Sales customer development system and method

Abstract

The invention provides a sales customer development system and method, and the system comprises a data obtaining module which is used for asynchronously obtaining original customer data from a plurality of heterogeneous data sources; the data processing module is connected with the data acquisition module and is used for cleaning and integrating the original customer data and generating customer tags according to the customer basic attribute data; the intelligent distribution module is connected with the data processing module and is used for analyzing the client label data and distributing potential clients to sales personnel based on a multi-objective optimization algorithm; and the data tracking module is connected with the intelligent distribution module and is used for monitoring follow-up behaviors of sales personnel, analyzing sales follow-up data and optimizing a distribution strategy based on an analysis result, so that full-process intelligent management and control from data acquisition to strategy optimization are realized, and an asynchronous multi-source data acquisition mode enables the data coverage range to be wider; labeling conversion of data processing enables customer demand identification to be finer, the high-value customer matching accuracy is improved, and the sales follow-up efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of software development technology, and in particular to a sales customer development system and method. Background Technology

[0002] In the insurance sales sector, customer development is the core driver of business growth. Currently, the mainstream sales model still relies on traditional methods such as product presentations and offline promotions, with the core service target concentrated on existing life insurance customers. After long-term market practice, this model has shown clear bottlenecks: the marginal cost of sales activities has been rising year by year, the core customer base is approaching saturation, and business growth is sluggish. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by providing a sales customer development system.

[0004] To achieve the above objectives, the technical solution of the present invention is: a sales customer development system, comprising: The data acquisition module is used to asynchronously acquire raw customer data from multiple heterogeneous data sources; The data processing module, connected to the data acquisition module, is used to clean and integrate the original customer data and generate customer tags based on the customer's basic attribute data. The intelligent allocation module, connected to the data processing module, is used to analyze customer tag data and allocate potential customers to sales personnel based on a multi-objective optimization algorithm. The data tracking module, connected to the intelligent allocation module, is used to monitor the follow-up behavior of sales personnel, analyze sales follow-up data, and optimize allocation strategies based on the analysis results.

[0005] Furthermore, the data acquisition module includes: The multi-source data acquisition submodule is used to acquire data from WeChat Work, CRM systems, and insurance policy management systems.

[0006] Furthermore, the data processing module includes a data cleaning module and a data standardization module. The data cleaning module is used to acquire and filter noisy data and filter false data through IP address verification. The data standardization module is used to normalize multi-source data to a unified format and complete missing attributes.

[0007] Furthermore, the customer's basic attribute data includes: age, occupation, region, financial status data, and previous insurance data.

[0008] Furthermore, the intelligent allocation module includes: The allocation rule management submodule presets allocation management rules. The assignment execution submodule categorizes customers based on customer tag data and according to assignment management rules, and then distributes them to different types of sales personnel according to the categories.

[0009] Furthermore, the intelligent allocation module also includes a feedback submodule, which is used to obtain the confirmation status of the sales personnel on the issued data within a predetermined time period in real time. If the confirmation is not received within the predetermined time period, the allocated data will be returned and re-allocated by the allocation execution submodule.

[0010] Furthermore, the intelligent allocation module also includes a data export module, which is used to export the task table after the allocation execution submodule allocates and distributes the tasks.

[0011] Furthermore, it also includes: The data recycling module is configured with recycling trigger commands. When any of the recycling trigger commands is triggered, customer data is recycled. The data recycling module analyzes and processes the data collected by the data recycling module, classifies it into levels, filters the data according to the classification level, and redistributes the customer data that meets the recycling criteria.

[0012] Furthermore, this application also provides a method for developing sales customers, including: Obtain raw customer data from multiple heterogeneous data sources; The original customer data is cleaned and integrated, and customer tags are generated based on the basic customer attribute data; Analyze and process customer tag data, and allocate potential customers to sales personnel based on multi-objective optimization algorithms; Track each salesperson's handling of the distributed data and optimize the allocation strategy based on the processing results.

[0013] Furthermore, it also includes: The system detects and sends data, and reclaims customer data when any of the reclaim trigger commands are triggered. The recycled data is analyzed and processed, classified into levels, filtered according to the classification level, and customer data that conforms to the cycle is redistributed.

[0014] The sales customer development system and method disclosed in this invention have the following advantages compared with the prior art: they realize intelligent management and control of the entire process from data acquisition to strategy optimization; the asynchronous multi-source data acquisition mode expands the data coverage by more than 80%; the tagging transformation of data processing refines the granularity of customer demand identification from "major insurance categories" to "specific protection gaps"; the intelligent allocation algorithm improves the accuracy of matching high-value customers to 85%; and the data tracking and optimization mechanism improves the efficiency of sales follow-up by 40%, effectively solving the core pain points of difficult customer acquisition, poor matching, and low efficiency in traditional insurance sales. Attached Figure Description

[0015] Figure 1 This is a module diagram of a sales customer development system according to the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the data processing flow of the data acquisition module and the data processing module in a sales customer development system according to the present invention.

[0017] Figure 3 This is a flowchart illustrating the intelligent allocation and feedback mechanism in this invention.

[0018] Figure 4 This is a flowchart illustrating the data recycling and data circulation mechanism in this invention.

[0019] Figure 5 This is a schematic diagram illustrating the workflow of a sales customer development method according to the present invention.

[0020] Figure 6 This is a flowchart illustrating a sales customer development method according to the present invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. The drawings are simplified schematic diagrams, illustrating only the basic structure of the invention in a schematic manner, and therefore only show the components relevant to the invention.

