An insurance dialogue recommendation and intelligent evaluation system and method

Abstract

The present application relates to the technical field of artificial intelligence, and specifically discloses a kind of insurance speech technique recommendation and intelligent evaluation system and method.The technical scheme of the present application generates pre-generated speech corresponding to the portrait combination in batches offline based on the insurance customer attribute label system calling large language model, matches and pushes pre-generated speech according to the sales rhythm stage in the first call, dynamically generates the next call speech based on the last call content when not the first call, and after the call ends, the recording is transcribed into text for structured scoring and improvement point identification to generate historical communication summary, while periodically extracting the excellent speech in the call of the performance excellent seat into the speech library, realizing the personalized recommendation and dynamic adjustment of the insurance speech, improving the comprehensiveness of call quality evaluation and feedback timeliness, and promoting the continuous optimization of the overall speech ability of the team.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an insurance sales pitch recommendation and intelligent assessment system and method. Background Technology

[0002] In property insurance telesales, communication between agents and customers has long relied on three traditional models. The first is a standardized script system, pre-developed by the business department and stored in a structured database for agents to study and reference before making calls. While the script is organized according to sales process stages, it uses the same script for all customers without tailoring it to individual customer characteristics. The second model involves manual quality control and coaching. After agents are trained, coaches manually sample call recordings and provide improvement suggestions. However, due to limited manpower, the sampling coverage is low, making it impossible to conduct timely and comprehensive quality assessments for every call. The third model relies primarily on word-of-mouth transmission from high-performing agents or centralized training sessions, resulting in a lack of systematic extraction and consolidation of excellent communication experience and script skills.

[0003] The aforementioned traditional model has several shortcomings in practical application. Standardized scripts are only broadly categorized according to sales process stages, failing to differentiate recommendations based on customers' actual attributes. This results in agents using only uniform scripts during calls, leading to limited talking points, poor customer communication experience, and difficulty in further improving conversion rates. Call quality assessment relies on manual sampling, which is inefficient and lacks coverage. Lessons learned from numerous calls cannot be accumulated, and personalized improvement directions for agents cannot be identified in a timely and accurate manner. Furthermore, there is a lack of systematic monitoring methods to determine whether agents are communicating according to recommended scripts and the quality of their communication. The effectiveness of script recommendations lacks closed-loop evaluation, making continuous optimization based on actual usage data impossible. The excellent script handling strategies of top-performing agents in different business scenarios and objection types lack automated extraction, refinement, and distribution mechanisms, making it difficult to continuously translate them into a shared capability for the entire team.

[0004] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an insurance sales pitch recommendation and intelligent evaluation system and method.

[0006] Firstly, the present invention provides an insurance sales script recommendation and intelligent evaluation system, the technical solution of which is as follows: The pre-generated script module is used to call the large language model to generate pre-generated scripts and key points corresponding to the profiles of each insurance customer offline, based on the attribute tag system and standard script library of insurance customers, and store them in the script library. The script matching and recommendation module is used to obtain the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the number of calls is the first time, the pre-generated script corresponding to the profile combination of the insurance customer is matched from the script library and pushed to the insurance agent terminal. If the number of calls is not the first time, the large language model is called to generate the next script based on the dialogue content of the previous call and pushed to the insurance agent terminal. The ASR transcription module is used to transcribe the call recording into call text after the call ends, and to preprocess the call text to obtain the preprocessed call text. The script adoption evaluation module is used to call the large language model to perform structured scoring on the actual scripts used by insurance agents in the preprocessed call text, identify areas for improvement, and generate insurance agent reviews. Based on the structured scoring, a structured summary of a single call is generated, and the structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output. The high-performing experience extraction module is used to periodically obtain the transaction call texts of high-performing insurance agents, call the large language model to extract excellent scripts from the transaction call texts, and store them in the script library after review. The dialogue data stored in the dialogue database is used by the dialogue matching and recommendation module to perform matching and retrieval based on the profile combination, and by the dialogue pre-generation module as the basis for generating dialogues offline in batches.

[0007] The beneficial effects of the insurance sales script recommendation and intelligent evaluation system of the present invention are as follows: This invention's system generates pre-generated scripts offline in batches based on a large language model and customer profiles, using an insurance customer attribute tagging system. During the first call, the system matches and pushes pre-generated scripts according to the sales rhythm stage. For subsequent calls, it dynamically generates the next script based on the previous conversation. After the call, the recording is transcribed into text for structured scoring and improvement point identification to generate a historical communication summary. Simultaneously, it periodically extracts excellent scripts from top-performing agents' closing calls and stores them in a script library. This solves the problems of traditional standard scripts lacking customer-specific adaptation, resulting in limited talking points and poor communication experience; insufficient coverage of manual quality inspection and delayed identification of personalized improvement directions; and a lack of systematic accumulation and distribution of excellent communication experience. It achieves personalized recommendation and dynamic adjustment of insurance scripts, improves the comprehensiveness and timeliness of call quality assessment, and promotes continuous optimization of the team's overall script capabilities.

[0008] Based on the above solution, the insurance sales script recommendation and intelligent evaluation system of the present invention can be further improved as follows.

[0009] In one alternative approach, the insurance customer attribute tagging system includes six dimensions: sales pace, business category, instant messaging addition status, claims attention, service sensitivity, and service sales status. The attribute values ​​of each dimension are combined to form various combinations of insurance customer profiles.

[0010] The beneficial effects of adopting the above optional methods are as follows: further construct an attribute tag system through six dimensions: sales rhythm, business category, instant messaging addition status, claims attention, service sensitivity, and service sales status. The combination of attribute values ​​of each dimension forms a variety of profile combinations, providing a refined customer classification basis for generating differentiated communication scripts and improving the comprehensiveness of profile construction.

[0011] In one alternative approach, the sales rhythm is automatically divided based on the expiration date of the insurance customer's auto insurance policy. The business categories include outbound new business and renewal business. The instant messaging addition status indicates whether the insurance customer has added the agent's instant messaging account. The claims attention level indicates whether the insurance customer shows concern about claims matters. The service sensitivity indicates the insurance customer's level of interest in additional services. The service sales status indicates the insurance customer's progress in receiving and using service marketing activities.

[0012] The beneficial effects of adopting the above optional methods are as follows: the sales rhythm is further clarified based on the automatic division of the car insurance policy expiration date, the business categories are distinguished between outbound new business and renewal business, and the other dimensions correspond to the specific meanings of instant messaging addition status, claims attention, service sensitivity and service sales status, so that the tag data has a quantifiable source and improves the accuracy and practicality of profile construction.

[0013] In one alternative approach, the pre-generated script includes three parts: an opening statement, a profile-featured script, and a closing action. Specifically, the script pre-generation module is used to call the large language model to extract the key points of the script for each part of the pre-generated script.

[0014] The beneficial effects of adopting the above optional method are as follows: the pre-generated script is further divided into three parts: opening remarks, profile feature script, and closing action. The large language model is used to extract the key points of the script for each part, making the script content clear and the structure standardized. This makes it easier for agents to quickly grasp the key points of communication before the call, and improves the convenience of script generation and use.

[0015] In one alternative approach, the script pre-generation module is specifically used for: When the large language model is invoked to generate the pre-generated script, the corresponding script is matched as the opening line based on the insurance customer's sales status and instant messaging addition status. The corresponding script is matched as the profile feature script based on the insurance customer's claims attention and service sensitivity. After the matching is completed, the opening line, the profile feature script and the closing action are seamlessly connected in a conversational style.

