Insurance sales intelligent partner training system and method based on multiple agents

Abstract

The invention discloses an insurance sales intelligent partner training system and method based on multiple agents, and relates to the technical field of large language models, the system comprises five kinds of agents, namely a client agent, a comment agent, a lesson making agent, a review agent and a knowledge extraction agent, the client agent loads client information and a script, interacts with an agent and asks when ambiguity is replied; the commenting agent comments and replies according to the target table, and performs abnormal verbal skill and anti-cheating detection; the lesson making agent generates a training material by combining the knowledge points and the customer portraits; the review agent analyzes the grouped partner training result, evaluates team ability and deduces learning materials; the knowledge extraction agent extracts knowledge from the multi-source data, and generates reuse materials or optimization suggestions. According to the system, through simulation interaction and multi-dimensional feedback, an agent is helped to improve skills directionally, the training process can be optimized, the cost is finally reduced, communication specifications are guaranteed, an efficient training system is constructed for insurance enterprises, and the service and conversion efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of large language model technology, and in particular to an intelligent training system and method for insurance sales based on multi-agent systems. Background Technology

[0002] In the development of the insurance industry, agents' sales skills and customer communication abilities are key factors affecting sales performance, thus related training is highly valued. Currently, insurance agent training largely relies on traditional training systems. The training content of these systems is mainly generated manually, covering basic sales knowledge, product information, and limited customer communication cases. The examination process is also mostly graded manually, judging the agent's mastery of the training content based on preset fixed standards.

[0003] However, existing technologies have significant drawbacks: On the one hand, manually compiled training content is difficult to cover all possible customer needs and communication scenarios in insurance sales, and cannot provide personalized training courses based on the actual situation of different agents' knowledge reserves and skill gaps, resulting in rigid and inflexible training programs that fail to accurately meet agents' improvement needs; on the other hand, under the manual scoring model, scoring standards are easily affected by subjective factors and lack a unified and standardized evaluation system, which not only easily leads to misjudgments but also causes inconsistent training effectiveness evaluation results among different agents, failing to effectively guarantee training quality. Ultimately, it is difficult to effectively help agents improve their sales skills and customer communication abilities, and cannot meet the industry's actual need to improve sales performance through high-quality training. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a multi-agent-based intelligent training system and method for insurance sales, comprising the following:

[0005] Firstly, this application provides a multi-agent-based intelligent training system for insurance sales, the system comprising:

[0006] Customer intelligent agent, review intelligent agent, lesson creation intelligent agent, review intelligent agent, and knowledge extraction intelligent agent;

[0007] The client intelligent agent is used to load client role information and training script, interact and practice with the agent, and autonomously ask questions or challenges when the agent's response is ambiguous.

[0008] The review agent is used to load the target table of the corresponding script, and to provide single-sentence reviews, stage reviews, or overall reviews of the agent's responses, while also performing abnormal wording detection and anti-cheating detection.

[0009] The course-making intelligent agent is used to generate exam questions, training materials, exam points, and scripts by combining knowledge points and customer profiles;

[0010] The review agent is used to analyze the overall evaluation results of the agent group training, assess the overall ability of the team, and recommend personalized learning materials;

[0011] The knowledge extraction agent is used to extract structured knowledge from multi-source data, including product materials, training materials, and online logs, to generate reusable training materials or optimization suggestions.

[0012] Optionally, the customer role information loaded by the customer intelligent agent includes:

[0013] Name, title, age, gender, occupation, location, whether it is a new customer, customer type, customer group category, and potential information, including education level, annual income, hobbies, investment preferences, asset size, family structure, and personality traits.

[0014] Optionally, the training script loaded by the customer intelligent agent contains multiple stages, each stage defining reference scripts for sales personnel and customers, as well as goals and corresponding script segments related to the scripts.

[0015] Optionally, the single-sentence comments of the review agent are used to generate results that include test point scores, improvement suggestions, and recommended phrases;

[0016] The phase review is used to provide feedback on the completed phases, including phase scores, strengths and weaknesses analysis, and improvement suggestions, when the agent completes the current phase goal and moves to the next phase.