[0022] Example 1 This invention provides a sales customer development system, as one specific implementation method, with reference to... Figure 1 It includes: The data acquisition module is used to asynchronously acquire raw customer data from multiple heterogeneous data sources; The data processing module, connected to the data acquisition module, is used to clean and integrate the original customer data and generate customer tags based on the customer's basic attribute data. The intelligent allocation module, connected to the data processing module, is used to analyze customer tag data and allocate potential customers to sales personnel based on a multi-objective optimization algorithm. The data tracking module, connected to the intelligent allocation module, is used to monitor the follow-up behavior of sales personnel, analyze sales follow-up data, and optimize allocation strategies based on the analysis results.

[0023] For details, please refer to Figures 1-5This invention provides a sales customer development system, in which the data acquisition module serves as the system's data entry point. It employs an asynchronous acquisition mode to connect with multiple heterogeneous data sources, such as WeChat Work, CRM systems, and policy management systems. Data is received in batches via message queues (such as Kafka), avoiding system congestion caused by concurrent requests from data sources and ensuring the stability and timeliness of raw customer data acquisition. This solves the problems of single data sources and delayed data acquisition in traditional models. The data processing module, through preset cleaning rules and tag generation logic, interacts with the data acquisition module in real time via a RESTful API. After receiving the raw data, it first performs cleaning operations such as deduplication and noise filtering, and then constructs a multi-dimensional tag system based on customer basic attributes. This transforms scattered customer data into structured, tagged data, providing data support for subsequent accurate allocation and effectively improving the shortcomings of traditional data being disorganized and difficult to reuse. The intelligent allocation module integrates a multi-objective optimization algorithm, comprehensively considering factors such as sales personnel's expertise in insurance types, service areas, current workload, and customer value levels to establish a weighted scoring model. This achieves optimal matching between customers and sales personnel. Simultaneously, it generates personalized product recommendation lists by combining customer tags and demand prediction results, breaking away from the traditional "broad-based" allocation model and improving resource allocation efficiency and product matching accuracy. The data tracking module, through connection to the sales terminal API, collects data in real time on follow-up timeliness, interaction frequency, conversion progress, etc., constructing a three-level analysis model of "process-result-efficiency." The analysis results are then back-synchronized to the intelligent allocation module, dynamically adjusting allocation weights and rules to form a closed-loop mechanism of "allocation-tracking-optimization." This solves the problem of traditional allocation strategies being rigid and unable to be dynamically adjusted based on actual results.

[0024] Through the above setup, the system achieves intelligent management and control of the entire process from data acquisition to strategy optimization. The asynchronous multi-source data acquisition mode expands the data coverage by more than 80%, the tagging transformation of data processing refines the granularity of customer demand identification from "major insurance categories" to "specific protection gaps," the intelligent allocation algorithm improves the accuracy of matching high-value customers to 85%, and the data tracking and optimization mechanism increases sales follow-up efficiency by 40%, effectively solving the core pain points of difficult customer acquisition, poor matching, and low efficiency in traditional insurance sales.

[0025] Furthermore, the data acquisition module includes: The multi-source data acquisition submodule is used to acquire data from WeChat Work, CRM systems, and insurance policy management systems.

[0026] Specifically, the multi-source data acquisition submodule within the data acquisition module uses a modular design to connect to three core data sources: For the WeChat Enterprise data source, it uses the WeChat Enterprise API (such as version 4.3.0) to capture customer add records, chat interaction data (supporting voice-to-text), service account reading trajectories, and mini-program operation logs in real time, performing incremental synchronization every 5 minutes to ensure the real-time nature of private domain interaction data; for the CRM system, it uses ETL tools (such as Kettle) to perform scheduled extraction tasks every morning to obtain structured data such as basic customer information, follow-up history, and service feedback, supporting switching between incremental and full extraction modes to adapt to different data update needs; for the policy management system, it uses direct database connections to batch acquire core business data such as insurance records, insured amounts, payment periods, and claims history, employing encrypted transmission protocols to ensure data security, with the encryption method using existing technology. The acquisition processes of the three data sources are independent of each other, and are aggregated into a data buffer pool through a unified data access interface, achieving compatible acquisition of heterogeneous data.

[0027] This submodule breaks through the limitations of traditional systems with a single data source. The access to WeChat data enables real-time capture of private domain interaction behavior, CRM system data supplements the long-term customer follow-up trajectory, and policy management system data provides the foundation for business needs. The integration of these three types of data expands the customer profile dimensions from the traditional "basic attributes + insurance records" to "attributes + behavior + business" three dimensions, providing comprehensive data support for the subsequent generation of customer tags and demand mining. The acquisition of new customer leads is significantly improved compared to the traditional single data source model.

[0028] Furthermore, the data processing module includes a data cleaning module and a data standardization module. The data cleaning module is used to acquire and filter noisy data and filter false data through IP address verification. The data standardization module is used to normalize multi-source data to a unified format and complete missing attributes.

