Intelligent Management Methods and Systems for Foreign Trade Customers
By using multi-dimensional deduplication and dynamic adjustment of order-grabbing permissions based on historical attribution trajectories, the problems of inaccurate customer identification and unclear order-grabbing permission control in CRM systems have been solved, enabling precise management and efficient allocation of customer resources and improving the customer management efficiency of foreign trade CRM systems.
Patent Information
- Application Number
- CN202611381387.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-07
- Publication Date
- 2026-11-10
AI Technical Summary
Existing CRM systems rely on a single customer deduplication dimension, resulting in inaccurate customer identification and unclear control over order-grabbing permissions for customers in the public pool. This leads to inefficient customer resource management, duplicate data consuming storage resources, and unreasonable allocation of high-value customers, resulting in wasted manpower and internal conflicts.
By constructing a multi-dimensional search and reset reliability assessment mechanism and dynamically adjusting the order-grabbing permission judgment parameters based on the customer's historical affiliation trajectory, we can achieve accurate identification of customer identity and differentiated management of public sea customers.
It enables accurate identification of customer identities, optimizes the allocation of customer resources, improves the management efficiency of the CRM system, avoids redundant follow-ups and internal conflicts, and increases customer conversion rates.
Smart Images

Figure CN123070960A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of customer relationship management and computer information processing technology, specifically relating to an intelligent management method and system for foreign trade customers. Background Technology
[0002] With the continuous development of global trade, foreign trade enterprises need to constantly acquire new customers and manage them through CRM (Customer Relationship Management) systems. Customer acquisition channels are becoming increasingly diversified, including offline trade shows, B2B platforms (such as Alibaba.com and Made-in-China.com), social media (such as Facebook and LinkedIn), search engine marketing, and customs data procurement. While multi-channel customer acquisition brings enterprises abundant customer resources, it also leads to the possibility that the same customer may enter the system multiple times through different channels, resulting in duplicate data entry and chaotic allocation of customer resources.
[0003] The existing customer deduplication and public domain management solutions for CRM systems have the following main technical defects:
[0004] First, the plagiarism detection dimensions are limited. Existing solutions typically only check for plagiarism using a single field such as customer name or email address. However, customer names can exist in various forms, including English abbreviations, Chinese translations, and spelling variations. Checking only one field can easily lead to missed detections, resulting in low accuracy. Furthermore, existing solutions do not cover plagiarism detection for social media accounts. Social media (such as WhatsApp, Facebook, and LinkedIn) is a crucial channel for international trade communication, and the same customer cannot be effectively identified when accessing the system through different social media accounts.
[0005] Second, the control over order-grabbing authority for clients in the open market is unclear. The existing solution lacks a dynamic control mechanism for the ownership and order-grabbing authority of clients in the open market. It cannot set different order-grabbing thresholds based on the actual value and follow-up difficulty of clients, resulting in a lack of effective differentiation between high-ability and low-ability salespersons when grabbing orders. This may lead to situations where low-ability personnel grab high-quality clients and waste leads. It also lacks the ability to adjust personalized order-grabbing parameters based on the client's historical follow-up situation.
[0006] The limited scope of duplicate checking leads to inaccurate customer identification, which in turn undermines the reliable data foundation for determining the ownership of customers in the open market. Furthermore, the lax control over order-grabbing permissions for open market customers further exacerbates the irrationality of customer resource allocation. These combined defects result in inefficiency in customer resource management within the foreign trade CRM system. Large amounts of duplicate customer data consume storage resources, multiple salespeople repeatedly follow up on the same customer, causing wasted manpower and internal conflicts, and high-value customers experience reduced conversion rates due to irrational allocation.
[0007] Therefore, there is an urgent need for a method that can accurately identify duplicate customers from multiple dimensions and dynamically and intelligently manage the order-grabbing permissions of customers in the open market, in order to improve the accuracy and efficiency of customer management in foreign trade CRM systems.
[0008] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the embodiments of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0009] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.
[0010] This invention provides a method and system for intelligent management of foreign trade customers. It achieves accurate identification of customer identity by constructing a multi-dimensional search and deduplication reliability assessment mechanism, triggers differentiated public sea classification strategies based on confidence level, and dynamically adjusts the order-grabbing permission judgment parameters by utilizing the customer's historical affiliation trajectory. This solves the technical problems of existing CRM systems having a single deduplication dimension and unclear control over public sea customer order-grabbing permissions.
[0011] In some embodiments, a method for intelligent management of foreign trade customers is provided, including the following steps:
[0012] Obtain customer information from multiple dimensions of the customer to be checked for duplicates, perform database duplicate checks on each dimension, identify existing customers that match each dimension, calculate the matching confidence sub-score for each dimension, and weight and fuse the matching confidence sub-scores of each dimension that match the same existing customer, and use the highest weighted fusion score as the comprehensive confidence score.
[0013] The plagiarism detection results are divided into multiple confidence levels based on the comprehensive confidence score, and the customers to be checked are classified into the public customer pool or associated with the existing customer status according to the preset customer classification trigger strategy corresponding to each confidence level.
[0014] In response to the customer to be deduplicated being categorized into the public customer pool, the public customer intelligent classification processing strategy is executed, and a customer allocation operation is performed. The customer allocation operation includes: obtaining the customer's historical affiliation; when the customer is detected to be in an order-grabbing state, adjusting the order-grabbing permission judgment parameters according to the historical affiliation trajectory; based on the adjusted order-grabbing permission judgment parameters, combined with the user ability assessment score and the user-customer matching degree assessment score, determining the target user with order-grabbing qualifications, and allocating the customer to the target user.