[0017] The overall review is used to provide a total score, assessment of strengths and weaknesses, and suggestions for improvement after the agent has completed all stages.

[0018] Optionally, the abnormal speech detection dimensions of the review agent include:

[0019] The system checks whether the content provided by the current agent violates authenticity, ethical standards, company policies, or is irrelevant to the objective.

[0020] Optionally, the anti-fraud detection is used to detect the similarity between the agent's script and the standard script. If the similarity exceeds a preset similarity threshold, the corresponding target score is canceled and a result containing evaluation and optimization suggestions is generated.

[0021] Optionally, the knowledge extraction agent's analysis of product materials includes extracting information on the company's products' functions, parameters, and usage scenarios;

[0022] Training material generation includes summarizing frequently asked questions and best practices based on historical training data;

[0023] Online log analysis includes identifying user behavior patterns or common errors and feeding them back to product iterations.

[0024] Secondly, this application provides a multi-agent-based intelligent training method for insurance sales, the method comprising:

[0025] Based on the course-making intelligent agent, test questions, training materials, test points and scripts are generated by combining knowledge points and customer profiles;

[0026] The client intelligent agent loads client role information and training scripts, and conducts interactive drills with the agent. When the agent's response is ambiguous, the agent will ask questions or raise doubts.

[0027] Based on the question-and-answer results between the customer agent and the agent, the review agent loads the target table of the corresponding script and performs single-sentence review, stage review, and overall review on the agent's response, while also performing abnormal wording detection and anti-fraud detection.

[0028] The overall evaluation results of the intelligent agent's group training sessions were reviewed and analyzed to assess the team's overall capabilities and recommend personalized learning materials.

[0029] The knowledge extraction agent extracts structured knowledge from multi-source data such as product materials, training materials, and online logs to generate reusable training materials or optimization suggestions.

[0030] Optionally, the customer role information loaded by the customer agent includes:

[0031] Name, title, age, gender, occupation, location, whether it is a new customer, customer type, customer group category, and potential information, including education level, annual income, hobbies, investment preferences, asset size, family structure, and personality traits.

[0032] Optionally, the training script loaded by the customer intelligent agent contains multiple stages, each stage defining reference scripts for sales personnel and customers, as well as goals and corresponding script segments related to the scripts.

[0033] Optionally, the single-sentence comments of the review agent are used to generate results that include test point scores, improvement suggestions, and recommended phrases;

[0034] Phase review is used to provide feedback on completed phases, including phase scores, strengths and weaknesses analysis, and improvement suggestions, when an agent completes the current phase and moves to the next phase.

[0035] The overall review is used to provide a total score, assessment of strengths and weaknesses, and suggestions for improvement after the agent has completed all stages.

[0036] Optionally, the abnormal speech detection dimensions of the review agent include:

[0037] The system checks whether the content provided by the current agent violates authenticity, ethical standards, company policies, or is irrelevant to the objective.

[0038] Optionally, the anti-fraud detection is used to detect the similarity between the agent's script and the standard script. If the similarity exceeds a preset similarity threshold, the corresponding target score is canceled and a result containing evaluation and optimization suggestions is generated.

[0039] Optionally, the knowledge extraction agent's analysis of product materials includes extracting information on the company's products' functions, parameters, and usage scenarios;

[0040] Training material generation includes summarizing frequently asked questions and best practices based on historical training data;

[0041] Online log analysis includes identifying user behavior patterns or common errors and feeding them back to product iterations.

[0042] Thirdly, this application provides a device comprising a memory and a processor, the memory for storing instructions or code, and the processor for executing the instructions or code to cause the device to perform the multi-agent-based intelligent coaching method for insurance sales described in any of the implementations of the second aspect.

[0043] Fourthly, this application provides a computer-readable storage medium storing code, wherein when the code is executed, a device running the code implements the multi-agent-based intelligent coaching method for insurance sales described in any of the implementations of the second aspect above.