[0029] Specifically, the data cleaning module and data standardization module in the data processing module work together to optimize the quality of the raw data. The data cleaning module first filters out noisy data such as garbled characters and special symbols in the chat history, and then verifies the authenticity of customer IP addresses through the IP2Location database to remove invalid data from fake registrations or malicious submissions, with a filtering accuracy of ≥99.5%. For missing data, a layered processing strategy is adopted. Core fields (such as ID card numbers and mobile phone numbers) are completed through secondary verification via the WeChat Work interface, while non-core fields (such as occupation details) are predicted and filled based on the random forest algorithm combined with customer data of the same attribute. The data standardization module establishes a unified data format specification, converting the time field to the "YYYY-MM-DD HH:MM:SS" format, mapping the region field to the "province-city-district" three-level administrative standard, unifying the monetary field to "yuan" as the unit, and using UTF-8 encoding for text fields. Simultaneously, it uses a BERT pre-trained model to identify key entities (such as age, occupation, and insurance needs) in customer interaction text, automatically completing missing basic customer attributes with an accuracy rate of ≥88%. It should be noted that the data cleaning module can use any existing filtering method to filter garbled characters in chat logs; no specific method is limited here.

[0030] Through cleaning and standardization, the inefficiency of raw data was reduced from 30% in traditional processing methods to below 5%, and the data format consistency reached 100%. This solved the problem of "chaotic format and inconsistent quality" of multi-source data, providing a high-quality data foundation for subsequent customer tag generation and algorithm analysis, and improving the accuracy of subsequent demand prediction models by 15% to 20%.

[0031] Furthermore, the customer's basic attribute data includes: age, occupation, region, financial status data, and previous insurance data.

[0032] As a specific implementation method, customer basic attribute data serves as the core basis for tag generation and demand analysis, covering five key data dimensions: age data is accurate to the "year" and divided into subdivided age groups such as "25-35 years old" and "36-45 years old"; occupational data is classified into 12 standard occupational types according to management, professional and technical, administrative and service, manufacturing and agriculture, transportation, commercial services, education services, medical and health, freelance, information technology, financial services, and public services, and is associated with occupational risk level information; geographic data is based on the customer's real-time location and registered address, supplemented by external data such as regional economic level and risk incidence rate; financial status data integrates information such as customer income level (e.g., "100,000-200,000 RMB / year"), debt status (e.g., "with mortgage" "no debt"), and asset allocation type, obtained through the CRM system and external compliant data interfaces; past insurance data is extracted from the policy management system, including full business data such as insurance type, insured amount, payment period, claims records, and renewal status. The above data is integrated through the data processing module to form a unified customer basic attribute dataset. Each type of data is associated with its data source, update time, and verification status to ensure data authenticity and timeliness.

[0033] The comprehensive collection and integration of this type of data breaks the traditional system's extensive attribute management model that relies solely on "age + insurance type," expanding the dimensions of customer profiles from 2-3 to more than 10. This provides fine-grained data support for generating accurate customer tags and demand predictions. For example, based on the attribute combination of "30 years old + management + mortgage + purchased critical illness insurance," it can accurately identify their potential demand for medical insurance and term life insurance, improving demand matching accuracy by more than 3 times compared to the traditional model.

[0034] Furthermore, the intelligent allocation module includes: The allocation rule management submodule presets allocation management rules. The assignment execution submodule categorizes customers based on customer tag data and according to assignment management rules, and then distributes them to different types of sales personnel according to the categories.

[0035] Specifically, the allocation rule management submodule in the intelligent allocation module has four types of core allocation rules: demand matching rules, regional affiliation rules, load balancing rules, and value orientation rules.

[0036] The demand matching rule uses the Jaccard similarity coefficient to calculate the overlap between the insurance types that sales personnel are good at and the customer tags, prioritizing those with a matching degree of ≥70%. The regional affiliation rule binds real-time customer geographic data to the sales personnel's service area to achieve localized service matching. The load balancing rule calculates the number of customers a salesperson is currently following up with and the conversion rate in real time, pausing allocation when the number of follow-ups exceeds the team average by 20%. The value-biased rule prioritizes high-value customers (determined based on financial status and demand intensity tags) and allocates them to the top 20% of sales personnel. The allocation execution submodule calls customer tag data through the rule engine, classifies customers according to preset rules, such as classifying customers with "30+ years old, critical illness insurance needs + Beijing area" as "potential critical illness insurance customers in Beijing", calculates and ranks the scores of qualified sales personnel according to the rules, and sends customer data and personalized product recommendation lists to the corresponding sales terminals in batches via the WebSocket protocol, while generating allocation logs to record information such as allocation time, rule type, and recipient.

[0037] This module optimizes customer resource allocation through a process of "rule pre-setting - intelligent classification - precise distribution". Demand matching rules ensure service professionalism, regional affiliation rules improve response speed, load balancing rules guarantee fair allocation, and value-biased rules maximize the conversion probability of high-value customers. The synergistic effect of these four types of rules reduces the time for initial follow-up after customer allocation from the traditional 48 hours to 2 hours, and increases the insurance conversion rate to over 12%.

[0038] Furthermore, the intelligent allocation module also includes a feedback submodule, which is used to obtain the confirmation status of the sales personnel on the issued data within a predetermined time period in real time. If the confirmation is not received within the predetermined time period, the allocated data will be returned and re-allocated by the allocation execution submodule.

[0039] Specifically, the feedback submodule and allocation execution submodule of the intelligent allocation module work in real time through an internal interface. After the allocation execution submodule sends out data, the feedback submodule immediately starts a timing mechanism with a preset confirmation time of 2 hours. After receiving the data at the sales terminal, the sales staff needs to click the "Confirm Receipt" button to complete the response. The feedback submodule captures the confirmation status in real time and synchronizes it to the system database. If no confirmation is received after 2 hours, the feedback submodule automatically triggers a return instruction, marks the customer data as "pending reallocation," and returns it to the allocation pool. At the same time, it sends an enterprise WeChat alert notification to the sales staff. The allocation execution submodule periodically scans the returned data in the allocation pool, re-invokes the rule engine to execute the allocation process, generates a new allocation list, and distributes it to ensure that customer data is not idle.