[0016] The beneficial effects of adopting the above optional method are as follows: when generating pre-generated scripts, the opening remarks are matched according to the sales status and instant messaging status, and the profile feature scripts are matched according to the claims attention and service sensitivity. Then, the three parts are seamlessly connected in a conversational style, so that the generated scripts are in line with the actual attributes of customers and are expressed naturally and fluently, thereby improving the adaptability of script recommendations.

[0017] In one alternative approach, the dialogue matching and recommendation module is specifically used for: The current sales rhythm stage of the insurance customer is calculated based on the expiration date of the insurance customer's auto insurance policy, and the number of calls made by the insurance customer within the same sales rhythm stage is obtained as the number of calls. When the insurance customer enters the next sales rhythm stage, the number of calls is reset to the first time.

[0018] The beneficial effects of adopting the above optional method are as follows: further calculate the current sales rhythm stage of the insurance customer based on the expiration date of the car insurance policy, take the number of calls made in the same stage as the number of calls, and reset it to the first time when entering the next stage, so that the script recommendation strategy is accurately aligned with the policy sales cycle, and realizes the automatic judgment of the switching of scripts between the first and non-first calls.

[0019] In one alternative approach, the script adoption evaluation module is specifically used for: After receiving the preprocessed call text, semantic matching evaluation is performed on the opening remarks, profile feature-based dialogue, and closing action sections of the actual dialogue, comparing them with the pre-generated dialogue or the key points of the dialogue. Each part is given a structured score, which is then weighted to obtain an overall communication quality score. Based on the overall communication quality score, the areas for improvement are identified, and the insurance agent's review is generated.

[0020] The beneficial effects of adopting the above-mentioned optional method are as follows: the actual dialogue is further divided into three parts: opening remarks, dialogue with profile characteristics, and closing actions. The semantic matching evaluation is carried out against the pre-generated dialogue or dialogue key points. After obtaining the structured scores of each part, the overall communication quality score is obtained by comprehensive weighting. Based on this, the improvement points are identified and agent comments are generated, making the call quality evaluation more granular.

[0021] Secondly, this invention provides a method for recommending and intelligently evaluating insurance sales scripts. The method employs the insurance sales script recommendation and intelligent evaluation system provided by this invention, and the technical solution is as follows: Based on the insurance customer attribute tag system and standard script library, the large language model is called to generate pre-generated scripts and key points of the scripts corresponding to the profiles of each insurance customer offline in batches and store them in the script library. The system obtains the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the number of calls is the first time, it matches the pre-generated script corresponding to the profile combination of the insurance customer from the script library and pushes it to the insurance agent terminal. If the number of calls is not the first time, it calls the large language model to generate the next script based on the dialogue content of the previous call and pushes it to the insurance agent terminal. After the call ends, the call recording is transcribed into a text file, and the text file is preprocessed to obtain a preprocessed text file. The large language model is invoked to perform structured scoring on the actual dialogue used by the insurance agent in the preprocessed call text, identify areas for improvement, and generate insurance agent reviews. Based on the structured scoring, a structured summary of a single call is generated, and the structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output. The system periodically retrieves the transaction texts of top-performing insurance agents, calls the large language model to extract excellent scripts from the transaction texts, and stores them in the script library after review.

[0022] The beneficial effects of the insurance sales pitch recommendation and intelligent evaluation method of the present invention are as follows: This invention's method utilizes an insurance customer attribute tagging system to call a large language model to offline batch generate pre-generated scripts corresponding to customer profiles. During the first call, pre-generated scripts are matched and pushed according to the sales rhythm stage. For subsequent calls, scripts for the next call are dynamically generated based on the previous conversation. After the call ends, the recording is transcribed into text for structured scoring and improvement point identification to generate a historical communication summary. Simultaneously, excellent scripts from top-performing agents' closing calls are periodically extracted and stored in a script library. This solves the problems of traditional standard scripts lacking customer-specific adaptation, resulting in limited talking points and poor communication experience; insufficient coverage of manual quality inspection and delayed identification of personalized improvement directions; and a lack of systematic accumulation and distribution of excellent communication experience. It achieves personalized recommendation and dynamic adjustment of insurance scripts, improves the comprehensiveness and timeliness of call quality assessment, and promotes continuous optimization of the team's overall script capabilities.

[0023] Thirdly, the technical solution of an electronic device according to the present invention is as follows: It includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the insurance sales pitch recommendation and intelligent assessment method of the present invention.

[0024] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows: The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the insurance sales pitch recommendation and intelligent assessment method of the present invention.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of the present invention more obvious and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an embodiment of the insurance sales pitch recommendation and intelligent evaluation system of the present invention; Figure 2 This is a schematic diagram of the overall architecture; Figure 3 A schematic diagram of the logic flow for adaptive switching based on sales rhythm; Figure 4 A schematic diagram of the evaluation process for the adoption assessment module based on a large language model; Figure 5 A schematic diagram of the six-dimensional attribute tagging system and the combination of profile features and sales scripts for insurance customers; Figure 6 This is a schematic diagram of the monthly cycle operation process of the high-performing experience extraction module; Figure 7 This is a flowchart illustrating an embodiment of the insurance sales pitch recommendation and intelligent evaluation method of the present invention; Figure 8 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0027] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0028] Figure 1This diagram illustrates the structure of an embodiment of an insurance sales script recommendation and intelligent assessment system 100 provided by the present invention. Figure 1 As shown, the insurance sales script recommendation and intelligent evaluation system 100 includes: The pre-generated script module 101 is used to call the large language model offline to generate pre-generated scripts and key points corresponding to the profiles of each insurance customer based on the attribute tag system and standard script library of insurance customers, and store them in the script library.

[0029] Among them, insurance customers refer to natural persons who purchase or intend to purchase insurance products and accept insurance telephone sales services; for example, Mr. A's car insurance policy expires on July 1, 2026, and Mr. A answers a sales call made by Ms. B, an insurance agent.

[0030] The attribute tagging system refers to a set of categorized attributes used to describe the multidimensional characteristics of insurance customers. Each attribute value is used to create a differentiated customer profile. For example, Mr. A's attribute tags include "sales pace: first day," "business category: renewal business," "instant messaging addition status: not added," "claims attention level: high," "service sensitivity: interested in service," and "service sales status: issued but not received." The standard script library refers to a collection of general script texts pre-developed by the business department that comply with compliance requirements and sales processes, for agents to learn and reference. For example, the standard script library stores the opening template "Hello, I am a customer service representative from Insurance Company C. Are you Mr. A?", the objection handling template "You mentioned a price issue; we can introduce the coverage to you," and the closing template "Please check the link first, then decide whether to proceed."

[0031] In this context, the profile combination refers to the customer classification result formed by combining the values ​​of multiple attribute tags of an insurance customer. For example, Mr. A's attribute tag value combination is "first day plus renewal business plus no instant messaging plus high claims attention plus service interest plus issued but not claimed". This combination is Mr. A's profile combination.

[0032] The pre-generated script refers to a three-part personalized recommendation script text generated offline in batches by a large language model based on the profile of the insurance customer before the call takes place. For example, for Mr. A's profile, a pre-generated script is generated in advance, with the opening line being "Hello Mr. A, your car insurance expires today. We have a renewal plan that includes value-added services. Would it be convenient to add me on WeChat to send it to you?", the profile feature script being "You have previously been concerned about claims. Our services include free roadside assistance and a replacement vehicle," and the closing action being "Please check the details through the link first, and then proceed if you think it's suitable." The script key points refer to the core semantic information extracted from each segment of the pre-generated script, which is used by agents for flexible reference during actual calls. For example, the key points of the opening line are "Car insurance expires today, renewal plan, add me on WeChat to send it to you," the key points of the profile feature script are "Claims concern, roadside assistance, replacement vehicle," and the key points of the closing action are "Check the link, and then proceed if it's suitable."

