns can be handled with a single abstract rule based on semantic understanding, reducing the number of logic blocks to 1.

[0154] (9) Relationship with existing systems Conventional RPA and AI agents do not require modification of existing systems, but this comes at the cost of limited automation scope. This embodiment also eliminates the need for modification of existing systems while achieving a broader scope of automation by utilizing semantic understanding.

[0155] (10) Implementation Approach Traditional RPA and AI agents required "adapting tasks to the tool," presupposing standardization and formalization. This embodiment, however, adopts the opposite approach: "adapting the tool to the tasks," eliminating the need to modify existing workflows.

[0156] (11) Final decision-making power In conventional systems, the final decision-making power rested with the system, making human intervention difficult. In this embodiment, the final decision-making power always remains with humans, and AI and RPA are solely in a supporting role. This contributes to the construction of a highly reliable and transparent automation foundation.

[0157] The technical essence of this embodiment lies in the fact that it is based on a new understanding that applies "information processing based on semantic understanding," which humans normally perform, to automation, rather than the "formal processing" on which conventional technologies relied.

[0158] When humans recognize data, they understand and judge it based on the "meaningful content" it represents, rather than on formal information such as its position on the screen, column order, or item name. For example, various expressions such as "customer name," "name," "name," and "customer name" are instantly understood by humans to have the same meaning, "customer's name," even though their formats differ. Furthermore, different data representation formats such as CSV, JSON, and web screen displays can all be recognized uniformly as "data." This ability to understand meaning allows humans to perform the same processing even if the position or order changes, as long as the meaning remains the same.

[0159] However, conventional RPA lacked this semantic understanding capability and relied on formal elements such as screen coordinates, DOM structure, column positions, and string matching of item names to operate. Therefore, even a slight change in the input data format would render the existing logic inoperable, necessitating the implementation of entirely new logic. As a result, maintenance costs increased exponentially, becoming a barrier to field adoption. Furthermore, if an error occurred during processing, the process would stop, requiring the arrival of a human operator to resolve the issue, as human intervention was not integrated into the workflow. The resulting time loss and rework due to the complete reprocessing were significant.

[0160] On the other hand, generative AI possesses high semantic comprehension capabilities based on natural language processing. It can grasp the identity of different expressions such as "customer name," "name," and "full name," and interpret data based on its meaning in the real world. This embodiment combines the capabilities of this generative AI with the automated processing mechanism of RPA, and further adopts a new structure that places humans at the center of the business process flow, thereby realizing a collaborative automation system that "processes automatically while understanding meaning like a human, and allows for immediate human intervention to continue when errors occur."

[0161] In other words, the core of this embodiment lies in the following two points. (1) General processing based on semantic understanding Because the generation AI interprets the meaning of the data, it can process unnormalized data, different screen display formats, and file formats (both input and output) uniformly with the same rules and minimal logic.

[0162] (2) Cooperative operation via an interactive control interface Because the system employs a structure where humans are always at the heart of the workflow, and AI and RPA function as its exoskeleton, it is possible to continue processing through immediate human intervention in the event of uncertainty or exceptions during processing.

[0163] These two components fundamentally solve the problems of conventional mechanical automation, such as formal dependency, vulnerability, maintenance burden, and time loss and rework in the event of errors. This provides a novel and original technological system with a high degree of flexibility and practical adaptability that could not be achieved with conventional technologies.

[0164] The essence of this embodiment is not limited to a specific algorithm, program structure, or technical implementation method, but rather lies in the technical philosophy itself, based on the fundamental ideas described below.

[0165] Firstly, this embodiment introduces a new automation concept of "cooperative operation between humans, AI, and RPA." While conventional automation technologies aimed for complete system automation by excluding humans, this embodiment places humans at the center, retaining human judgment while employing a structure in which AI and RPA complement and enhance the work in an exoskeleton-like manner.

[0166] Secondly, this embodiment is based on the idea of ​​"understanding and processing meaning rather than form." While conventional technologies rely on formal information such as screen coordinates, DOM, column positions, and item names for processing, this embodiment uses the semantic understanding capabilities of the generation AI to uniformly interpret the semantic content contained in different data formats and screen representations, thereby absorbing differences in format.

[0167] Thirdly, this embodiment presents a new paradigm that shifts from "automation that eliminates humans" to "automation that supports humans." This creates an environment where business processes can continue even in situations requiring sophisticated exception handling and decision-making, while simultaneously enhancing human capabilities.

[0168] Fourthly, this embodiment is based on the concrete and practical idea of ​​"minimizing time loss and rework in the event of errors." By using a conversational control structure that enables semantic understanding by generational AI and immediate human intervention, it is possible to avoid stopping or restarting the entire process, and to continue the process with partial corrections.

[0169] All of the above ideas are original and do not exist in prior art, and are important elements that underpin the technical value and patentability of this embodiment.

[0170] It should be noted that this embodiment is not limited to the embodiments described above, and various other applications and modifications can be taken as long as they do not depart from the gist of this embodiment as described in the claims. For example, the above-described embodiment is a detailed and specific explanation of the system configuration in order to clearly illustrate this embodiment, and is not necessarily limited to having all the configurations described. Furthermore, it is possible to add, delete, or replace some of the configurations in this embodiment with other configurations. Furthermore, the control lines and information lines shown are those deemed necessary for explanatory purposes, and not all control lines and information lines are necessarily shown in the actual product. In reality, it is safe to assume that almost all components are interconnected. [Explanation of symbols]

[0171] 10…Operator terminal, 20…Client application, 21…Integrated display function, 22…Suggestion display function, 23…Process operation function, 24…Integrated authentication function, 25…Business system linkage function, 30…Support server, 40…Backend application, 41…Generative AI processing function, 42…Automated processing function, 43…Process management function, 44…Knowledge management function, 45…Integrated authentication management function, 46…Authentication linkage function, 47…Authentication information management function, 48…Screen information acquisition function, 50…Business system group, 60…Computer network

Claims

1. A method for supporting the automated processing of business processes that link multiple business systems, The generating AI grasps the processing status on the operator terminal and proposes the subsequent processing content, The steps include receiving the operator's decision regarding the proposed processing content from the operator terminal, The process includes the step of controlling the automated processing of the business process by RPA in response to the received decision, A method for supporting the automation of business processes executed by information processing equipment.

2. The business process automation support method according to claim 1, wherein the screens of the multiple business systems are displayed integrally within a browser built into an application running on the operator terminal, and the generating AI grasps the processing status on the operator terminal based on its understanding of the meaning of the display structure within the browser.

3. The business process automation support method according to claim 1, further comprising the step of receiving a decision from the operator terminal during the control of the automated process by the RPA, and performing control including stopping, reversing, skipping, confirming, correcting, or resuming the automated process in accordance with the decision.

4. The method for supporting business process automation according to claim 1, wherein the automated processing by RPA is controlled based on the generation AI's understanding of the meaning of the data from the source business system and / or the meaning of the data output destination.

5. A system that supports the automated processing of business processes that link multiple business systems, The generating AI grasps the processing status on the operator terminal and proposes the subsequent processing content, The steps include receiving the operator's decision regarding the proposed processing content from the operator terminal, The information processing device includes an information processing device that performs processing including the step of controlling the automated processing of the business process by RPA in response to the received decision, Business process automation support system.

Images