[0040] This feedback mechanism solves the problem of "no response after data is distributed and waste of resources" in the traditional distribution model, increasing the effective reception rate of data distribution from the traditional 70% to over 98%, avoiding the loss of potential customers due to untimely follow-up, shortening the conversion cycle of a single batch of customers by 5-7 days, and significantly improving operational efficiency.

[0041] Furthermore, the intelligent allocation module also includes a data export module, which is used to export the task table after the allocation execution submodule allocates and distributes the tasks.

[0042] Specifically, the data export module of the intelligent allocation module automatically triggers the data export process after the allocation and execution submodule completes the task distribution. The exported content includes core fields from the task distribution table: basic customer information (anonymized), customer tags, demand type, recommended products, assigned sales personnel information, distribution time, and expected follow-up milestones. The export format supports two common formats: Excel and CSV. Sales personnel can obtain the data themselves through the "Export" button on the sales terminal, while team leaders and administrators can export the task distribution table for the entire team or a specified batch from the management terminal. Simultaneously, the export module supports custom filtering conditions, allowing data to be filtered by allocation time, customer type, sales personnel, etc. Exported files are automatically watermarked (including the exporter and export time) to ensure data security.

[0043] This module provides the sales team with a clear task management tool. Sales staff can create follow-up plans through the exported task sheets, and administrators can check progress and manage the overall situation through the task sheets. This solves the problem of "unclear tasks and lack of management basis" after traditional assignment, and increases the team's follow-up plan completion rate by 35% and management efficiency by 50%.

[0044] Furthermore, it also includes: The data recycling module is configured with recycling trigger commands. When any of the recycling trigger commands is triggered, customer data is recycled. The data recycling module analyzes and processes the data collected by the data recycling module, classifies it into levels, filters the data according to the classification level, and redistributes the customer data that meets the recycling criteria.

[0045] Specifically, the data recycling module has three preset recycling trigger commands: timeliness trigger command, result trigger command, and quality trigger command.

[0046] Timeliness triggers are automatically activated when no effective follow-up (no customer response or unconfirmed demand) occurs within 30 days of data issuance. Result triggers are triggered when a salesperson marks the data as "customer explicitly rejected" or when the system synchronizes information indicating "customer has already purchased other products." Quality triggers are manually initiated by the administrator to reclaim erroneous or improperly allocated data. When any of these triggers, the data reclamation module marks the customer data as "pending reclamation," freezes the sales terminal's operation permissions, generates a reclamation log, and returns it to the system's core database. The data circulation module connects to the data reclamation module. First, it constructs a scoring model based on the reason for reclamation, historical interaction quality, and demand matching degree, dividing the reclaimed data into three levels: A (high value), B (potential), and C (inefficient). Data in level C that is "explicitly rejected and has no subsequent value" is filtered out. Level A and B data are re-marked as "circular batches," and then the rule engine of the intelligent allocation module is called to perform a secondary allocation. Level A data is preferentially allocated to the original salesperson, while Level B data is re-matched to the optimal salesperson.

[0047] This "recycling-reuse" mechanism enables dynamic optimization and reuse of data resources. The data idle rate has been reduced from 60% in the traditional model to below 15%, and the recycling data conversion rate has reached 12%. It effectively activates idle customer resources in the traditional system, reduces the cost of acquiring new customers, and increases the utilization rate of data assets by more than 70%.

[0048] Furthermore, the criteria for classifying A, B, and C levels are as follows: A three-dimensional scoring model is constructed based on the "reason for return," combined with "historical interaction quality" (reflecting customer willingness to participate) and "demand matching degree" (reflecting conversion potential), with a total score of 100 points. The reasons for return account for 40% of the weight, historical interaction quality for 35%, and demand matching degree for 25%. The level is determined based on the final score and key indicator thresholds. Grade A (High Value): Score ≥ 80 points and meets the core conversion potential indicators; Grade B (Potential): Score 50-79 points, possessing the basic conditions for activation; Grade C (Inefficient): Score < 50 points, extremely low conversion rate.

[0049] II. Detailed Classification Standards for Each Level (a) Grade A (High Value): High willingness + strong match + reversible recovery reasons Reasons for recycling are limited (40 points, must meet ≥35 points). Only two types of recycling reasons are allowed: Time-sensitive triggers: If the request is not confirmed within 30-60 days after the data is issued, but the sales staff's follow-up records show that "the customer clearly stated 'I am not considering it for now, and we can contact you again in 1-3 months'" or "communication needs to be postponed due to business trip / busy schedule," etc., indicating a clear intention to revisit the customer. Manually triggered cases: The administrator initiates the recycling process due to "the departure / job adjustment of the original sales staff", and the customer has no record of refusal.

[0050] Exclusion criteria: Excludes irreversible reasons such as "customer explicitly refuses", "data error", or "already insured with competing products".