[0015] In some embodiments, a foreign trade customer intelligent management system is provided, comprising:
[0016] The deduplication module is used to obtain customer information from multiple dimensions of the customer to be deduplicated, perform database deduplication on each dimension, determine the existing customers matched by each dimension and calculate the matching confidence sub-score for each dimension, and weight and fuse the matching confidence sub-scores of each dimension that match the same existing customer, and use the highest weighted fusion as the comprehensive confidence score.
[0017] The confidence level and classification module is used to divide the deduplication results into multiple confidence levels according to the comprehensive confidence score, and classify the deduplication customers to be deduplicated into the public customer pool or associate the deduplication customers to be deduplicated with the existing customer ownership status according to the preset customer classification trigger strategy corresponding to each confidence level.
[0018] The public pool classification processing module is used to execute the public pool customer intelligent classification processing strategy in response to the customer to be deduplicated being classified into the public pool customer pool;
[0019] The customer allocation module is used to perform a customer allocation operation in response to the customer to be deduplicated being classified into the public customer pool. The customer allocation operation includes:
[0020] The attribution trajectory recording unit is used to obtain the customer's historical attribution trajectory;
[0021] The order-grabbing parameter adjustment unit is used to adjust the order-grabbing permission determination parameters according to the historical attribution trajectory when the customer is detected to have entered the order-grabbing state.
[0022] The target user determination unit is used to determine target users who are qualified to bid for orders based on the adjusted order-grabbing permission judgment parameters, combined with user ability assessment scores and user-customer matching degree assessment scores, and to assign the customers to the target users.
[0023] In some embodiments, an electronic device is provided, comprising: at least one processor; and
[0024] A memory that is communicatively connected to the at least one processor;
[0025] The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the processor to perform the foreign trade customer intelligent management method described in the first aspect.
[0026] The intelligent management method, system, and electronic device for foreign trade customers provided in this disclosure can achieve the following technical effects:
[0027] The intelligent management method and system for foreign trade customers provided in this disclosure have the following technical effects:
[0028] In the customer identification process, customer information from multiple dimensions is obtained and deduplicated separately. The matching results of each dimension are then grouped and merged according to the existing customer IDs and weighted to obtain a comprehensive confidence score. This makes deduplication no longer rely on simple matching of a single field, but rather comprehensively judges the same customer entity based on information from multiple dimensions. This effectively solves the problem of missed detection caused by a single deduplication dimension and the problem of false detection caused by the mixing of different customer dimensions.
[0029] In the subsequent processing of the plagiarism detection results, the comprehensive confidence score is converted into a multi-level confidence level classification. Based on the preset customer classification trigger strategy corresponding to each level, a macro-classification decision is executed—classifying customers into the public customer pool or associating them with the existing customer status. This establishes a causal link between the output of the plagiarism detection stage and the decision of the classification stage. The plagiarism detection result is no longer an isolated judgment endpoint but an intelligent trigger that drives subsequent classification decisions. At the same time, by preset differentiated classification trigger strategies for different confidence levels, the system can directly associate existing customers in high-confidence scenarios to avoid duplicate processing, use plagiarism detection information to guide the classification direction in medium-confidence scenarios, and execute standard classification in low-confidence scenarios. This achieves the probabilistic transmission and differentiated adaptation of plagiarism detection results to classification decisions.
[0030] In the order-grabbing management stage after a customer enters the public customer pool, the intelligent classification and processing strategy for public customers is triggered in response to the customer being categorized into the public customer pool, in order to further determine the specific department affiliation of the customer within the public pool. Based on this, by acquiring and recording the complete historical affiliation trajectory of the customer in the system and dynamically adjusting the order-grabbing permission judgment parameters according to this trajectory, the order-grabbing decision has historical perception capabilities: information such as the number of times the customer has repeatedly released the product, the reasons for the release, and similar successful experiences identified based on the candidate user's historical follow-up records are all included in the adjustment criteria for the order-grabbing threshold. This allows salespeople with different abilities to be given differentiated order-grabbing qualifications when facing customers of different quality and follow-up difficulty, thereby achieving precise matching of customer resources and salesperson capabilities.
[0031] The above general description and the description below are exemplary and explanatory only, and are not intended to limit the embodiments of this disclosure. Attached Figure Description
[0032] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein:
[0033] Figure 1 This is a schematic diagram of the structure of the intelligent management system for foreign trade customers provided in this embodiment of the disclosure;
[0034] Figure 2 This is a flowchart illustrating the intelligent management method for foreign trade customers provided in this embodiment of the disclosure;
[0035] Figure 3 This is a schematic diagram of the intelligent classification and processing of offshore customers provided in an embodiment of this disclosure;
[0036] Figure 4 This is a schematic diagram of the customer allocation process provided in an embodiment of this disclosure;
[0037] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this disclosure. Detailed Implementation
[0038] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.
[0039] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.
[0040] Unless otherwise stated, the term "multiple" means two or more.
[0041] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.
[0042] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.
[0043] The term "correspondence" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.
[0044] It should be noted that the methods and systems described in this disclosure involve the collection and processing of customer information (including but not limited to company name, contact information, email address, social media account, etc.), and are executed after obtaining the user's legal authorization.
[0045] Figure 1 This embodiment of the present disclosure provides an intelligent management system for foreign trade customers. The system includes: a deduplication module, a confidence level classification and categorization module, a high seas categorization processing module, and a customer allocation module.