[0033] The script library refers to a database that centrally stores pre-generated scripts, key points of scripts, standard scripts, and excellent scripts extracted later. It supports conditional retrieval and matching based on profile combinations. For example, the script library stores pre-generated scripts corresponding to Mr. A's profile combination and his three key points of scripts, as well as standard script templates and excellent objection handling scripts from top-performing agents, indexed by tags.

[0034] Specifically, an attribute tagging system for insurance customers is constructed, a standard script library is obtained, and the attribute values ​​of each dimension in the attribute tagging system are combined to form multiple profile combinations. For each profile combination, a large language model is invoked, and based on the script content in the standard script library, the corresponding three-part pre-generated script is generated offline in batches. The pre-generated script includes an opening, a profile feature script, and a closing action. The core points are extracted from each pre-generated script as script key points, and the pre-generated script and script key points are stored in the script library.

[0035] The script matching and recommendation module 102 is used to obtain the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the number of calls is the first time, the pre-generated script corresponding to the profile combination of the insurance customer is matched from the script library and pushed to the insurance agent terminal. If the number of calls is not the first time, the large language model is called to generate the next script based on the dialogue content of the previous call and pushed to the insurance agent terminal.

[0036] The sales rhythm phase refers to nine time phases automatically divided based on the expiration date of the insurance customer's car insurance policy, used to determine the sales process at which the call is taking place; for example, Mr. A's car insurance policy expires on July 1, 2026, and the sales rhythm phase on July 1, 2026 is determined as "Day 1".

[0037] Herein, the call count refers to the cumulative number of calls that an insurance agent has already made with the same insurance customer within the same sales rhythm phase; for example, before Miss Yi calls Mr. Jia on July 1, 2026, it is queried that within this "first day" phase, Mr. Jia has not been called yet, so the call count is zero, and this call is the first call. The insurance agent terminal refers to an electronic device used by the insurance agent to make and receive calls, view recommended scripts, and input information; for example, Miss Yi uses a softphone terminal of the telemarketing system on a computer to dial Mr. Jia's mobile phone number, and the pre-generated script text pushed is simultaneously displayed on the terminal screen.

[0038] Specifically, the current sales rhythm phase of the insurance customer and the call count within the sales rhythm phase are acquired. If the call count is the first time, a corresponding pre-generated script is matched from the script library according to the portrait combination of the insurance customer, and the pre-generated script is pushed to the insurance agent terminal. If the call count is not the first time, the dialogue content of the previous call is acquired, a large language model is invoked to generate the next call script according to the dialogue content of the previous call, and the next call script is pushed to the insurance agent terminal.

[0039] The ASR transcription module 103 is configured to transcribe the call recording into call text after the call ends, and perform preprocessing on the call text to obtain a preprocessed call text.

[0040] Herein, the call recording refers to a complete audio file automatically recorded when the insurance agent and the insurance customer communicate by telephone; for example, the entire call between Miss Yi and Mr. Jia is recorded as an MP3 format recording file with a duration of about 5 minutes. The call text refers to the original text content obtained by transcribing the call recording through automatic speech recognition; for example, the call recording between Mr. Jia and Miss Yi is transcribed into an original text draft containing dialogue contents such as "Hello, is that Mr. Jia?" and "Yes, who is that?".

[0041] Herein, the preprocessed call text refers to a standardized text obtained by performing operations such as removing modal particles, removing repeated words, correcting wrong characters, and adjusting word order on the original call text, which is used for inputting into the large language model for evaluation; for example, after preprocessing, the original text "Hmm, well, I'll think about it" is standardized to "I'll think about it".

[0042] Specifically, automatic speech recognition transcription is triggered after the call ends, the call recording is transcribed into call text, preprocessing operations are performed on the call text, and the preprocessing operations include removing modal particles, removing repeated words, correcting common wrong characters, and adjusting word order, so as to obtain the preprocessed call text.

[0043] The script adoption evaluation module 104 is used to call the large language model to perform structured scoring on the actual scripts used by insurance agents in the preprocessed call text, identify areas for improvement, and generate insurance agent reviews. Based on the structured scoring, a structured summary of a single call is generated, and the structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output.

[0044] Here, "insurance agent" refers to a customer service representative employed by an insurance company who specializes in telephone sales. For example, Ms. B is an agent at the car insurance telemarketing center of Insurance Company C, with employee number 023. "Actual script" refers to the actual language spoken by the insurance agent during a call with an insurance customer. For example, Ms. B's actual opening line during the call was, "Hello Mr. A, your car insurance expires today. We have a renewal discount available; would you mind listening to it?"

[0045] Structured scoring refers to the quantitative score given by the large language model based on the semantic matching degree of the insurance agent's actual speech relative to the pre-generated speech or key points of the speech, according to preset scoring dimensions (opening remarks, profile-featured speech, closing actions) and grading standards. For example, Ms. Yi's actual opening remarks scored 75 points, profile-featured speech scored 68 points, and closing actions scored 55 points. Areas for improvement refer to the most typical speech defect identified from the lower-scoring dimensions, used to guide agents to improve communication quality. For example, for the closing action score of 55, the identified area for improvement is "failure to proactively guide the customer to add an instant messaging account, and failure to provide low-threshold action suggestions." Insurance agent feedback refers to the evaluative comments given to the insurance agent's communication performance in concise sentences based on the scoring results of each speech. For example, Ms. Yi's feedback is "the opening remarks were natural and fluent, the response to objections was not specific enough, and the closing process needs stronger guidance."

[0046] The structured summary refers to a structured record generated after a single call, including three parts: individual scores, overall score, reason analysis, areas for improvement, and agent feedback. For example, the structured summary for this call includes: an opening statement of 75 points with the reason "complete self-introduction but no mention of adding instant messaging"; profile-characteristic dialogue of 68 points with the reason "mentioned claims but did not elaborate on service content"; closing actions of 55 points with the reason "no specific action guidance provided"; an overall score of 65 points; areas for improvement "lack of guidance in the closing process"; and feedback "overall performance is average, closing needs improvement." The historical communication summary refers to a summary of the overall communication profile and pace for the same insurance customer, generated by merging the structured summaries of all previous calls. For example, merging the summaries of Mr. Jia's current call with the previous call generates a historical communication summary: "The agent failed to successfully add instant messaging twice; the customer showed interest in claims and additional services, but no deal was closed in either instance."

[0047] The evaluation result refers to the complete set of evaluation data output by the script adoption evaluation function, which includes a structured summary, historical communication summary, scores for each part, points for improvement, and comments. For example, the evaluation result includes a complete structured summary of this call, Mr. A's historical communication summary, and specific improvement suggestions for Ms. B.

[0048] Specifically, a large language model is invoked to perform structured scoring on the actual dialogue used by insurance agents in the preprocessed call text. The structured scoring includes semantic matching evaluation of the opening remarks, profile-featured dialogue, and closing actions of the actual dialogue against pre-generated dialogue or key points of the dialogue, obtaining structured scores for each of the three parts. The structured scores of the three parts are then combined and weighted to obtain an overall communication quality score. Based on the overall communication quality score, areas for improvement are identified and insurance agent reviews are generated. A structured summary of a single call is generated based on the structured scores. The structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output.