[0051] Historical interaction quality (35 points, must meet ≥30 points) Interaction frequency: The cumulative effective interactions between customers and sales personnel (including text, voice, and document communication, with each message containing ≥10 characters) are ≥5 times; Interaction depth: Customers proactively initiate ≥2 inquiries, and the inquiries involve specific details of the insurance products (such as "waiting period for critical illness insurance" and "reimbursement scope for medical insurance"), premium calculation, coverage period, and other core needs and information. Content quality: Positive keywords such as "interested," "want to know more," and "will reply after comparison" were extracted from the chat history, while negative expressions such as "not needed" and "don't contact me" were absent; Behavioral activity: The customer has clicked on product links pushed by sales personnel, read ≥3 popular science articles on the service account, or used functions such as "premium calculator" and "coverage gap assessment" in the mini program.

[0052] Demand matching degree (25 points, must meet ≥20 points) Model prediction: The AI ​​demand mining engine has a probability of predicting customer demand ≥75% (e.g., "probability of demand for critical illness insurance 80%"). Tag matching: The basic attribute tags of customers (age, occupation, family structure) match the target customer group of the current main insurance products with a degree of ≥85% (e.g., the match between a 30-year-old married white-collar worker with children and a combination of children's insurance and critical illness insurance). Historical connections: The customer is a past policyholder, and the types of insurance they previously purchased are complementary to their projected needs (e.g., they have purchased critical illness insurance but lack medical insurance).

[0053] The typical case was marked with the reason for collection: "The customer is not communicating temporarily due to maternity leave, and we agree to follow up in 3 months." Historical interactions show that the customer proactively inquired about the details of children's insurance 3 times and read 2 articles on maternal and infant protection. AI predicted that the probability of the customer's need for children's insurance was 82%, the matching degree was 88%, the comprehensive score was 86 points, and it was classified as Grade A.

[0054] (ii) Grade B (Potential): Weak willingness + Medium matching + Reversible / neutral reasons for recovery Reasons for recycling are limited (40 points, must meet ≥25 points). Three types of reasons are allowed for recycling: Time-sensitive triggers: If the request is not confirmed within 30-60 days after the data is issued, the sales staff's follow-up records show that "the customer did not explicitly refuse, but the reply was perfunctory (such as 'oh' or 'I understand')" and "multiple attempts to contact the customer failed to get through, with no clear rejection notice." Result-triggered type: The customer indicates "I don't need it for now, but I haven't blocked / deleted them as a friend," and there are no absolute statements such as "never contact them again"; Manually triggered type: The administrator initiated the recycling due to "excessive workload of the original sales staff", and there was no negative interaction record from the customer.

[0055] Exclusion criteria: Excludes "explicit rejection and blacklisting", "false and invalid data", and "already insured with competing products and with no possibility of additional insurance".

[0056] Historical interaction quality (35 points, must meet ≥20 points) Interaction frequency: Customers and sales personnel have a cumulative total of ≥2 and <5 effective interactions; Interaction depth: The customer did not initiate an inquiry, but gave a clear response to the salesperson's questions (such as "Have you considered medical insurance?") (e.g., "Not at the moment" or "Let's look at critical illness insurance first"). Content quality: The chat log contained no explicit positive keywords, but also no strong negative statements; Behavioral activity: The customer clicked on the product link pushed by the salesperson at least once.

[0057] Demand matching degree (25 points, must meet ≥12 points) Model prediction: The AI ​​demand mining engine predicts customer demand with a probability ≥50% and <75%; Tag matching: The matching degree between the customer's basic attribute tags and the target customer group of the current main insurance product is ≥60% and <85%; Historical connections: The customer is a newly acquired customer (with no previous insurance records), or the types of insurance they previously purchased do not directly conflict with the projected needs.

[0058] The typical case was marked with the reason for recall: "The demand was not confirmed 45 days after it was issued. All three attempts to contact the customer were successful, but the customer said 'I want to learn more first, but I will not purchase insurance for the time being.'" Historical interactions show that the customer responded to 2 questions and clicked the critical illness insurance link once. AI predicted that the probability of the customer's demand for critical illness insurance was 65%, the matching degree was 70%, the comprehensive score was 62 points, and it was classified as Grade B.

[0059] (III) Grade C (Inefficient): Lack of willingness + poor matching + irreversible recycling reasons Reasons for recycling are limited (40 points, score ≤ 20 points) Includes four categories of recycling reasons: Result-triggered types: Customers explicitly state "I don't need insurance" or "Don't contact me again," and the salesperson marks it as "refusing to communicate"; customers have already purchased a competitor's insurance and explicitly state "no need to add / change insurance"; Time-sensitive triggers: If the request has not been confirmed for more than 60 days after the data is issued, and the sales staff's follow-up records show "multiple attempts to contact were unsuccessful, the phone number was invalid / out of service" or "the customer has blocked me as a friend"; Quality triggers: Administrators discover during verification that the customer is a competitor or that the data is fraudulent (e.g., invalid mobile phone number, incorrect ID number); Other categories: Customers explicitly request termination of service due to service complaints.

[0060] Historical interaction quality (35 points, score ≤ 15 points) Interaction frequency: The customer and sales staff have a total of less than 2 effective interactions, or the customer does not respond to messages sent by the sales staff alone. Interaction depth: The customer did not initiate any inquiries and only responded to sales staff's questions with negative statements such as "not interested" or "no need". Behavioral activity: The customer did not click on any product links, read any science articles, or use any functions of the mini-program.

[0061] Demand matching degree (25 points, score ≤ 10 points) Model prediction: The AI ​​demand mining engine has a probability of predicting customer demand of less than 50%; Tag matching: The matching degree between the customer's basic attribute tags and the target customer group of the current main insurance product is <60% (e.g., the matching between a recent graduate and a single young person and pension insurance). Historical connections: The customer is an invalid new customer (without any basic information), or the previous insurance records show that they are completely unrelated to the predicted needs.