[0046] The deduplication module retrieves customer information from multiple dimensions for the customer to be deduplicated, performs database deduplication on each dimension, identifies existing customers matching each dimension, and calculates a matching confidence sub-score for each dimension. The matching confidence sub-scores for each dimension matching the same existing customer are weighted and merged, with the highest weighted merge score used as the overall confidence score. This module is deployed in the backend service layer of the CRM system. It receives customer deduplication requests submitted from the frontend page via a RESTful API interface. After parsing and validating the request parameters, it calls the database query interface for each dimension to perform deduplication. Regardless of whether the customer is already in the system database, the deduplication module performs the same database query operation to calculate the matching degree. Finally, it summarizes and calculates the deduplication results for each dimension and passes them to the confidence leveling and classification module.
[0047] The confidence level classification module is used to divide the deduplication results into multiple confidence levels based on the comprehensive confidence score, and classify the customers to be deduplicated into the public customer pool or associate them with the existing customer status according to the preset customer classification trigger strategy corresponding to each confidence level. In this embodiment, there are three confidence levels: a first confidence level, a second confidence level, and a third confidence level. This module receives the comprehensive confidence score from the deduplication module and performs branch processing based on the value of the score: if the comprehensive confidence score is 0 (i.e., no existing customer records are matched in all dimensions), the customer is directly determined to be a new customer, and standard public customer intelligent classification processing is performed (without adding deduplication-related weighting factors); if the comprehensive confidence score is greater than 0, it is compared with the preset first threshold and second threshold to determine the confidence level to which the current deduplication result belongs. When the comprehensive confidence score is higher than the first threshold, it belongs to the first confidence level; when it is between the second threshold and the first threshold (inclusive), it belongs to the second confidence level; and when it is lower than the second threshold, it belongs to the third confidence level. The preset customer classification trigger strategy for each confidence level is as follows: When the customer to be checked is at the first confidence level, it is determined that the customer to be checked and the existing customer are the same customer. There is no need to execute the public classification process. The customer to be checked is directly associated with the belonging status of the existing customer. When the customer is at the second confidence level, the public customer intelligent classification process is executed, and the department of the existing customer matched by the duplicate check is used as a weighting factor in the department matching degree calculation. When the customer is at the third confidence level, the public customer intelligent classification process is executed, and no duplicate check-related weighting factors are added.
[0048] The public customer pool classification module is used to execute an intelligent public customer pool classification strategy in response to the customer to be deduplicated being classified into the public customer pool. Upon receiving a trigger signal from the confidence level and classification module indicating that a customer has been classified into the public customer pool, this module initiates the intelligent public customer pool classification process. This process extracts the customer feature vector of the customer to be classified and the department feature vectors of each department in the system. For each department, it calculates a matching score with the customer to be classified. The matching score includes at least a weighted sum of regional matching score, industry matching score, product line matching score, channel source matching score, and departmental load factor score. If the highest matching score exceeds the classification threshold, the customer is classified into the department's public pool corresponding to that department. If the matching scores of all departments are lower than the classification threshold, the customer is classified into the company's public pool. After completing the departmental affiliation determination, this module transmits the classification result (department public pool or company public pool) to the customer allocation module.
[0049] A customer allocation module is used to perform a customer allocation operation in response to the customer to be deduplicated being categorized into the public customer pool. This module further includes:
[0050] The attribution trajectory recording unit is used to acquire the customer's historical attribution trajectory. This historical attribution trajectory includes each change in the customer's attribution status between the public and private pools, the corresponding triggering event, follow-up results, and follow-up duration. This unit automatically records the status change in the attribution trajectory table of the database each time the customer's attribution status changes, providing a data basis for adjusting order-grabbing parameters. Attribution status includes three types: private pool customer (with a clearly defined owner), departmental public pool customer (belonging to a specific department's public pool), and company public pool customer (belonging to a company-wide public pool).
[0051] The order-grabbing parameter adjustment unit is used to adjust the order-grabbing permission determination parameters based on the customer's historical affiliation trajectory when the customer is detected to have entered an order-grabbing state. This unit monitors customer affiliation status change events. When a customer enters an order-grabbing state (e.g., is released back to the departmental or company public pool), it reads the customer's historical affiliation trajectory from the affiliation trajectory recording unit. Based on information such as the number of releases, reasons for releases, and similar successful experiences in the trajectory, it dynamically adjusts the order-grabbing permission determination parameters. The order-grabbing permission determination parameters include a comprehensive score threshold for determining whether a candidate user is eligible to grab orders and a priority ranking weight for the customer's order-grabbing recommendation. The adjusted parameters are then passed to the target user determination unit.
[0052] The target user determination unit is used to determine target users qualified to bid for orders based on the adjusted order-grabbing permission judgment parameters, combined with user ability assessment scores and user-customer matching degree assessment scores, and then assign the customers to the target users. This unit integrates the adjusted order-grabbing qualification judgment threshold output by the order-grabbing parameter adjustment unit, the ability scores and matching degree scores of each candidate user, filters out users qualified to bid for orders, and determines the final target users, completing the entire process from customer deduplication to customer assignment.
[0053] The foreign trade customer intelligent management system disclosed in this embodiment receives customer information through a deduplication module and performs multi-dimensional deduplication. It then transmits the comprehensive confidence score to a confidence leveling and classification module. This module classifies customers into levels based on the confidence score and performs macro-classification decisions—classifying customers into the public customer pool or associating them with existing customers. In response to a customer being classified into the public customer pool, the public customer classification processing module is triggered to execute a public customer intelligent classification processing strategy to determine the customer's specific department affiliation within the public pool. Subsequently, the customer allocation module is triggered to perform customer allocation operations. The attribution trajectory recording unit, order-grabbing parameter adjustment unit, and target user determination unit collaboratively complete the entire allocation process from attribution trajectory recording and order-grabbing parameter adjustment to target user determination. A causal linkage chain of "deduplication → leveling and classification → public micro-classification → customer allocation" is formed among the modules, jointly realizing complete intelligent management from customer deduplication to customer allocation.