[0049] The high-performing experience extraction module 105 is used to periodically obtain the transaction call texts of high-performing insurance agents, call the large language model to extract excellent scripts from the transaction call texts, and store them in the script library after review.

[0050] Among them, "top-performing insurance agents" refers to outstanding telephone sales agents who rank highly in performance and have a high conversion rate within a certain period; for example, Mr. Ding, the top-ranked agent in a branch of Insurance Company C last month, was identified as a top-performing insurance agent. "Contact transcript" refers to the text content obtained by transcribing the recording of a successful insurance policy transaction by a top-performing insurance agent; for example, the transcript of Mr. Ding's most recent transaction was approximately 800 words long.

[0051] Among them, excellent sales scripts refer to effective responses to specific objection scenarios automatically extracted from the transaction texts of top-performing insurance agents by a large language model; for example, the excellent sales script extracted from Mr. Ding's transaction call is "You feel it is more expensive than last year because roadside assistance and replacement vehicle services have been added this year, so it only costs a few cents more per day."

[0052] The dialogue data stored in the dialogue database is used by the dialogue matching and recommendation module 102 to perform matching and retrieval based on the profile combination, and by the dialogue pre-generation module 101 as the basis for generating dialogues in offline batches.

[0053] The matching retrieval function refers to the process by which the script matching and recommendation function searches for and retrieves corresponding pre-generated scripts in the script library based on the profile of the insurance customer. For example, the matching retrieval function accurately matches the pre-generated script text in the script library based on Mr. A's profile combination of "first-day renewal, no instant messaging added, high claims attention, service interest, and disbursement not yet claimed." The generation basis refers to the data sources relied upon by the script pre-generation function when calling the large language model to generate scripts in batches, including the standard script library and historical evaluation data returned from the script library. For example, the script pre-generation function uses general script templates in the standard script library and script segments that performed well in previous evaluation feedback as the generation basis to generate new pre-generated scripts for Mr. A.

[0054] Specifically, on a monthly basis, the recordings of the top three performing insurance agents from each institution are obtained. The recordings are then transcribed into text using automatic speech recognition. A large language model is used to extract excellent scripts for handling objections from the text. These scripts are then refined and polished to be more conversational, with the output word count kept under 150 characters. After review by business experts, the excellent scripts are stored in a script library according to the corresponding format of top-performing experience themes and responses.

[0055] The technical solution of this embodiment utilizes an insurance customer attribute tagging system to call a large language model to offline batch generate pre-generated scripts corresponding to the customer profile. During the first call, the pre-generated scripts are matched and pushed according to the sales rhythm stage. For subsequent calls, the scripts for the next call are dynamically generated based on the content of the previous conversation. After the call ends, the recording is transcribed into text for structured scoring and identification of areas for improvement to generate a historical communication summary. At the same time, excellent scripts from high-performing agents' closing calls are periodically extracted and stored in the script library. This solves the problems of traditional standard scripts not being adapted to customer characteristics, resulting in monotonous talking points and poor communication experience; insufficient coverage of manual quality inspection and lag in the identification of personalized improvement directions; and lack of systematic accumulation and distribution of excellent communication experience. It realizes personalized recommendation and dynamic adjustment of insurance scripts, improves the comprehensiveness of call quality assessment and the timeliness of feedback, and promotes the continuous optimization of the team's overall script capabilities.

[0056] In one alternative approach, the insurance customer attribute tagging system includes six dimensions: sales pace, business category, instant messaging addition status, claims attention, service sensitivity, and service sales status. The attribute values ​​of each dimension are combined to form various combinations of insurance customer profiles.

[0057] The sales rhythm refers to nine time phases automatically calculated and divided based on the expiration date of the auto insurance policy, including the first call day, the first call period, 7 days before the expiration date, 3 days before the expiration date, the first day, the first week, the first 2 weeks, the release period, and the last 10 days until the expiration date. For example, Mr. A's auto insurance policy expires on July 1, 2026, so the current sales rhythm is the first day phase. The business category refers to the categories in insurance telephone sales business divided according to customer source and renewal status, including outbound new business and renewal business. For example, Mr. A is an existing customer who purchased insurance from Insurance Company C last year and is renewing his policy this year, so the business category is renewal business. The instant messaging addition status refers to whether the insurance customer has added the insurance agent's instant messaging account (e.g., WeChat account). For example, Mr. A has not yet added Ms. B's WeChat account, so the instant messaging addition status is "not added".

[0058] Among these, claims attention refers to whether the insurance customer actively expresses concern about the claims process, claim amount, or claims speed during communication; for example, Mr. A asked at the beginning, "Is it convenient to make a claim after an accident?", so his claims attention is "high." Service sensitivity refers to the insurance customer's willingness and liking for the additional services provided by the insurance company; for example, Mr. A responded "That's not bad" after hearing about roadside assistance, indicating his interest in the service, so his service sensitivity is "interested in the service." Sales status refers to the insurance customer's progress in receiving and using the marketing activities conducted by the insurance company around additional services; for example, Insurance Company C recently issued free car wash vouchers to renewal customers, and Mr. A has received the notification but has not yet claimed them, so his sales status is "issued but not claimed."

[0059] Among the above-mentioned optional methods, an attribute tag system is further constructed through six dimensions: sales rhythm, business category, instant messaging addition status, claims attention, service sensitivity, and service sales status. The combination of attribute values ​​of each dimension forms a variety of profile combinations, providing a refined customer classification basis for generating differentiated communication scripts and improving the comprehensiveness of profile construction.

[0060] In one alternative approach, the sales rhythm is automatically divided based on the expiration date of the insurance customer's auto insurance policy. The business categories include outbound new business and renewal business. The instant messaging addition status indicates whether the insurance customer has added the agent's instant messaging account. The claims attention level indicates whether the insurance customer shows concern about claims matters. The service sensitivity indicates the insurance customer's level of interest in additional services. The service sales status indicates the insurance customer's progress in receiving and using service marketing activities.

[0061] Among them, the expiration date of the car insurance policy refers to the end date of the current coverage period of the insurance customer's car insurance policy. The sales rhythm stage is automatically calculated based on this date. For example, Mr. A's car insurance policy expiration date is July 1, 2026. Therefore, July 1, 2026 is designated as the "first day" stage.

[0062] Outbound new customer service refers to sales calls made by agents to potential customers who have never purchased car insurance from the same insurance company, with the aim of developing new customers. For example, Ms. Yi calls Mr. Wu, a new car owner who has not previously purchased insurance from Insurance Company C; this call constitutes outbound new customer service. Renewal service refers to sales calls made by agents to existing customers who have previously purchased car insurance from the same insurance company and whose policies are about to expire or have already expired, with the aim of maintaining their existing policies. For example, Mr. Jia purchased car insurance from Insurance Company C last year and is contacting them again this year to discuss renewal; this is considered renewal service.

[0063] Instant messaging accounts refer to third-party application accounts used for real-time text or voice communication, through which agents and customers communicate via non-telephone channels. For example, Ms. B wants to add Mr. A's WeChat account to send him a link to a renewal insurance plan. Claims matters refer to specific matters related to claims during the insurance application or claims process, such as reporting an accident, submitting materials, assessing and verifying damages, and making payments. For example, Mr. A's question, "Do I need to pay for the car repairs myself first?" falls under the category of claims matters.

[0064] Additional services refer to non-claims-related services offered by insurance companies to enhance customer experience beyond basic insurance coverage. For example, Mr. A's insurance plan includes free car wash and chauffeur services, which are additional services. The level of interest refers to the customer's level of liking for the additional services, categorized as interested, uninterested, or not mentioned. For example, Mr. A's response to the free car wash service, "That's good," indicates an interested level.