[0044] This application provides a multi-agent-based intelligent training system for insurance sales. It includes: a customer agent, a review agent, a course-creating agent, a review agent, and a knowledge extraction agent. The customer agent loads customer role information and training scripts, interacts with the agent, and autonomously asks questions or challenges when the agent's responses are ambiguous. The review agent loads the target table corresponding to the script, provides single-sentence, stage, or overall reviews of the agent's responses, and performs abnormal sales pitch detection and anti-cheating detection. The course-creating agent combines knowledge points and customer profiles to generate exam questions, training materials, exam points, and scripts.

[0045] The review agent is used to analyze the overall evaluation results of the agent group training, assess the overall ability of the team, and recommend personalized learning materials; the knowledge extraction agent is used to extract structured knowledge from multi-source data, including product materials, training materials, and online logs, to generate reusable training materials or optimization suggestions.

[0046] In this way, by simulating real customer interaction scenarios through a customer intelligence agent, and combining precise role information and scripted processes to achieve immersive training, while providing real-time feedback on agent performance from multiple dimensions through a feedback intelligence agent, and supplemented by course creation, review, and knowledge extraction intelligence agents, agents can repeatedly train their sales skills in a highly simulated environment, targeting weaknesses for improvement. Furthermore, the system can continuously optimize training content and processes based on actual business data. This achieves the goal of improving the relevance and efficiency of insurance agent sales capabilities, reducing training costs, and ensuring standardized and professional sales communication. Thus, it can help insurance companies build a standardized and efficient agent training system, helping them quickly cultivate sales talent that meets business needs, thereby improving customer service quality and business conversion efficiency. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A schematic diagram of the structure of an intelligent insurance sales coaching system based on multi-agent technology provided in this application embodiment;

[0049] Figure 2 A schematic diagram of another intelligent insurance sales coaching system based on multi-agent technology provided in this application embodiment;

[0050] Figure 3 A flowchart of a multi-agent-based intelligent coaching method for insurance sales provided in this application embodiment. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0052] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0053] Figure 1 This is a schematic diagram of a multi-agent-based intelligent insurance sales coaching system provided in an embodiment of this application. (Combined with...) Figure 1 As shown in the embodiment of this application, the insurance sales intelligent coaching system 100 based on multiple agents includes: a customer agent 110, a review agent 120, a course creation agent 130, a review agent 140, and a knowledge extraction agent 150. An agent is an intelligent entity with autonomous perception, decision-making, and action capabilities, capable of autonomously interacting, completing tasks, or achieving goals in a specific environment. In this application, the agents are intelligent modules built based on a large language model. Each agent undertakes a specific function, collaboratively realizing the full-process management of insurance sales intelligent coaching. The functions of each agent are briefly described below:

[0054] The customer intelligence agent is used to load customer role information and training scripts, and interact with agents to practice. It autonomously asks questions or challenges when the agent's responses are ambiguous. The review intelligence agent is used to load the target table corresponding to the script, and provide single-sentence, stage-based, or overall reviews of the agent's responses, while also detecting abnormal language and preventing cheating. The course creation intelligence agent is used to generate exam questions, training materials, key points, and scripts by combining knowledge points and customer profiles. The review intelligence agent is used to analyze the overall review results of the agent group training sessions, assess the team's overall ability, and recommend personalized learning materials. The knowledge extraction intelligence agent is used to extract structured knowledge from multi-source data, including product materials, training materials, and online logs, to generate reusable training materials or optimization suggestions.

[0055] The above embodiments provide a brief description of the multi-agent-based intelligent insurance sales coaching system disclosed in this application. To enable those skilled in the art to clearly grasp the specific structure of this system, the following description, in conjunction with... Figure 2 This paper elaborates on the intelligent training system for insurance sales based on multi-agent technology. Figure 2 This is a schematic diagram of another intelligent insurance sales coaching system based on multi-agent technology, provided as an embodiment of this application.