[0054] Example 2
[0055] Corresponding to Embodiment 1 above, Embodiment 2 of the present invention provides a method for intelligent management of foreign trade customers, referring to... Figure 2 As shown, the method includes the following steps:
[0056] S1: Obtain customer information from multiple dimensions of the customer to be checked for duplicates, perform database duplicate checks on each dimension, determine the existing customers matched by each dimension, and calculate the matching confidence sub-score for each dimension. Weightedly fuse the matching confidence sub-scores of each dimension that match the same existing customer, and use the highest weighted fusion score as the comprehensive confidence score.
[0057] When a salesperson enters a new customer information into the CRM system, the system needs to compare the input information with the information of existing customers in the database to determine whether the customer already exists, thus avoiding duplicate entries. The output of the deduplication module is a comprehensive confidence score, which will serve as input data for the subsequent confidence grading module. "Comprehensive confidence score of 0" and "Comprehensive confidence score greater than 0" represent two different deduplication results: "no existing customer matched" and "at least one existing customer matched," respectively. The subsequent S2 will use this score for grading.
[0058] The system's plagiarism detection module receives customer plagiarism requests submitted by users through the front-end interface. Users can input customer information from one or more dimensions in the front-end interface. Specifically, these dimensions include six aspects: company name, company phone number, contact person's phone number, company email address, contact person's email address, and social media accounts. After parsing the input customer information, the plagiarism detection module performs database plagiarism checks on each dimension.
[0059] For company name deduplication, the module performs a fuzzy matching query in the customer basic information table, supporting matching of various name formats such as Chinese and English, abbreviations, and full names. For example, "ABC Trading Co., Ltd" and "ABCTrading LLC" can be identified as similar. The module compares the input company name with the company names of existing customers in the database, calculates the name similarity, and returns a list of customers with high similarity to the input name, along with the corresponding similarity score for each record.
[0060] For deduplication of company phone numbers and contact person phone numbers, the deduplication module matches the corresponding database fields and performs format normalization processing (such as unifying country codes and removing separators) to identify identical phone numbers with different formats as matches. The deduplication module calculates the degree of match between the input phone number and existing phone numbers in the database, assigning a higher sub-score to records that are complete matches.
[0061] For deduplication checks on company email and contact email addresses, the deduplication module performs a precise match after normalization processing (such as removing the tag part in email aliases). It compares the input email address with the email addresses of existing customers in the database, and assigns a higher sub-score when there is a complete match.
[0062] For deduplication of social media accounts, the system first identifies the platform type to which the social media account belongs. For example, when a WhatsApp account and a contact's phone number are associated with the same existing customer through the same phone number, the system only counts the match with the higher similarity in these two dimensions as a sub-score of the match confidence score; the corresponding match result in the other dimension is not scored to avoid duplicate counting. Specifically, if the contact's phone number dimension is an exact match (sub-score = 100) and the WhatsApp account also matches the same customer through that phone number, then only the contact's phone number dimension scores 100 points, and the social media account dimension scores 0 points for that match record; if the contact's phone number dimension does not match but the WhatsApp account does, then the social media dimension scores normally. For Facebook and LinkedIn accounts, the system performs exact and fuzzy matches respectively using the account URL, username, and user ID fields recorded in the social media association table. When the same contact's accounts on multiple social media platforms match the same existing customer, the match results on each platform are counted in the corresponding sub-score of the match confidence score, but deduplication is required to avoid duplicate counting. Unlike traditional customer information dimensions such as company name, phone number, and email, social media account dimensions have the following unique characteristics: First, strong reliance on foreign trade communication scenarios. In foreign trade, social media tools such as WhatsApp, Facebook, and LinkedIn are the main channels for sales staff to communicate with overseas customers in real time. Customers may initiate inquiries multiple times through different social media accounts. Without deduplication of social media accounts, different social media accounts of the same customer will be misidentified as multiple different customers, resulting in scattered customer resources and duplicate follow-ups. Second, platform heterogeneity of account formats. The account formats of different social media platforms vary greatly (e.g., WhatsApp accounts are associated with phone numbers, LinkedIn accounts are associated with custom URLs or email addresses, and Facebook accounts are associated with usernames or IDs). Normalization and matching strategies need to be designed separately for each platform, rather than using a uniform string matching method. Third, the complexity of cross-platform association identification. The same customer may register with different usernames, email addresses, or phone numbers on different social media platforms. The system needs to cross-verify the account information of each social media platform with existing phone number and email dimensions (e.g., WhatsApp accounts are associated with phone numbers, and LinkedIn accounts are associated with email addresses) to form multi-dimensional cross-verification to improve the accuracy of deduplication. By incorporating social media accounts into a multi-dimensional deduplication system, this application can effectively identify duplicate customers that cannot be found through traditional fields but can be linked through social media accounts, filling the gap in this dimension of existing CRM systems.
[0063] After each dimension's plagiarism check is completed, the plagiarism check module calculates the matching confidence sub-score S for that dimension. i(Scores are based on a percentage system, ranging from 0 to 100.) If the input information for a certain dimension is empty or no record is found in the database, the sub-score for that dimension is 0; if it is a complete match, the sub-score is 100; if it is a partial match (such as a fuzzy match with similar but not completely identical records), a sub-score between 0 and 100 is assigned based on the degree of similarity (e.g., 85% similarity is recorded as 85 points, and so on). It should be noted that when checking for duplicates in the database for each dimension, the unique identifier (customer ID) of the existing customer matched must also be recorded. When multiple existing customers are matched for a certain dimension, the system only retains the matching record with the highest confidence sub-score under that dimension and its corresponding customer ID; other matching records are not used. Matching a single information dimension is insufficient to determine that they are the same customer; cross-verification across multiple dimensions is required to achieve a high confidence level.