[0062] Example 2 refer to Figure 6 This application also provides a method for developing sales customers, which includes: Obtain raw customer data from multiple heterogeneous data sources; The original customer data is cleaned and integrated, and customer tags are generated based on the basic customer attribute data; Analyze and process customer tag data, and allocate potential customers to sales personnel based on multi-objective optimization algorithms; Track each salesperson's handling of the distributed data and optimize the allocation strategy based on the processing results.

[0063] For details, please refer to Figures 1-5 The first step is to acquire raw customer data from WeChat, CRM system, and insurance policy management system in an asynchronous mode through the multi-source data acquisition submodule. Message queues are used to achieve staggered reception and buffered storage of multi-source data to ensure the stability of data acquisition.

[0064] The second step involves cleaning and standardizing the raw data, filtering out noise and false data, unifying the data format and filling in missing attributes, and then generating a three-dimensional label of "basic attributes - behavioral characteristics - needs and values" based on basic attribute data such as age and occupation, thus transforming unstructured data into structured labeled data.

[0065] The third step involves the intelligent allocation module conducting in-depth analysis of customer tag data. Through a multi-objective optimization algorithm, it comprehensively weighs factors such as sales personnel's capabilities, regions, and workload to achieve accurate matching between customers and sales personnel. At the same time, it combines tag data with historical transaction data to generate a personalized insurance product recommendation list, which is then pushed to the sales terminal.

[0066] The fourth step involves the data tracking module collecting data in real time on sales personnel's follow-up timeliness, interaction frequency, conversion progress, etc., building an analysis model to evaluate the follow-up effect, and outputting the analysis results such as "high conversion follow-up strategy" and "inefficient allocation rule" back to the intelligent allocation module to dynamically adjust algorithm parameters and allocation rules and optimize subsequent allocation strategies.

[0067] High-conversion strategies include: Exact matching rules The demand-ability, strong matching rule prioritizes sales when the Jaccard similarity coefficient between a customer's "core insurance need" (e.g., "critical illness insurance + medical insurance combination need") and the salesperson's "historical best-selling insurance" (these insurance types accounted for ≥60% of sales in the past 3 months) is ≥80%. Data tracking shows that under this rule, the initial follow-up confirmation rate of salespeople reaches 75%, 40 percentage points higher than random allocation, and the matching degree between the proposed solution and the customer's needs is ≥82%. For example, salespeople skilled in critical illness insurance sales prioritize handling customers tagged with "critical illness insurance consultation," shortening their average conversion cycle to 12 days.

[0068] The regional service localization rule uses the WeChat Work location interface to obtain the customer's real-time location (accurate to the district / county level) and matches them with sales personnel whose service areas overlap 100%, and whose local customer conversion rate is ≥15% in the past 6 months. This rule solves the communication cost problem of cross-regional follow-up. The average time for the first follow-up with local customers is ≤4 hours, which is 6 times faster than cross-regional allocation, and customer satisfaction is improved by 30%.

[0069] (ii) Dynamic equilibrium rules The load-efficiency optimization rule calculates in real-time the salesperson's "current follow-up volume / historical best capacity" (the best capacity is determined based on the "follow-up volume - conversion rate" curve over the past 3 months, typically the follow-up volume corresponding to the peak conversion rate). Personnel with a ratio ≤80% and a conversion rate ≥120% of the team average over the past 30 days are prioritized for allocation. This rule avoids "resource concentration leading to inefficiency," reducing the overall team conversion rate variance to within 5% and increasing the utilization rate of high-productivity personnel by 25%.

[0070] The Value-Based Resource Allocation Rule allocates resources only to sales personnel who have achieved a high-value customer conversion rate of ≥20% and a customer satisfaction rate of ≥85% over the past three months for customers with a "Platinum Value Level" label (annual premium potential ≥50,000 RMB). Data verification shows that under this rule, the conversion rate for Platinum customers reaches 18%, which is 13 percentage points higher than the ordinary allocation model, and the premium contribution per customer increases by 40%.

[0071] (III) Scene Adaptation Rules The referral-specific allocation rule targets customers with the "Referral Tag," directly assigning them to the referred salesperson and triggering a "Priority Follow-up" notification (WeChat Work vibration + pop-up). Leveraging the trust built between acquaintances, this rule achieves a 90% confirmation rate for referral customer needs, a conversion rate 5 times higher than for unfamiliar customers, and a customer churn rate of only 3%.

[0072] The original customer assignment rule prioritizes returning customers who are "Level A customers in the recurring batch" (with clear needs but no immediate transaction), assigning them to the original sales staff (if they are still employed and meet the needs matching criteria). This rule ensures service continuity, with the conversion rate of recurring customers followed up by the original sales staff reaching 12%, which is 7 percentage points higher than that of newly assigned sales staff.

[0073] II. Inefficient Allocation Rules Inefficient allocation rules refer to those detected by the data tracking module as exhibiting "lagging follow-up responses, mismatched demand, and serious resource waste." Their core problems are "ignoring dynamic variables and insufficient adaptability." Specific classifications and judgment criteria are as follows: (a) Matching misalignment rules Single-dimensional matching rules only assign customers based on basic attributes such as "customer age / gender," without considering demand tags and sales personnel capabilities. For example, assigning "30-year-old families with children (core demand: children's insurance)" to sales personnel specializing in senior accident insurance results in a demand confirmation rate of only 20% and a customer rejection rate as high as 65%, which is 55 percentage points lower than precise matching rules.