[0065] Among them, service marketing activities refer to customer outreach and activation activities carried out by insurance companies around additional services, including coupon distribution, service card gifting, usage reminders, etc. For example, Insurance Company C pushed an activity to its renewal customers to "receive a free car wash coupon for the whole year". Mr. A has received the activity notification but has not yet clicked to claim it.

[0066] Among the above optional methods, the sales rhythm is further clarified to be automatically divided based on the expiration date of the car insurance policy, the business categories are distinguished as new outbound business and renewal business, and the other dimensions correspond to the specific meanings of instant messaging addition status, claims attention, service sensitivity and service sales status, so that the tag data has a quantifiable source and improves the accuracy and practicality of profile construction.

[0067] In one alternative approach, the pre-generated script includes three parts: an opening statement, a profile-featured script, and a closing action. Specifically, the script pre-generation module 101 is used to call the large language model to extract the key points of the script for each part of the pre-generated script.

[0068] The opening remarks, or "promotional dialogue," are the first part of the conversation, used by the agent to introduce themselves, confirm the customer's identity, and state the purpose of the communication. For example, Ms. B's opening remarks might be: "Hello Mr. A, I'm Ms. B from Insurance Company C. Your car insurance expires today, and I'd like to introduce you to the renewal plan." The profile-based dialogue, the second part, addresses and elaborates on the personalized characteristics of the insurance customer, such as their claims focus and service sensitivity. For example, Ms. B's profile-based dialogue might be: "Mr. A, you're concerned about claims. Our services include a fast claims channel and a free replacement vehicle, which are very practical." The closing action, the third part, guides the insurance customer to take specific actions, such as adding them on instant messaging, viewing the plan, or completing payment. For example, Ms. B's closing action might be: "Please add me on WeChat first, and I'll send you the detailed plan and link. Don't rush to make a decision."

[0069] Among the above-mentioned optional methods, the pre-generated script is further divided into three parts: opening remarks, profile feature script, and closing action. The large language model is used to extract the key points of the script for each part, making the script content clear and the structure standardized. This makes it easier for agents to quickly grasp the key points of communication before the call, and improves the convenience of script generation and use.

[0070] In one alternative embodiment, the script pre-generation module 101 is specifically used for: When the large language model is invoked to generate the pre-generated script, the corresponding script is matched as the opening line based on the insurance customer's sales status and instant messaging addition status. The corresponding script is matched as the profile feature script based on the insurance customer's claims attention and service sensitivity. After the matching is completed, the opening line, the profile feature script and the closing action are seamlessly connected in a conversational style.

[0071] Specifically, when calling the large language model to generate pre-generated scripts, the large language model matches corresponding scripts from the script library as opening remarks based on the insurance customer's sales status and instant messaging addition status. It also matches corresponding scripts from the script library as profile feature scripts based on the insurance customer's claims attention and service sensitivity. After matching, the large language model adjusts the opening remarks, profile feature scripts, and closing actions to colloquial language to achieve seamless connection between the three parts.

[0072] In the above-mentioned optional methods, when generating pre-generated scripts, the opening remarks are matched based on the sales status and instant messaging status, and the profile-featured scripts are matched based on the claims attention and service sensitivity. The three parts are then seamlessly connected in a conversational style, so that the generated scripts are in line with the actual attributes of the customers and are expressed naturally and fluently, thus improving the adaptability of the script recommendations.

[0073] In an alternative embodiment, the speech matching and recommendation module 102 is specifically used for: The current sales rhythm stage of the insurance customer is calculated based on the expiration date of the insurance customer's auto insurance policy, and the number of calls made by the insurance customer within the same sales rhythm stage is obtained as the number of calls. When the insurance customer enters the next sales rhythm stage, the number of calls is reset to the first time.

[0074] Specifically, the script matching and recommendation module 102 calculates the current sales rhythm stage of the insurance customer based on the expiration date of the insurance customer's auto insurance policy, and accumulates the number of calls made by the insurance customer within the same sales rhythm stage, using the accumulated number of calls as the call count; the script matching and recommendation module 102 monitors whether the insurance customer has entered the next sales rhythm stage, and if so, resets the call count to the first time.

[0075] Among the above optional methods, the current sales rhythm stage of the insurance customer is further calculated based on the expiration date of the car insurance policy. The number of calls made within the same stage is taken as the number of calls, and the number is reset to the first time when entering the next stage. This makes the script recommendation strategy accurately aligned with the policy sales cycle and realizes the automatic judgment of the switching between the first and non-first call scripts.

[0076] In an alternative embodiment, the script adoption evaluation module 104 is specifically used for: After receiving the preprocessed call text, semantic matching evaluation is performed on the opening remarks, profile feature-based dialogue, and closing action sections of the actual dialogue, comparing them with the pre-generated dialogue or the key points of the dialogue. Each part is given a structured score, which is then weighted to obtain an overall communication quality score. Based on the overall communication quality score, the areas for improvement are identified, and the insurance agent's review is generated.

[0077] Semantic matching evaluation refers to the process by which the large language model compares the actual script of the insurance agent with the pre-generated script or key points of the script to determine whether the actual script covers the core points. For example, the semantic matching evaluation compares Ms. Yi's actual script "Do you have any questions about the claims?" with the key points of the script "Fast claims plus a replacement vehicle", and finds that "replace vehicle" is missing, so the profile feature script score is low.

[0078] The overall communication quality score refers to the total score obtained by comprehensively weighting the structured scores of the opening remarks, profile descriptions, and closing actions. The output range is from 30 to 85 points. For example, Ms. Yi's three scores are 75, 68, and 55 points, respectively. After weighting, the overall communication quality score is 65 points.

[0079] Specifically, the script adoption evaluation module 104 receives the pre-processed call text, extracts the actual scripts used by the insurance agents from the pre-processed call text, and breaks down the actual scripts into an opening statement, a profile feature script, and a closing action script. Semantic matching evaluation is performed against the pre-generated scripts or script key points to obtain structured scores for the opening statement, profile feature script, and closing action script. The structured scores of the three parts are comprehensively weighted to obtain an overall communication quality score. Improvement points are identified based on the overall communication quality score, and insurance agent reviews are generated based on the structured scores of the three parts and the overall communication quality score.

[0080] In the above-mentioned optional methods, the actual script is further divided into three parts: opening remarks, profile feature scripts, and closing actions. The scripts are then compared with pre-generated scripts or key points of the scripts for semantic matching evaluation. After obtaining the structured scores of each part, the overall communication quality score is obtained by comprehensive weighting. Based on this, areas for improvement are identified and agent comments are generated, making the call quality evaluation more granular.

[0081] like Figure 2 As shown, the script pre-generation module 101 and the high-performing experience extraction module 105 run as offline batch tasks, merging the generated script data into the script library. The script matching and recommendation module 102 reads scripts from the script library and pushes them to the telemarketing system agent terminals. After the call ends, the call recording is processed by the ASR transcription module 103 and then enters the script adoption evaluation module 104 through asynchronous task scheduling. The evaluation results are fed back to the script library to form a closed-loop optimization.