[0064] After calculating the sub-scores of the matching confidence score for each dimension, the system groups and merges the existing customer IDs matched by each dimension: the sub-scores of the matching confidence score for all dimensions matching the same customer ID are grouped into the same group, and the sub-scores within each group are weighted and summed according to preset weights to obtain the comprehensive confidence score corresponding to that customer ID. If only one customer ID is matched, the comprehensive confidence score corresponding to that customer ID is the final comprehensive confidence score, and that customer is a candidate matched existing customer; if multiple customer IDs are matched, the highest comprehensive confidence score is taken as the final comprehensive confidence score, and the existing customer corresponding to that customer ID is used as a candidate matched existing customer. Sub-scores of dimensions that do not match the candidate matched existing customer (i.e., sub-scores that match other customer IDs) are not included in the calculation of the comprehensive confidence score for the candidate matched existing customer. Through the above merging process, it is ensured that the comprehensive confidence score always corresponds to a specific existing customer entity.
[0065] The overall confidence score is calculated using the following formula:
[0066]
[0067] Among them, S i W represents the matching confidence sub-score for the i-th dimension (1≤i≤6) after customer ID merging. This sub-score is the matching score relative to the existing customers in the aforementioned candidate matches. i Let C be the weight of the i-th dimension, and the sum of the weights of all dimensions is 1. total The value range is 0 to 100. The weights of each dimension are determined as follows: First, a baseline weight W is set for each dimension based on business experience or historical data analysis. base_iThen, the baseline weight is dynamically adjusted based on the information entropy of the dimension field in the system and its historical deduplication accuracy. Specifically, the weight is first adjusted based on the information entropy E of each dimension field in the system's customer database. i Determine the benchmark weight W base_i Higher information entropy indicates higher discriminative power and a greater baseline weight for that dimension field. Simultaneously, the historical deduplication accuracy (A) of each dimension field is periodically calculated. i (The proportion of records deemed a true match by humans among those identified as matches in this dimension), and the final weight W. i = W base_i ×(1 +β·(A i - A avg ), where β is the adjustment coefficient, A avg This represents the average plagiarism detection accuracy across all dimensions. When the accuracy of a particular dimension exceeds the average accuracy, its weight is increased; conversely, it is decreased. The baseline weight W for each dimension is... base_i The initial values are set by the system administrator based on the distinguishability of each information dimension for customer identification in foreign trade business scenarios, as follows: Company name 0.20 (due to name changes, hence the second highest weight), Company phone number 0.15 (due to potential changes in phone numbers), Contact person's phone number 0.15 (due to potential changes in phone numbers), Company email address 0.25 (due to its uniqueness and stability, it is the most reliable dimension for identifying customer identity), Contact person's email address 0.15, Social media account 0.10 (due to the diverse account formats and complex cross-platform associations, its reliability as a sole basis for identification is relatively low), and the sum of the weights of each benchmark is 1. The above benchmark weights can be manually adjusted through the management backend according to actual business scenarios, and the system also dynamically corrects them according to the aforementioned information entropy and deduplication accuracy.
[0068] After the above grouping and merging, if the sub-scores of the matching confidence scores for each dimension within each group corresponding to all customer IDs are 0, then the overall confidence score is 0, indicating that no existing customer record has been matched, and the customer is determined to be a brand new customer; when the sub-scores of the matching confidence scores for at least one dimension within at least one customer ID group are greater than 0, the overall confidence score is greater than 0, indicating that at least one existing customer record has been matched, and this score will be passed to the confidence grading module for confidence level determination.
[0069] Through the aforementioned multi-dimensional weighted fusion mechanism, this step integrates the originally isolated deduplication results of each dimension into a unified confidence score, providing a quantitative basis for subsequent confidence leveling and differentiated processing.
[0070] For example, a salesperson from a foreign trade company enters a new customer information into the CRM system, including the company name "ABCTrading Co., Ltd," the company email address "info@abctrading.com," and the contact phone number "+1-***-*******." They also enter the customer's LinkedIn account "abc-trading." The deduplication module finds a similarity match for the existing customer "ABC Trading LLC" in the company name dimension (sub-score 85); a match for the email address of the same existing customer "ABC Trading LLC" in the company email dimension (sub-score 100); no match in the contact phone number dimension (sub-score 0); and a match in the LinkedIn account of the same existing customer in the social media dimension (sub-score 100). Other dimensions have sub-scores of 0 because no information was entered. Since all the deduplication results point to the same customer ID, the system uses this customer ID as a candidate for matching with existing customers and groups the sub-scores of each dimension into the same weighted fusion. Assuming the company name has a weight of 0.20, the company email has a weight of 0.25, the social media has a weight of 0.20, and the combined weight of other dimensions is 0.35 but the sub-scores are 0, then the overall confidence score C is... total = 0.20×85 + 0.25×100 + 0.20×100 = 62 points.
[0071] In another scenario, a salesperson enters a completely new customer whose company name, phone number, email address, social media accounts, and other information do not match any existing records in the database. In this case, the sub-scores for each dimension are all 0, and the overall confidence score is 0. The system directly classifies the customer as a completely new customer without going through the confidence leveling process. The customer is directly classified into the public customer pool and the standard public customer intelligent classification processing is directly executed (without adding weighting factors related to deduplication).