[0056] The client intelligent agent module is the core execution unit for coaching interaction. It contains a role information loading submodule, a script parsing submodule, and an interaction response submodule, with the following specific functions:

[0057] The Role Information Loading Submodule loads preset customer role information, including name, title, age, gender, occupation, location, whether it's a new customer, customer type, customer group category, and potential information. This potential information includes education level, annual income, hobbies, investment preferences, asset size, family structure, and personality traits. For example, a customer role might have the name "Su Wei," be addressed as "General Manager Su," "General Manager Su Wei," or "Mr. Su," be 35 years old, have a high-income professional status, be located in Shenzhen, and have potential information such as a doctoral degree, an annual income of 220,000 yuan, and an aggressive investment preference. This submodule supports flexible configuration and updates of role information, allowing agents to add or modify customer attribute fields through the backend management interface. This multi-dimensional, customizable role information highly replicates the characteristics of real insurance customers, enabling agents to tailor their communication styles to different clients during training sessions (e.g., focusing on professional policy interpretation for high-income professionals and emphasizing cost-effectiveness for ordinary white-collar workers). This avoids the traditional training model where information cannot be changed, significantly improving the adaptability and practicality of communication skills.

[0058] The script analysis submodule loads and parses the practice script. The loaded script contains multiple stages (such as scheduling appointments, small talk and topic introduction, real-time discovery and consensus building, product solution explanation, and objection handling). Each stage defines reference scripts for sales personnel (agents) and customers, along with related objectives and corresponding script snippets. For example, in the scheduling stage, the script defines reference scripts for sales personnel such as "Hello, Mr. Su! I'm Zhang from xxx Life Insurance, which Mr. Gao mentioned to you. It's a pleasure to contact you," and reference scripts for customers such as "Hello, Zhang, what can I do for you?" It also links to objectives such as "greeting" and "mentioning the referrer," along with corresponding script snippets. This submodule automatically identifies the stage divisions and objective relationships within the script, generating a structured practice flowchart. In this way, the structured script provides a standardized process framework for training, guiding agents to gradually master the logic of the entire insurance sales process and avoid missing key links in communication (such as not agreeing on a meeting time when scheduling an appointment); while the binding of the script with the goal allows agents to clearly know the training focus of each sentence, improving the direction and efficiency of the training.

[0059] The evaluation module is the core unit for assessing training effectiveness. It includes a target table loading submodule, a hierarchical evaluation submodule, and a risk detection submodule. These submodules work together to achieve standardized, multi-dimensional evaluation of the agent's scripts. Specific functions are as follows:

[0060] The target table loading submodule is used to load the target table for the corresponding coaching script, providing a unified evaluation basis for tiered reviews. The target table contains the evaluation points for each stage (such as "using honorifics when greeting customers" and "introducing core information about tax incentive policies"), the weight of each point (such as a maximum of 1 point for a single point), the judgment rules (such as "must mention both the referrer and one's own identity"), and sample scripts. For example, in "Stage_0_Target_0" (greeting point), the judgment criteria of "using honorifics when greeting customers and introducing oneself" are clearly defined, as well as the rule that no points are awarded for this point if honorifics are not used, ensuring the objectivity and consistency of subsequent review results.

[0061] The tiered review submodule loads the evaluation criteria provided by the target table and generates review results in three dimensions: single sentence, stage, and overall. This enables a full-process evaluation from details to the whole. Specifically, single sentence reviews include generating results for each agent's response, including "key point score," "improvement suggestions," and "recommended scripts." The key point score is based on the completion status of the corresponding key points in the target table (e.g., 1 point for completing the "use honorifics" key point, 0 points for not completing it); improvement suggestions focus on the shortcomings in the single sentence (e.g., "not mentioning one's own organization may reduce customer trust"); and recommended scripts provide optimized examples that fit the current scenario (e.g., "Hello, Mr. Su! I am Zhang from xxx Insurance").

[0062] The phase review includes a comprehensive evaluation of the completed phases when the agent achieves the current phase goals and moves to the next phase. The results include a "phase score," "strengths and weaknesses analysis," and "improvement suggestions." The phase score is calculated by summing the scores of all assessment points for that phase (e.g., the appointment phase has a maximum score of 10 points, and the agent actually scores 8 points). The strengths and weaknesses analysis objectively summarizes the agent's performance highlights (e.g., "successfully reducing the client's wariness through referrals and clearly agreeing on a meeting time") and shortcomings (e.g., "failing to proactively mention the actual benefits of tax incentives to the client"). The improvement suggestions provide targeted optimization directions for the phase's shortcomings (e.g., "In future phases, you could add specific wording like 'Tax incentives can help you save XX yuan in taxes' to enhance client interest").