[0082] The pre-generation module 101 for insurance scripts employs offline batch processing, generating the script and key points before the call takes place. This module first constructs an attribute tagging system for insurance customers, including six dimensions: sales pace, business category, instant messaging addition status, claims attention, service sensitivity, and sales service status. The sales pace comprises nine stages: first call day, first call period, 7 days before the first call, 3 days before the first call, first day, first week, first 2 weeks, release period, and the last 10 days until the expiry date. Business categories include new outbound business and renewal business. Instant messaging addition status includes yes and no. Claims attention includes high claims attention and not mentioned. Service sensitivity includes service interest, service disinterest, and not mentioned. Sales service status includes not issued, issued but not claimed, claimed but not used, and used. The attribute values ​​of the six dimensions are combined using a Cartesian product, resulting in 9 × 2 × 2 × 2 × 3 × 4 = 3240 possible customer profile combinations.

[0083] The pre-generated script module 101 organizes each recommended script into a three-part structure: opening remarks, profile-specific scripts, and closing actions. Each part must be seamlessly connected in a conversational style. Utilizing a large language model, based on a standard script library provided by the business department and insurance customer attribute tags, it generates corresponding personalized scripts for each profile combination. The large model employs role-based positioning engineering, positioning the model as a telephone sales trainer. The generation process follows these rules: the opening remarks are matched to the corresponding script based on the insurance customer's service and sales status and instant messaging addition status; if no match is found, a generic script is used. The profile-specific scripts are matched to the insurance customer's claims focus and service sensitivity, combined and refined into a conversational expression. The closing actions directly use generic scripts. Compliance constraints are applied throughout the entire process; the generated scripts do not contain sensitive information such as customer names, ID numbers, or license plate numbers, and do not use prohibited expressions such as "full coverage," "100% payout," "cashback," or "red envelopes."

[0084] The pre-generation module 101 extracts core points from the three-part dialogue scripts generated using a large language model, focusing on the opening remarks, profile-based dialogue, and closing actions. Each key point is limited to 30 characters. A total of 3240 × 3 = 9720 dialogue key points are generated. All generated dialogue combinations and key points are stored in a dialogue database, indexed by tags, supporting efficient conditional filtering and matching retrieval.

[0085] The script matching and recommendation module 102 provides agents with the most suitable scripts in actual dialing scenarios. Its core logic is an adaptive switching mechanism based on the sales rhythm state, with the key judgment criterion being the number of calls within the same sales rhythm stage. For example... Figure 3 As shown, after an agent initiates a call, the module automatically calculates the current sales rhythm stage based on the insurance customer's car insurance policy expiration date and obtains the number of calls the insurance customer has made within the same sales rhythm stage as the call count. The module determines whether this is the first call within the rhythm stage. If it is the first call, it filters the script library based on the insurance customer's profile combination, accurately matches the corresponding pre-generated standard script, and pushes it to the agent's terminal. If it is not the first call, it obtains the dialogue content of the previous call, calls the large language model to generate the next call script in real time based on the actual content of the previous call, and pushes it to the agent's terminal. The next call script is generated in real time based on the customer's real feedback and includes three parts: opening remarks, profile-featured script, and closing action. When the insurance customer enters the next sales rhythm stage, the call count is reset to the first call, and the first call of the new rhythm reuses the pre-generated standard script corresponding to the new stage.

[0086] The script adoption evaluation module 104 runs in near real-time after the call ends, analyzing the automatically transcribed text based on a large language model. For example... Figure 4As shown, after the call ends, the ASR transcription module 103 is automatically triggered to transcribe the call recording into text. The text is then preprocessed, including removing interjections, correcting common typos, and adjusting word order, resulting in a preprocessed text. The preprocessed text is then input into the large language model evaluation engine.

[0087] The script adoption evaluation module 104 executes three sub-tasks. Sub-task one involves scoring the script in three parts: opening remarks, scripts highlighting specific characteristics, and closing actions. The script is evaluated semantically against standard scripts or key points. The scoring mechanism employs a strict principle, with a default starting score of 60 points. Full compliance earns a score above 90, while missing key points deducts below 50. Specific score ranges are as follows: 91-100 points indicate word-for-word or only interjection differences, with all key points complete and in the correct order. 81-90 points indicate all core points are complete, with slight variations in expression but no omissions. 71-80 points indicate core points are complete but one expression is incorrect or the order is incorrect. 61-70 points indicate one missing core point or numerous expression problems. 51-60 points indicate two missing core points or overall disorganized expression. 41-50 points indicate only a few key points are hit, with most missing. A score of 31 to 40 indicates that only the keywords are matched, while the content is off-topic. A score of 0 to 30 indicates that the content is basically irrelevant or completely missing. In addition, there are 3 bonus points: 5 to 10 points for the presentation of the event script, 10 to 15 points for payment guidance in the closing action, and 5 to 10 points for self-introduction and mention in instant messaging in the opening remarks.

[0088] Subtask two involves overall communication quality assessment and identification of areas for improvement. An overall communication quality score will be generated by combining the scores from the three parts, aiming for a score between 30 and 85. For aspects scoring below 50, the most typical area for improvement will be identified, including offensive language, logical inconsistencies, errors in key information, excessively long or continuous speech exceeding 15 seconds, and ignoring explicit customer objections. Improvement suggestions will be provided in a gentle manner, with a word count of no more than 50 words.

[0089] Subtask three involves agent feedback. For each script type, provide feedback of no more than 20 words based on the score. For scores of 70 and above, focus on praise; for scores between 50 and 70, praise first and then point out areas for improvement; for scores below 50, point out areas for improvement.

[0090] The script adoption evaluation module 104 generates a structured summary of each call based on structured scoring. After each call, a structured call summary is automatically generated, including: opening remarks score and reason analysis, profile-specific script score and reason analysis, closing action script score and reason analysis, overall communication quality score and reason analysis, a list of areas for improvement, and this review. Agents and managers can visually view the communication quality profile of each call on the front end.

[0091] The script adoption evaluation module 104 performs cross-call fusion and summarization of the content of all calls to the same insurance customer, generating a historical communication summary. The historical communication summary includes the agent's communication rhythm (i.e., which services were introduced, which products were introduced, which customer objections were answered, whether the agent proactively added the customer's instant messaging, whether non-motorized products were proactively recommended, whether the agent proactively scheduled appointments with the customer, and whether a quote was provided). It also includes the customer's communication rhythm (i.e., what aspects were proactively inquired about, what objections were raised, and whether the agent's attitude was indifferent). Multiple communication contents from the same sales serial number are automatically fused and summarized.

[0092] like Figure 5 As shown, the six-dimensional attribute tag system includes nine values ​​for sales rhythm, two values ​​for business category, two values ​​for instant messaging addition status, two values ​​for claims attention, three values ​​for service sensitivity, and four values ​​for service-sales status. These are combined using a Cartesian product to form 3240 customer profiles. Each profile corresponds to a three-part script, and after extracting the key points of each part, a total of 9720 key points are generated and stored in the script library. When an agent makes a call, the corresponding script is accurately matched from the script library based on the real-time six-dimensional tag combination of the insurance customer.