[0072] S2: Based on the comprehensive confidence score, the deduplication results are divided into multiple confidence levels, and according to the preset customer classification trigger strategy corresponding to each confidence level, the customers to be deduplicated are classified into the public customer pool, or the customers to be deduplicated are associated with the existing customer ownership status.
[0073] The confidence leveling module receives the overall confidence score from the plagiarism detection module and performs branching processing based on the score value:
[0074] If the overall confidence score is 0 (i.e. no existing customer records are matched in any dimension), the customer is directly determined to be a brand new customer. There is no need to enter the confidence level process. The public sea customer intelligent classification processing module directly performs standard public sea customer classification processing (without adding weighting factors related to deduplication).
[0075] If the overall confidence score is greater than 0, it is compared with a preset first threshold and a second threshold to determine the confidence level of the plagiarism detection result. The first threshold is greater than the second threshold, and their initial values can be set by the system administrator based on the similarity distribution statistics in historical plagiarism detection records (e.g., using the 85th percentile as the first threshold and the 55th percentile as the second threshold). In the initial stage where historical data is lacking, empirical values can be used (e.g., setting the first threshold to 85 and the second threshold to 55). When the overall confidence score is higher than the first threshold, it belongs to the first confidence level; when it is between the second threshold and the first threshold (inclusive), it belongs to the second confidence level; when it is lower than the second threshold, it belongs to the third confidence level. The confidence levels of the first, second, and third confidence levels decrease sequentially.
[0076] The specific customer classification trigger strategies for each confidence level are as follows:
[0077] First Confidence Level (High Confidence): The system determines that the customer to be checked for deduplication is almost certainly the same entity as the candidate matching existing customer identified in step S1. At this point, the system does not need to execute the public pool classification process and directly associates the customer to be checked with the existing customer's ownership status. If the existing customer is a private pool customer, the new customer directly enters the private pool and does not enter the public pool; if the existing customer is a departmental or company public pool customer, the new customer follows suit and enters the corresponding public pool. Simultaneously, the system establishes a "strong association" tag for the two customers in the database, recording the association basis and confidence score. The ownership status includes private pool ownership (assigned to a specific user), departmental public pool ownership (belonging to a departmental public pool), and company public pool ownership (belonging to a company-level public pool).
[0078] Second confidence level (medium confidence): The system determines that the customer to be checked for duplicates may be related to the candidate matching customers identified in step S1, but this is not yet certain. At this point, the system performs intelligent classification processing for the public customer pool, but in the department matching degree calculation, the department to which the candidate matching customer belongs is directly used as a weighting factor, and the matching score of that department receives an additional increase Δ = α · C. total Where α is a preset coefficient, C totalTo ensure a higher probability that a customer will be assigned to the same department as an existing customer in the candidate matching process, the final matching score for that department is the sum of the base matching score and the increment. If the sum exceeds 100, it is truncated to 100 to ensure the matching score remains within the preset range of 0-100. By utilizing customer identity clues obtained during the deduplication process, the system guides the public pool classification direction, ensuring that new customers are assigned to the same department as existing customers in the candidate matching process. This avoids customer resource fragmentation, duplicate follow-ups, and internal order collisions caused by different information forms of the same customer being dispersed to different departments due to insufficient search reliability. Considering both the guidance effect and algorithm robustness, the value of α is set to 0.1 < α < 0.6, with an optimal range of 0.2 to 0.5 (inclusive). The system can dynamically adjust this value based on historical data analysis and business feedback. If the value of α is too high (e.g., ≥0.6), the plagiarism detection guidance weight will completely cover other matching factors. If the value is too low (e.g., ≤0.1), the guidance weight will be too small and will not have an effective impact on the department matching score.
[0079] For the third confidence level of the plagiarism check results, the system determines that the customer to be checked is likely a new customer. At this time, the system performs standard intelligent classification processing for public customers, without adding any weighting factors related to plagiarism checks.
[0080] Furthermore, to ensure the confidence level grading mechanism adapts to changes in the business environment, the aforementioned first threshold and second threshold for level determination are not fixed. The system periodically calculates the subsequent order conversion rate for customers at the second and / or third confidence levels, i.e., the proportion of customers at each confidence level who ultimately complete a transaction. If the conversion rate for a certain confidence level is significantly higher than expected, it indicates that the threshold setting for that level may be too strict, and the system automatically lowers the threshold appropriately to allow more customers to enter that level and receive the corresponding classification and processing strategy; conversely, if the conversion rate is significantly lower than expected, it indicates that the threshold setting may be too lenient, and the system automatically raises the threshold appropriately. The subsequent order conversion rate for new customers (with a comprehensive confidence score of 0) is calculated separately and is not included in the adjustment of the first and second thresholds. For example, the conversion rate R for customers at each confidence level is calculated on a monthly basis. j (j=1, 2, 3). Set the target conversion rate T for each level. j (Set by the administrator based on business objectives). Threshold adjustment adopts a feedback-based rule-driven correction mechanism. When R... j >T j When the value is × 1.2, the conversion rate of this level is determined to be significantly higher than expected, and the corresponding level determination threshold is adjusted downwards by Δ = min(3, (R j - T j () × 10) points; when R j < Tj When the value is × 0.8, the conversion rate of this level is determined to be significantly lower than expected, and the corresponding level determination threshold is adjusted upwards by Δ = min(3, (T j - R j () × 10) points; the single adjustment range shall not exceed 3 points, and the cumulative adjustment range shall not exceed ±20% of the initial threshold.