[0063] The overall review includes summarizing the evaluation data from each stage after the agent completes all training phases, generating an overall result that includes "Total Score Review," "Strengths and Weaknesses Assessment," and "Improvement Suggestions." The total score review is the weighted average of the scores from each stage (if each stage has the same weight, the five stages are scored 8, 7, 9, 6, and 8 points respectively, for a total score of 7.6 points); the strengths and weaknesses assessment summarizes core strengths (e.g., "skilled at guiding conversations through customer interests, clear approach to handling objections") and core weaknesses (e.g., "lack of data support and insufficient persuasiveness in product solution explanations") from a holistic perspective; the improvement suggestions cover optimization directions throughout the entire process (e.g., "suggest adding a case study of 'customers in the same customer group paying XX yuan annually and enjoying XX protection' during the product explanation phase, while strengthening the explanation of the calculation logic of tax incentive policies"). In this way, subjective biases are eliminated by using unified evaluation standards, and the three-level evaluation of "single sentence, stage, and overall" enables immediate correction, stage review, and overall optimization, allowing agents to accurately identify their shortcomings and improve them efficiently. At the same time, the anomaly detection and anti-cheating mechanism not only intercepts compliance risks in advance but also avoids formalized training, ensuring that the training results are genuine and effective, which can significantly improve the agents' communication standardization, professionalism, and practical application ability.

[0064] The risk detection submodule includes an abnormal dialogue detection unit and an anti-cheating detection unit. The abnormal dialogue detection unit detects whether the content provided by the current agent violates the principles of authenticity, ethical standards, company policies, and is irrelevant to the objective. When an anomaly is detected, a counter is triggered to accumulate. When the count reaches a preset threshold, the training session is automatically terminated and the training session is deemed unqualified.

[0065] The anti-fraud detection unit uses a text similarity algorithm to detect the similarity between the agent's script and the standard script. If the similarity exceeds a preset threshold, the corresponding target score is canceled, and a result containing evaluation (such as "[High similarity to the standard script detected, please express the content in your own words] Lacking specific examples, it is recommended to add similar examples") and optimization suggestions (such as "Reference script: I have already processed the tax refund for Mr. X before, and he received a tax refund of XXX yuan").

[0066] The knowledge extraction intelligent agent module is the core unit for system knowledge accumulation and iteration. It includes a multi-source data access submodule, a structured extraction submodule, and a knowledge output submodule, with the following specific functions:

[0067] The multi-source data access submodule supports access to various types of data, including product materials, training materials, and online logs. Product materials include insurance product terms and conditions, coverage description documents, etc.; training materials include historical training records and instructor courseware, etc.; and online logs include agent operation logs and customer interaction logs, etc. This submodule can achieve batch import and real-time synchronization of data through API interfaces, file uploads, and other methods.

[0068] The structured extraction submodule is used to parse and extract information from multi-source data, including product material analysis, extracting information on the functions, parameters, and usage scenarios of company products, such as the payment period, payment method, and coverage amount of a certain pension insurance product, as well as the usage scenarios suitable for "young and middle-aged white-collar workers in large and medium-sized enterprises and institutions"; training material generation, summarizing high-frequency questions (such as "what to do if a customer questions the returns of their personal pension account") and best practices (such as "answering questions by combining tax incentive policies with case studies of compound interest returns in accounts") based on historical training data; and online log analysis, identifying user behavior patterns (such as agents spending too much time on average during the "product solution explanation" stage) or common errors (such as frequently omitting key information such as "tax incentive amount calculation").