[0093] The next call script generation module is invoked when a next call script needs to be generated. Unlike the offline batch generation method of the script pre-generation module 101, it adopts an online real-time generation method. The next call script generation module obtains the automatically transcribed text of the previous call via speech recognition. Using a large language model, it extracts the reasons for refusal, price sensitivity, and objection types expressed by the customer in the previous call. It then integrates the extracted customer feedback information with the corresponding script templates in the standard script library to generate a three-part next call script that continues the content of the previous call. The opening of the next call script includes statements that follow up on the customer's feedback in the previous call and the purpose of this call. The objection handling section includes targeted responses to the customer's previous objections. The closing action section includes low-threshold action guidance statements based on the customer's previous communication status. When a second or subsequent call is required within the same sales rhythm phase, the large language model is invoked, taking the customer's actual state expressed in the previous call as a starting point, combining it with the core information of the standard script, and generating a three-part script for the next communication according to the following conversion rules. The opening remarks should build upon the previous conversation, naturally reinforcing the reason for further communication and outlining the purpose of this interaction. Objection handling should acknowledge the customer's feelings and provide a substantive response based on standard script content. Closing actions should include easy-to-implement action suggestions, seamlessly connecting to instant messaging or payment guidance. The generation rules prohibit directly copying standard scripts without addressing customer pain points, ignoring previous objections, and using prohibited terms such as "extra," "cashback," or "red envelope." Output should use rich text format, with key content highlighted in color for easy identification by agents on the terminal.

[0094] like Figure 6 As shown, the top-performing experience extraction module 105 is an independent functional module that operates on a monthly cycle. First, the extraction theme is defined, selecting common difficulties in handling objections from insurance agents under different business models. This includes price and personal favor objections in outbound new business scenarios, and price and service impracticality objections in renewal business scenarios. Next, top-performing call data is filtered, extracting the recordings of the top 3 performing insurance agents from each institution's transaction calls, and obtaining the transaction call text through automatic speech recognition. Then, a large language model is used for experience extraction and polishing. This involves using a large language model to extract experience, combining it with reference script styles for the top-performing experience theme, and supplementing and polishing the input excellent script content with colloquial language. The output is controlled to around 150 words, while also filtering for compliant vocabulary and desensitizing prices. Finally, business experts conduct compliance review. After the generated content is reviewed by business experts, it is stored in the script library according to the corresponding format of the top-performing experience theme and excellent script responses. This is made available to all insurance agents on the front end as a reference script library for handling specific objection scenarios.

[0095] In another embodiment of the insurance sales script recommendation and intelligent evaluation system of the present invention, it specifically includes: The pre-generated script module 101 constructs an attribute tag system for insurance customers. The attribute tag system includes six dimensions: sales pace, business category, instant messaging addition status, claims attention, service sensitivity, and service sales status. The attribute values ​​of each dimension are combined to form various profile combinations. For each profile combination, a large language model is invoked, and based on the script content in the standard script library, the corresponding three-part pre-generated script is generated offline in batches. The pre-generated script includes an opening, a profile feature script, and a closing action. The core points are extracted from each pre-generated script as script key points, and the pre-generated script and script key points are stored in the script library.

[0096] The script matching and recommendation module 102 obtains the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the call is the first time, it matches the corresponding pre-generated script from the script library according to the insurance customer's profile and pushes the pre-generated script to the insurance agent terminal. If the call is not the first time, it obtains the dialogue content of the previous call, calls the large language model to generate the next script based on the dialogue content of the previous call, and pushes the next script to the insurance agent terminal.

[0097] After the call ends, the ASR transcription module 103 triggers automatic speech recognition and transcription to transcribe the call recording into a text. It then performs preprocessing operations on the text, including removing interjections, removing repeated words, correcting common typos, and adjusting word order, to obtain the preprocessed text.

[0098] The script adoption evaluation module 104 calls a large language model to perform structured scoring on the actual scripts used by insurance agents in the preprocessed call text. The structured scoring includes semantic matching evaluation of the opening remarks, profile-featured scripts, and closing actions of the actual scripts, compared with pre-generated scripts or key points, to obtain the structured scores for each of the three parts. The structured scores of the three parts are then weighted to obtain a weighted total score. Customer emotional features and agent expression features are extracted from the call text. Customer emotional features include the quantitative value of the customer's positive emotional state during the call, the number of times the customer proactively asks questions, and the number of times the customer expresses refusal. Agent expression features include the uniformity of the agent's speaking speed and the frequency of alternation between the agent and the customer's speech rounds. A comprehensive adjustment coefficient is calculated based on the customer emotional features and agent expression features, using the following formula: in, This represents the comprehensive adjustment coefficient. This represents a quantitative value indicating the level of customer emotional positivity. This indicates the frequency of the conversation between the agent and the customer. This indicates the current sales phase number for the insurance customer. Indicates the total number of sales phases and , Indicates the number of calls. This indicates that the customer refuses to specify the number of occurrences. This indicates the uniformity of the speaking speed among the agents. and It is the saturation constant. to For preset weighting coefficients, The baseline value for each round is used. A comprehensive adjustment coefficient is used to non-linearly correct the weighted total score, resulting in an overall communication quality score. The correction method involves truncating the sum of the weighted total score and the adjustment range multiplied by the comprehensive adjustment coefficient, with a truncation range of 30 to 85 points. Improvement points are identified based on the overall communication quality score. Insurance agent reviews are generated based on the structured scores from the three parts and the overall communication quality score. A structured summary of each call is generated based on the structured scores. The structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output.

[0099] The script adoption evaluation module 104 feeds back the evaluation results to the script library. The three parts of the evaluation results—structured score, points for improvement, and insurance agent comments—serve as the basis for correction when the script pre-generation module 101 generates scripts offline in batches.

[0100] The top-performing experience extraction module 105 periodically acquires the transaction call recordings of the top three performing insurance agents from each institution on a monthly basis. It then uses automatic speech recognition to transcribe the transaction call recordings into transaction call text. It calls a large language model to extract excellent scripts for objection scenarios from the transaction call text, refines and polishes these scripts to make them more conversational, and keeps the output word count within 150 words. After review by business experts, the excellent scripts are stored in the script library according to the corresponding format of top-performing experience themes and excellent script responses.

[0101] The technical solution of this embodiment constructs a multi-dimensional attribute tag system to form a profile combination and calls a large language model to offline batch generate pre-generated scripts and key points of the scripts. Based on the sales rhythm stage and the number of calls, the pre-generated scripts are matched during the first call, and the next script is dynamically generated for subsequent calls. After the call, the transcribed text is preprocessed, and the large language model is used to perform semantic matching and evaluation from three parts against the key points of the scripts. This evaluation utilizes factors such as the customer's emotional positivity, the frequency of speech rounds, the sales rhythm stage, the number of calls, the frequency of rejection statements, and the uniformity of the agent's speaking speed. The overall communication quality score is obtained by nonlinearly correcting the weighted total score with a comprehensive adjustment coefficient, identifying areas for improvement and generating agent reviews. Simultaneously, a single call summary is generated based on the structured scoring, and historical communication summaries are generated by merging all summaries. Furthermore, the text of successful calls by top-performing agents is periodically retrieved, and excellent scripts are extracted and stored in the script database using a large language model. This solves the problems of traditional standard scripts lacking differentiated adaptation, insufficient coverage of manual sampling quality inspection and lagging personalized improvement directions, and lack of systematic extraction and accumulation of excellent experience, thus improving the adaptability of script recommendations and the comprehensiveness of call quality assessment.

[0102] Figure 7 This diagram illustrates a flowchart of an embodiment of an insurance sales script recommendation and intelligent evaluation method provided by the present invention. This method employs the insurance sales script recommendation and intelligent evaluation system provided by the present invention. Figure 7 As shown, the method includes the following steps: S1. Based on the attribute tag system and standard script library of insurance customers, call the large language model to generate pre-generated scripts and key points of the scripts corresponding to the profiles of each insurance customer offline in batches and store them in the script library. S2. Obtain the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the number of calls is the first time, match the pre-generated script corresponding to the profile combination of the insurance customer from the script library and push it to the insurance agent terminal. If the number of calls is not the first time, call the big language model to generate the next script based on the dialogue content of the previous call and push it to the insurance agent terminal. S3. After the call ends, transcribe the call recording into a text file and preprocess the text file to obtain a preprocessed text file. S4. Call the large language model to perform structured scoring on the actual dialogue used by the insurance agent in the preprocessed call text, identify areas for improvement, and generate insurance agent reviews. Generate a structured summary of a single call based on the structured scoring, and merge the structured summaries of all calls to the same insurance customer to generate a historical communication summary and output the evaluation results. S5. Periodically obtain the transaction call texts of top-performing insurance agents, call the large language model to extract excellent scripts from the transaction call texts, and store them in the script library after review.