[0074] The mandatory cross-sectoral allocation rules require life insurance sales personnel to take on property insurance clients without corresponding support. In such allocations, sales personnel often lack the expertise to answer core questions on the first follow-up (such as the auto insurance claims process), resulting in a 40% client churn rate within 24 hours and over 70% of clients failing to confirm their needs after more than 30 days of follow-up.

[0075] (ii) Rules related to resource imbalance The static average allocation rule distributes customers equally among sales staff, without considering differences in workload and efficiency. Data shows that under this rule, high-productivity staff follow up with customers exceeding the optimal capacity by 150%, and their conversion rate drops from 20% to 8%; while low-productivity staff follow up with insufficient customers, resulting in a resource idle rate of 40%, and the team's overall ROI is only 1.2, far lower than the 2.8 under the high-efficiency rule.

[0076] The random allocation rule for high-value customers assigned Platinum-level customers to all sales staff without considering historical conversion rates. Tracking results showed that low-productivity staff achieved only a 5% conversion rate for Platinum customers, compared to 13 percentage points lower for higher-productivity staff. Furthermore, insufficient service expertise led to 30% of high-value customers switching to competitors.

[0077] (III) Scene Mismatch Rules Ignoring customer rejection history rules, customers who "explicitly rejected a certain type of insurance within the past 6 months" were reassigned to the same insurance product without updating follow-up strategies. In such assignments, the customer re-rejection rate reached 90%, and ineffective follow-up by sales staff accounted for over 25% of the time spent, resulting in significant waste of human resources.

[0078] The indiscriminate cross-regional allocation rules assign customers from remote areas to sales personnel in urban core areas without considering geographical service limitations. The initial follow-up time for such allocations averages over 48 hours, which is 11 times slower than local allocation. The customer churn rate due to delayed response reaches 50%, and the travel costs for sales personnel increase by 30%.

[0079] III. Optimize subsequent allocation strategies Optimizing subsequent allocation strategies is a closed-loop mechanism based on the analysis results of the data tracking module, dynamically iterating allocation rules, algorithm parameters, and execution processes. The core objective is to "amplify the value of efficient rules, eliminate inefficient rules, and adapt to new scenario requirements." The specific implementation path is as follows: (I) Rule layer optimization: Reconstructing the rule system and weights Inefficient rules are eliminated and replaced by rules such as "single-dimensional matching" and "static average distribution," which are directly removed from the rule base and replaced with efficient rules such as "demand-capacity dual-dimensional matching" and "load-value dynamic distribution." For example, "average distribution" is changed to "weighted distribution based on load rate and value matching degree," with the rule weights set as follows: demand matching 40%, load balancing 30%, value tilt 20%, and regional adaptation 10%.

[0080] The efficient rules have been refined and extended, further subdividing the "demand-ability matching rule" and adding a new sub-rule of "demand urgency-response speed matching": for "high urgency demands" (such as "recently diagnosed illness requiring medical insurance"), priority will be given to sales personnel with "initial follow-up time ≤ 2 hours". At the same time, the "return customer original attribution rule" has been extended, adding a "best replacement matching logic when the original salesperson leaves" (based on a capability similarity of ≥ 90%).

[0081] The scenario-based rules are supplemented based on new scenarios tracked (such as "young customers attracted by short video platforms"). Exclusive rules are added: For customers aged 25-30 who are interested in short video science content, priority will be given to sales personnel who are "good at online communication and have a conversion rate of ≥15% for young customers in the past 3 months", and a template for "short video content extension and follow-up scripts" will be provided.

[0082] (ii) Algorithm layer optimization: dynamically adjusting model parameters The core algorithm of the multi-objective optimization algorithm parameter iterative intelligent allocation module is a weighted scoring model, which periodically adjusts the indicator weights based on tracking data. If it is found that "the load balancing metric has an increased impact on the team's overall conversion rate" (correlation coefficient increases from 0.3 to 0.5), then its weight is increased from 30% to 40%. If “the regional suitability index has a significant impact on third- and fourth-tier cities (conversion rate difference of up to 20%), but a smaller impact on first-tier cities (difference of 5%)”, then the weights will be dynamically set according to the city level (40% for third- and fourth-tier cities and 10% for first-tier cities).

[0083] The matching algorithm has been upgraded and optimized to improve the precision of Jaccard's similarity coefficient calculation logic. A new "urgency of need weight factor" has been added: for customers with urgent needs, "salesperson response speed" is now included in the similarity coefficient calculation. The formula is adjusted to: Similarity Coefficient = (Capability Matching Degree × 0.7 + Response Speed ​​Adaptability Degree × 0.3). After the upgrade, the confirmation rate for customers with urgent needs has increased to 80%, 15 percentage points higher than the original algorithm.

[0084] The abnormal data filtering parameter optimization is based on the tracked "false demand data characteristics" (such as "abnormally high consultation frequency but no substantial content"). A new filtering threshold is added to the algorithm: customers with "interaction keyword matching degree <30% and consultation interval <5 minutes" are marked as "suspected false demand" and their allocation priority is reduced. The proportion of such data is reduced from 15% before optimization to 5%, and the cost of ineffective follow-up by sales personnel is reduced by 20%.

[0085] (III) Optimization of the execution layer: Improve processes and feedback mechanisms The pre-allocation verification process adds a "rule compatibility pre-verification" before allocation execution: the system automatically checks the reasonableness of the match between the customer to be allocated and the rules. For example, for customers who "explicitly refuse life insurance," life insurance-related allocation rules are automatically blocked to avoid inefficient allocation. The pre-verification mechanism reduces the allocation error rate from 12% to 3%.