[0069] The knowledge output submodule will extract information from the structured extraction submodule and generate reusable training materials (such as product knowledge manuals and frequently asked questions documents) or optimization suggestions (such as training improvement plans for agents' high-frequency errors). These materials will be synchronized to the course creation and review modules through data interfaces to provide knowledge support for training content generation and team capability improvement.

[0070] The course creation intelligent module is the core unit for generating training content. Based on the system's knowledge framework and training needs, it includes a knowledge point matching submodule, a customer profile association submodule, and a material generation submodule. Its specific functions are as follows:

[0071] The knowledge point matching submodule filters core knowledge points that are strongly related to insurance sales scenarios from the structured knowledge synchronized by the knowledge extraction intelligent agent module. This covers core product functions such as pension insurance payment and withdrawal rules, policy points such as the calculation logic of tax incentive policies, and communication skills such as the framework for handling objections. It also supports filtering priority knowledge points according to training objectives such as "basic training for new employees" and "elite sales improvement" to form a hierarchical knowledge pool.

[0072] The customer profiling submodule connects to the role information database of the customer intelligent agent module, matching appropriate knowledge points and training focuses based on different customer group characteristics (such as "Shenzhen high-income professionals," "young and middle-aged white-collar workers," and "parents of a three-person family"). For example, for customer profiles with "aggressive investment preferences," knowledge points such as "comparison of high-yield insurance products" and "asset allocation logic" are prioritized; for customer profiles with "new customer referrals," the focus is on linking content such as "referrer trust endorsement techniques" and "methods for mining new customer needs."

[0073] The material generation submodule automatically generates diverse training materials based on the matching results between knowledge points and customer profiles. These training materials include: test questions, which generate multiple-choice and short-answer questions around knowledge points, such as a calculation question for tax incentive policies: "A customer with an annual income of 220,000 yuan can deduct the maximum amount of tax per year under the tax incentive policy?"; training materials, which create knowledge point manuals, case compilations (such as "Insurance Cases of Customers in the Same Customer Group"), and sales script templates (such as "Product Explanation Scripts for High-Income Customers"); test points, which extract the core assessment points from the knowledge points, such as "Product explanations should include three core elements: payment period, coverage scope, and benefit calculation"; and scripts, which generate practice scripts adapted to specific customer profiles, such as a script for "35-year-old married high-income earners with children" that includes sections such as "family protection needs assessment" and "children's education fund planning". By accurately matching knowledge points with customer profiles, the problem of "general content that is out of touch with actual business development" in traditional training is avoided; automated material generation greatly reduces the cost of manual course production, and the generated test questions, scripts and other materials can be directly connected to practice and assessment scenarios, realizing the integration of "training, practice and assessment", allowing agents to quickly transform what they have learned into business development capabilities.

[0074] The intelligent agent module is the core unit for improving team capabilities. It focuses on the analysis and application of training data from agent teams. Internally, it includes sub-modules for group data access, capability assessment, and learning material recommendation. Specific functions are as follows:

[0075] The group data access submodule receives all data from agent group training sessions, including the overall evaluation results of each group (such as total score and score distribution at each stage), records of individual agents' weaknesses (such as "average score of only 6 points in the product solution explanation stage"), frequency of abnormal script triggers, etc. It also supports filtering data by group dimension (such as "newcomer group" and "regional team group") and time dimension (such as "monthly" and "quarterly") to form a structured data dashboard.

[0076] The capability assessment submodule analyzes the team's overall capability through data modeling. Specifically, it calculates the average score and score percentage for each stage by creating an overall capability profile, identifying the team's strengths (e.g., an average score of 8.5 in the appointment stage) and weaknesses (e.g., an average score of 5.8 in the objection handling stage). It identifies frequently occurring weaknesses by identifying common problems, such as "80% of agents omit revenue calculations during product explanations" and "60% of agents do not use customer honorifics". Through individual difference analysis, it compares individual scores within the group to identify those with outstanding capabilities (e.g., total scores above 9) and those needing improvement (e.g., total scores below 6), forming a tiered capability list.