[0081] Through the aforementioned confidence level-based triggering mechanism and threshold dynamic adjustment mechanism, this step implements a differentiated public sea classification strategy: direct association to avoid duplicate processing under high confidence, using deduplication information to guide classification direction under medium confidence, and executing standard classification under low confidence. This tiered processing method ensures processing efficiency in high-confidence scenarios, fully utilizes the attribution information contained in the deduplication results in medium-confidence scenarios, and ensures that the tiering standards always match the actual business results through automatic threshold adjustment.
[0082] S3: In response to the customer to be checked for duplicates being classified into the public customer pool, execute the public customer intelligent classification processing strategy and perform customer allocation operation.
[0083] When the macro-classification decision in step S2 classifies the customer to be deduplicated into the public customer pool, the system responds to this trigger signal and initiates the public customer intelligent classification processing strategy. In this embodiment, the public customer intelligent classification processing is implemented through public customer intelligent classification rules, which determine the department affiliation of customers who need to enter the public customer pool. Specifically, it applies to the following three scenarios: brand new customers with a comprehensive confidence score of 0, customers belonging to the second confidence level, and customers belonging to the third confidence level. (Refer to...) Figure 3 As shown, the rule includes the following:
[0084] S31. Extract the customer feature vector of the customer to be classified and the department feature vector of each department in the system. The customer feature vector includes the source channel, inquiry quality score, customer value level, industry, and region. The department feature vector includes the department's business scope, responsible area, responsible product line, and current customer load.
[0085] S32. Calculate the matching score between each department and the customer to be classified. The matching score includes at least the weighted sum of the regional matching score, industry matching score, product line matching score, channel source matching score, and departmental load factor score. If the highest matching score exceeds the classification judgment threshold, the customer is classified into the department's public pool corresponding to that department. If the matching scores of all departments are lower than the classification judgment threshold, the customer is classified into the company's public pool.
[0086] During the feature extraction phase, the system acquires the source channels of customers to be categorized (such as trade shows, B2B platforms, social media, etc.), inquiry quality scores (assessed based on the detail and professionalism of the customer's initial inquiry, normalized to a score range of 0-100), customer value levels (predicted by the system based on customer size, budget, and other information, normalized to a score range of 0-100), industry, and country / region. Simultaneously, the system acquires information on all departments: departmental business scope (the main business types of the department), responsible region (the geographical area the department is responsible for), responsible product lines (the product categories the department is responsible for), and current customer load (the ratio of the number of customers the department is currently following up with to the department's capacity, normalized to a score range of 0-100; the lower the load, the higher the score).
[0087] During the matching score calculation phase, a matching score is calculated for each department and the current customer. All matching scores are normalized to the range of 0-100 to ensure uniformity of measurement before weighted summation. The matching score determination rules for categorical features are as follows:
[0088] The rules for determining industry matching are as follows: The system has a pre-set industry classification hierarchy table, dividing industries into first-level industries (such as "manufacturing, service, information technology"), second-level industries (such as "electronics manufacturing, machinery manufacturing, financial services, software services"), and third-level industries (such as "consumer electronics manufacturing, semiconductor manufacturing, industrial machinery manufacturing"). A perfect match is defined as having the same third-level industry, scoring 100 points; a related match is defined as having the same second-level industry but different third-level industries, scoring 60 points; a mismatch is defined as having different first-level industries, scoring 0 points.
[0089] The rules for determining regional matching are as follows: The system pre-sets a regional hierarchy table, dividing geographical regions into major regions (such as "East China Region", "South China Region", "Europe Region", "North America Region"), provinces / countries (such as "Jiangsu Province", "Guangdong Province", "Germany", "United States"), and cities (such as "Nanjing City", "Shenzhen City", "Munich", "New York City"). A perfect match is defined as the customer's city being the same as a city within the department's responsible region, or the customer's province and country being the same as a province and country within the department's responsible region, and the department not being subdivided to the city level, earning 100 points; a partial match is defined as the customer's major region being the same as the department's responsible major region, but the specific province / country or city being different (i.e., the major region containing sub-regions), earning 50 points; a no-match is defined as the customer's major region being different from the department's responsible major region, earning 0 points.
[0090] The rules for determining product line matching are as follows: The system has a pre-set product classification hierarchy table, and the industry matching hierarchy judgment logic is referenced: a perfect match is defined as the three-level product categories being completely the same, which scores 100 points; related products are defined as the two-level product categories being the same but the three-level product categories being different, which scores 60 points; and a mismatch is defined as the one-level product categories being different, which scores 0 points.
[0091] The rules for determining the matching degree of channel sources are as follows: The system divides source channels into primary categories such as exhibitions, B2B platforms, social media, search engines, and customs data, and further subdivides specific channels under each primary category (e.g., "Munich Exhibition in Germany" belongs to the exhibition category, and "Alibaba.com" belongs to the B2B platform category). A perfect match is defined as the customer's source channel and the department's preferred channel belonging to the same specific channel or a highly relevant subcategory within the same primary category, which scores 100 points; otherwise, it scores 0 points.
[0092] The department load factor score is calculated as follows: Load factor score = max(0, (1 - load rate) × 100), where load rate = number of customers currently assigned to the department / department's preset capacity. The lower the load rate, the higher the score, encouraging the even distribution of customers to departments with lighter loads. When the department load rate exceeds 100%, the load factor score is recorded as 0, thus encouraging load balancing while avoiding excessive penalties for high-matching departments due to short-term overload. All matching scores are weighted and summed after the matching degree is determined. Among them, the regional matching score has the highest weight (30%), because the matching of geographical regions in foreign trade business directly affects the feasibility and efficiency of follow-up. Customers in the same region have significant advantages in terms of time zone, culture, and logistics. The industry matching score is second (25%), because industry knowledge is the foundation of effective communication and professional services. The product line matching score is third (20%), because product familiarity directly affects conversion efficiency. The channel source matching score (15%) and the department load factor score (10%) are auxiliary adjustment factors with relatively low weights, aiming to balance channel experience and load balancing.