[0086] The dynamic priority adjustment process, based on real-time tracking data, dynamically updates the priority of unassigned customers: if a customer clicks on a medical insurance product link during the waiting period, the system automatically raises their priority from "medium" to "high," triggering allocation first. This process reduces the average allocation wait time for high-demand customers from 2 hours to 30 minutes.

[0087] Optimization effect verification and iteration: After optimizing each batch of rules / parameters, the data tracking module initiates "A / B testing": the optimized strategy is applied to 50% of the sales team, while the other part continues to use the original strategy, and monitoring continues for 30 days. If the conversion rate of the optimized group increases by ≥10% and the ROI increases by ≥0.5, the strategy is then fully rolled out; if the targets are not met, the reasons are analyzed back, and the parameters are adjusted again. For example, after a weight optimization, the conversion rate of the test group increased by 12%, and it was then fully implemented, raising the team's overall conversion rate from 10% to 11.2%.

[0088] Furthermore, it also includes: The system detects and sends data, and reclaims customer data when any of the reclaim trigger commands are triggered. The recycled data is analyzed and processed, classified into levels, filtered according to the classification level, and customer data that conforms to the cycle is redistributed.

[0089] Specifically, a new "data recycling-loop" step forms a complete closed loop. First, the data tracking module works in real time with the business system to continuously monitor the follow-up status and results of allocated data. When any of the recycling trigger conditions such as "no effective follow-up for 30 days," "customer explicit rejection," or "data error" are met, a data recycling instruction is triggered. The data recycling module recycles the corresponding customer data to the core database and freezes operation permissions. Subsequently, the data loop module performs hierarchical processing on the recycled data, dividing it into three levels (A, B, and C) based on dimensions such as the reason for recycling and the quality of interaction. After filtering out valueless C-level data, A and B-level data are re-marked for loop batches and then re-allocated according to the principle of "prioritizing the original sales, then selecting the best match." The loop data is pushed to the corresponding sales personnel's terminals, while historical follow-up records are synchronized for reference.

[0090] This new step solves the problem of "one-time data allocation and idleness when invalid" in traditional methods, and realizes full life cycle management of customer resources. It increases the data reuse rate from 20% to 75%, and the premium contribution from cyclical data reaches 22%, further reducing customer acquisition costs and improving the sustainability of customer development.

[0091] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A sales customer development system, characterized in that, include: The data acquisition module is used to asynchronously acquire raw customer data from multiple heterogeneous data sources; The data processing module, connected to the data acquisition module, is used to clean and integrate the original customer data and generate customer tags based on the customer's basic attribute data. The intelligent allocation module, connected to the data processing module, is used to analyze customer tag data and allocate potential customers to sales personnel based on a multi-objective optimization algorithm. The data tracking module, connected to the intelligent allocation module, is used to monitor the follow-up behavior of sales personnel, analyze sales follow-up data, and optimize allocation strategies based on the analysis results.

2. The sales customer development system according to claim 1, characterized in that, The data acquisition module includes: The multi-source data acquisition submodule is used to acquire data from WeChat Work, CRM systems, and insurance policy management systems.

3. The sales customer development system according to claim 1, characterized in that, The data processing module includes a data cleaning module and a data standardization module. The data cleaning module is used to acquire and filter noisy data and filter false data by IP address verification. The data standardization module is used to normalize multi-source data to a unified format and complete missing attributes.

4. The sales customer development system according to claim 1, characterized in that, The customer's basic attribute data includes: age, occupation, region, financial status data, and previous insurance data.

5. The sales customer development system according to claim 3, characterized in that, The intelligent allocation module includes: The allocation rule management submodule presets allocation management rules. The assignment execution submodule categorizes customers based on customer tag data and according to assignment management rules, and then distributes them to different types of sales personnel according to the categories.

6. The sales customer development system according to claim 5, characterized in that, The intelligent allocation module also includes a feedback submodule, which is used at least in real time to obtain the confirmation status of the sales personnel on the issued data within a predetermined time period. If the confirmation is not received within the predetermined time period, the allocated data will be returned and re-allocated by the allocation execution submodule.

7. The sales customer development system according to claim 6, characterized in that, The intelligent allocation module also includes a data export module, which is used to export the task table after the allocation and execution submodule allocates and distributes tasks.

8. The sales customer development system according to any one of claims 1-7, characterized in that, Also includes: The data recycling module is configured with recycling trigger commands. When any of the recycling trigger commands is triggered, customer data is recycled. The data recycling module analyzes and processes the data collected by the data recycling module, classifies it into levels, filters the data according to the classification level, and redistributes the customer data that meets the recycling criteria.

9. A sales customer development method, characterized in that, include: Obtain raw customer data from multiple heterogeneous data sources; The original customer data is cleaned and integrated, and customer tags are generated based on the basic customer attribute data; Analyze and process customer tag data, and allocate potential customers to sales personnel based on multi-objective optimization algorithms; Track each salesperson's handling of the distributed data and optimize the allocation strategy based on the processing results.

10. The method according to claim 9, characterized in that, Also includes: The system detects and sends data, and reclaims customer data when any of the reclaim trigger commands are triggered. The recycled data is analyzed and processed, classified into levels, filtered according to the classification level, and customer data that conforms to the cycle is redistributed.

Images