[0077] The learning materials recommendation submodule, based on the capability assessment results, provides targeted recommendations of learning materials from the training material library generated by the course creation intelligent agent module. Specifically, for areas where the team is weak (such as objection handling), it recommends "Objection Response Script Manual" and "Case Studies of Answering Common Customer Questions"; for common issues (such as missing revenue calculations), it recommends "Product Revenue Calculation Tutorial Videos" and "Revenue Explanation Scripts"; and for individual differences, it recommends basic materials (such as "New Salesperson Sales Process Guide") for those needing improvement and advanced materials (such as "High-Net-Worth Client Sales Techniques") for those with outstanding capabilities. Through data analysis, it breaks through the limitations of traditional "judging team capabilities based on experience," accurately identifying the weaknesses of the team and individuals; targeted materials enable focused learning and improve training efficiency; and the accumulation of group data tracks team capability improvement trends, providing data support for insurance companies to develop training plans and optimize team configurations.

[0078] The above embodiments describe an intelligent insurance sales coaching system based on multi-agent technology provided in this application. In addition, this application also provides a method for using the system. Figure 3 A flowchart illustrating a multi-agent-based intelligent coaching method for insurance sales, provided as an embodiment of this application. (Combined with...) Figure 3 As shown in the embodiments of this application, the intelligent insurance sales coaching method based on multi-agent systems may include:

[0079] S301. Generate exam questions, training materials, exam points, and scripts based on the course-making intelligent agent, combining knowledge points and customer profiles;

[0080] S302. Load customer role information and training script through customer intelligent agent, conduct interactive drills with agent, and ask questions or challenges autonomously when the agent's response is ambiguous.

[0081] S303. The review agent loads the target table of the corresponding script based on the question and answer results of the customer agent and the agent, and performs single-sentence review, stage review, and overall review on the agent's response, while performing abnormal speech detection and anti-cheating detection.

[0082] S304. Review and analyze the overall evaluation results of the intelligent agent's group training sessions for agents, assess the overall ability of the team, and recommend personalized learning materials.

[0083] S305, the knowledge extraction intelligent agent extracts structured knowledge from multi-source data such as product materials, training materials, and online logs to generate reusable training materials or optimization suggestions.

[0084] In one implementation of this application, the customer role information loaded by the customer intelligent agent includes: name, title, age, gender, occupation, location, whether it is a new customer, customer type, customer group category, and potential information. The potential information includes education level, annual income, hobbies, investment preferences, asset size, family structure, and personality traits.

[0085] In one implementation of this application, the training script loaded by the customer intelligent agent includes multiple stages, each stage defining reference scripts for sales personnel and customers, as well as goals and corresponding script fragments related to the scripts.

[0086] In one implementation of this application, the single-sentence review by the review agent is used to generate a result that includes test point scores, improvement suggestions, and script recommendations;

[0087] The phase review is used to provide feedback on the completed phases, including phase scores, strengths and weaknesses analysis, and improvement suggestions, when the agent completes the current phase goal and moves to the next phase.

[0088] The overall review is used to provide a total score, assessment of strengths and weaknesses, and suggestions for improvement after the agent has completed all stages.

[0089] In one implementation of this application, the abnormal speech detection dimensions of the review agent include:

[0090] The system checks whether the content provided by the current agent violates authenticity, ethical standards, company policies, or is irrelevant to the objective.

[0091] In one implementation of this application, the anti-fraud detection is used to detect the similarity between the agent's script and the standard script. If the similarity exceeds a preset similarity threshold, the corresponding target score is canceled and a result containing evaluation and optimization suggestions is generated.

[0092] In one implementation of this application, the knowledge extraction agent's analysis of product materials includes extracting information on the company's products' functions, parameters, and usage scenarios.

[0093] Training material generation includes summarizing frequently asked questions and best practices based on historical training data;

[0094] Online log analysis includes identifying user behavior patterns or common errors and feeding them back to product iterations.

[0095] This application also provides corresponding devices and computer storage media for implementing the solutions provided in this application.

[0096] The device includes a memory and a processor. The memory stores instructions or code, and the processor executes the instructions or code to cause the device to perform the method described in any embodiment of this application.