[0103] It should be noted that the beneficial effects of the insurance sales script recommendation and intelligent evaluation method provided in the above embodiments are the same as those of the insurance sales script recommendation and intelligent evaluation system 100 described above, and will not be repeated here. Furthermore, the method and system embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the system embodiments, which will not be repeated here.

[0104] The insurance sales script recommendation and intelligent evaluation system 100 of the present invention can be a computer program (including program code) running on a computer device. For example, the insurance sales script recommendation and intelligent evaluation system 100 of the present invention is an application software that can be used to execute the corresponding steps in the insurance sales script recommendation and intelligent evaluation method of the present invention.

[0105] In some embodiments, the insurance sales pitch recommendation and intelligent evaluation system 100 of the present invention can be implemented in a combination of hardware and software. As an example, the insurance sales pitch recommendation and intelligent evaluation system 100 of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the insurance sales pitch recommendation and intelligent evaluation method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0106] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0107] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned insurance sales pitch recommendation and intelligent assessment methods. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the insurance sales pitch recommendation and intelligent assessment method shown in any embodiment of the present invention by calling the computer program.

[0108] In one alternative embodiment, an electronic device is provided, such as Figure 8 As shown, Figure 8 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0109] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0110] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0111] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0112] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0113] Among them, electronic devices can also be terminal devices. A terminal device can be any terminal device that can install applications and access web pages through applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0114] It should be noted that, Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0115] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-mentioned insurance sales pitch recommendation and intelligent assessment methods.

[0116] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0117] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned insurance sales pitch recommendation and intelligent assessment method.

[0118] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0119] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0120] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0121] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0122] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0123] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0124] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0125] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An insurance sales script recommendation and intelligent evaluation system, characterized in that, include: The pre-generated script module is used to call the large language model to generate pre-generated scripts and key points corresponding to the profiles of each insurance customer offline, based on the attribute tag system and standard script library of insurance customers, and store them in the script library. The script matching and recommendation module is used to obtain the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the number of calls is the first time, the pre-generated script corresponding to the profile combination of the insurance customer is matched from the script library and pushed to the insurance agent terminal. If the number of calls is not the first time, the large language model is called to generate the next script based on the dialogue content of the previous call and pushed to the insurance agent terminal. The ASR transcription module is used to transcribe the call recording into call text after the call ends, and to preprocess the call text to obtain the preprocessed call text. The script adoption evaluation module is used to call the large language model to perform structured scoring on the actual scripts used by insurance agents in the preprocessed call text, identify areas for improvement, and generate insurance agent reviews. Based on the structured scoring, a structured summary of a single call is generated, and the structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output. The high-performing experience extraction module is used to periodically obtain the transaction call texts of high-performing insurance agents, call the large language model to extract excellent scripts from the transaction call texts, and store them in the script library after review. The dialogue data stored in the dialogue database is used by the dialogue matching and recommendation module to perform matching and retrieval based on the profile combination, and by the dialogue pre-generation module as the basis for generating dialogues offline in batches.

2. The insurance sales script recommendation and intelligent evaluation system according to claim 1, characterized in that, The insurance customer attribute tagging system includes six dimensions: sales pace, business category, instant messaging addition status, claims attention, service sensitivity, and service sales status. The attribute values ​​of each dimension are combined to form various insurance customer profile combinations.

3. The insurance sales script recommendation and intelligent evaluation system according to claim 2, characterized in that, The sales rhythm is automatically divided according to the expiration date of the insurance customer's auto insurance policy. The business categories include outbound new business and renewal business. The instant messaging addition status indicates whether the insurance customer has added the agent's instant messaging account. The claims attention level indicates whether the insurance customer shows concern about claims matters. The service sensitivity indicates the insurance customer's interest in additional services. The service sales status indicates the insurance customer's progress in receiving and using service marketing activities.

4. The insurance sales script recommendation and intelligent evaluation system according to claim 2, characterized in that, The pre-generated script includes three parts: opening remarks, profile-featured script, and closing action. The script pre-generation module is specifically used to call the large language model to extract the key points of the script for each part of the pre-generated script.

5. The insurance sales script recommendation and intelligent evaluation system according to claim 4, characterized in that, The script pre-generation module is specifically used for: When the large language model is invoked to generate the pre-generated script, the corresponding script is matched as the opening line based on the insurance customer's sales status and instant messaging addition status. The corresponding script is matched as the profile feature script based on the insurance customer's claims attention and service sensitivity. After the matching is completed, the opening line, the profile feature script and the closing action are seamlessly connected in a conversational style.

6. The insurance sales script recommendation and intelligent evaluation system according to claim 1, characterized in that, The script matching and recommendation module is specifically used for: The current sales rhythm stage of the insurance customer is calculated based on the expiration date of the insurance customer's auto insurance policy, and the number of calls made by the insurance customer within the same sales rhythm stage is obtained as the number of calls. When the insurance customer enters the next sales rhythm stage, the number of calls is reset to the first time.

7. The insurance sales script recommendation and intelligent evaluation system according to claim 6, characterized in that, The script adoption evaluation module is specifically used for: After receiving the preprocessed call text, semantic matching evaluation is performed on the opening remarks, profile feature-based dialogue, and closing action sections of the actual dialogue, comparing them with the pre-generated dialogue or the key points of the dialogue. Each part is given a structured score, which is then weighted to obtain an overall communication quality score. Based on the overall communication quality score, the areas for improvement are identified, and the insurance agent's review is generated.

8. A method for recommending and intelligently evaluating insurance sales scripts, characterized in that, The method employs the insurance sales script recommendation and intelligent evaluation system as described in any one of claims 1 to 7, comprising: Based on the insurance customer attribute tag system and standard script library, the large language model is called to generate pre-generated scripts and key points of the scripts corresponding to the profiles of each insurance customer offline in batches and store them in the script library. The system obtains the current sales rhythm stage of the insurance customer and the number of calls within the sales rhythm stage. If the number of calls is the first time, it matches the pre-generated script corresponding to the profile combination of the insurance customer from the script library and pushes it to the insurance agent terminal. If the number of calls is not the first time, it calls the large language model to generate the next script based on the dialogue content of the previous call and pushes it to the insurance agent terminal. After the call ends, the call recording is transcribed into a text file, and the text file is preprocessed to obtain a preprocessed text file. The large language model is invoked to perform structured scoring on the actual dialogue used by the insurance agent in the preprocessed call text, identify areas for improvement, and generate insurance agent reviews. Based on the structured scoring, a structured summary of a single call is generated, and the structured summaries of all calls to the same insurance customer are merged to generate a historical communication summary, and the evaluation results are output. The system periodically retrieves the transaction texts of top-performing insurance agents, calls the large language model to extract excellent scripts from the transaction texts, and stores them in the script library after review.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory, the memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the insurance sales pitch recommendation and intelligent assessment method as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, implements the insurance sales script recommendation and intelligent evaluation method as described in claim 8.

Images