[0097] The computer storage medium stores code, and when the code is run, the device running the code implements the method described in any embodiment of this application.

[0098] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus a general-purpose hardware platform. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0099] It is understood that in the specific embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved need to obtain user permission or consent when the above embodiments of this application are applied to specific products or technologies, and the collection, use and processing of related data need to comply with the relevant laws, regulations and standards of relevant countries and regions.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0101] It should also be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0102] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-agent-based intelligent training system for insurance sales, characterized in that, The system includes a customer intelligent agent, a review intelligent agent, a course creation intelligent agent, a review intelligent agent, and a knowledge extraction intelligent agent; The client intelligent agent is used to load client role information and training script, interact and practice with the agent, and autonomously ask questions or challenges when the agent's response is ambiguous. The review agent is used to load the target table of the corresponding script, and to provide single-sentence reviews, stage reviews, or overall reviews of the agent's responses, while also performing abnormal wording detection and anti-cheating detection. The course-making intelligent agent is used to generate exam questions, training materials, exam points, and scripts by combining knowledge points and customer profiles; The review agent is used to analyze the overall evaluation results of the agent group training, assess the overall ability of the team, and recommend personalized learning materials; The knowledge extraction agent is used to extract structured knowledge from multi-source data, including product materials, training materials, and online logs, to generate reusable training materials or optimization suggestions.

2. The system according to claim 1, characterized in that, The customer role information loaded by the customer intelligent agent includes: Name, title, age, gender, occupation, location, whether it is a new customer, customer type, customer group category, and potential information, including education level, annual income, hobbies, investment preferences, asset size, family structure, and personality traits.

3. The system according to claim 1, characterized in that, The training script loaded by the customer's intelligent agent contains multiple stages. Each stage defines reference scripts for sales personnel and customers, as well as goals and corresponding script segments related to the scripts.

4. The system according to claim 1, characterized in that, The single-sentence comments of the review agent are used to generate results that include test point scores, improvement suggestions, and recommended phrases. The phase review is used to provide feedback on the completed phases, including phase scores, strengths and weaknesses analysis, and improvement suggestions, when the agent completes the current phase goal and moves to the next phase. The overall review is used to provide a total score, assessment of strengths and weaknesses, and suggestions for improvement after the agent has completed all stages.

5. The system according to claim 1, characterized in that, The abnormal speech detection dimensions of the commenting agent include: The system checks whether the content provided by the current agent violates authenticity, ethical standards, company policies, or is irrelevant to the objective.

6. The system according to claim 1, characterized in that, The anti-fraud detection is used to detect the similarity between the agent's script and the standard script. If the similarity exceeds a preset threshold, the corresponding target score is canceled and a result containing evaluation and optimization suggestions is generated.

7. The system according to claim 1, characterized in that, The knowledge extraction agent analyzes product materials by extracting information on the company's products' functions, parameters, and usage scenarios. Training material generation includes summarizing frequently asked questions and best practices based on historical training data; Online log analysis includes identifying user behavior patterns or common errors and feeding them back to product iterations.

8. A method for intelligent coaching in insurance sales based on multi-agent systems, characterized in that, The method applied to the multi-agent-based intelligent insurance sales coaching system of claim 1, the method comprising: Based on the course-making intelligent agent, test questions, training materials, test points and scripts are generated by combining knowledge points and customer profiles; The client intelligent agent loads client role information and training scripts, and conducts interactive drills with the agent. When the agent's response is ambiguous, the agent will ask questions or raise doubts. Based on the question-and-answer results between the customer agent and the agent, the review agent loads the target table of the corresponding script and performs single-sentence review, stage review, and overall review on the agent's response, while also performing abnormal wording detection and anti-fraud detection. The overall evaluation results of the intelligent agent's group training sessions were reviewed and analyzed to assess the team's overall capabilities and recommend personalized learning materials. The knowledge extraction agent extracts structured knowledge from multi-source data such as product materials, training materials, and online logs to generate reusable training materials or optimization suggestions.

9. A computing device, characterized in that, The computing device includes: a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the steps of the method as described in claim 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in claim 8.

Images