[0093] During the threshold determination phase, if a department's matching score exceeds the classification threshold, the customer is assigned to that department's public pool; if multiple departments exceed the threshold, the department with the highest score is selected; if no department exceeds the threshold, the customer is assigned to the company's public pool. The initial value of the classification threshold is set by the system administrator based on the number of departments and the customer load balancing goals of each business department, with a range of 60 to 80 points. After the threshold is set, the system periodically (e.g., quarterly) calculates the difference between the customer conversion rate of each department's public pool and the customer conversion rate of the company's public pool. If the conversion rate of the department's public pool is significantly lower than that of the company's public pool (with a difference of more than 10 percentage points), it indicates that the current classification threshold is set too low, resulting in too many customers with insufficient matching degree being classified into the department's public pool, diluting the overall customer quality of the department's public pool. In this case, the system will automatically raise the threshold by 2-5 points to increase the matching threshold for customers entering the department's public pool. Conversely, if the conversion rate of the department's public pool is significantly higher than that of the company's public pool (with a difference of more than 10 percentage points), it indicates that the current threshold setting may be too strict, limiting the number of potential high-quality customers entering the department's public pool. The system will automatically lower the threshold by 2-5 points to appropriately expand the coverage of the department's public pool.
[0094] For example, a foreign trade company has three sales departments: the East China Sales Department (responsible for the East China region, mainly dealing in electronic products, skilled in trade show channels, currently at 60% capacity), the South China Sales Department (responsible for the South China region, mainly dealing in machinery products, skilled in B2B platform channels, currently at 80% capacity), and the Overseas Sales Department (responsible for overseas markets, mainly dealing in all product categories, skilled in social media channels, currently at 50% capacity). A new customer is from the United States (overseas region), in the electronics industry, sourced from LinkedIn social media, and has a quality rating of 85. The system calculates the matching score for each department: Overseas Sales Department: Regional matching 100 points × 30% + Industry matching 60 points (electronics not a full range of products, belonging to a related industry) × 25% + Product line matching 100 points × 20% + Channel matching 100 points × 15% + Load factor 50 points (load rate 50%, load factor = (1-0.5) × 100 = 50 points) × 10% = 30 + 15 + 20 + 15 + 5 = 85 points; East China Sales Department: Regional matching 0 points × 30% + Industry matching 100 points × 25% + Product line matching 100 points × 20% + Channel matching 0 points × 15% + Load factor 40 points (load rate 60%, load factor = (1-0.6) × 100 = 40 points) × 10% = 0 + 25 + 20 + 0 + 4 = 49 points; South China Sales Department: Regional matching 0 points × 30% + Industry matching 0 points × 25% +Product line matching 0 points × 20% +Channel matching 0 points × 15% +Load factor 20 points (load rate 80%, load factor = (1-0.8) × 100 = 20 points) × 10% = 2 points. The overseas sales department's score of 85 points exceeds the classification threshold of 75 points, so this customer is classified into the overseas sales department's public sea.
[0095] In a preferred embodiment of the present invention, after deduplication and classification, and public sea categorization, a multi-department information attribution step is included to fully present the departmental affiliation information of cross-departmental users: Departmental public sea customers are selected from the public sea customers with assigned affiliations; the user ID set of each departmental public sea customer is extracted; all department IDs corresponding to each user ID are queried in batches through the user-department association table to obtain multiple department names to which each user belongs; the multiple department names to which each user belongs are aggregated into a department name list to serve as the user's complete departmental affiliation information in the departmental public sea customers. This multi-department information attribution step is triggered when the affiliation information of a departmental public sea customer changes (real-time mode), or is executed in batches by the system periodically (e.g., daily at midnight) (scheduled mode). For departmental public sea customers that have not yet been assigned affiliations, the multi-department information attribution step is not executed.
[0096] During the data collection phase, the system first filters out customer records with assigned users from all public customers across all departments. Then, it extracts the user IDs of these assigned users and deduplicates them to obtain a list of user IDs to be queried. In the batch query phase, the system uses a user-department relationship table to batch query all department IDs corresponding to each user ID in the aforementioned list of user IDs to be queried (this table supports many-to-many relationships, allowing a user to belong to multiple departments simultaneously). Subsequently, it batch queries the department information table to map department IDs to their corresponding department names. In the aggregation and display phase, the system iterates through the assigned users of each public customer in each department, collecting all department names for that user based on the above mapping relationship, forming a list of department names (displayed using commas or tags), which serves as the user's complete department affiliation information within the public customer pool.
[0097] In a specific application scenario, in a foreign trade CRM system, user "Li Si" holds positions in both the "East China Sales Department" and the "Overseas Sales Department." The system identifies a client, "ABC Trading LLC," under which Li Si is registered. During the multi-department information attribution process, the system queries the user-department relationship table to retrieve the list of department IDs corresponding to Li Si, and then maps this list to the department information table to obtain the list of department names ["East China Sales Department," "Overseas Sales Department"]. When displaying the client's attribution information, the front-end shows "Li Si (East China Sales Department, Overseas Sales Department)" instead of just "Li Si (East China Sales Department)," thus fully presenting Li Si's organizational position.
[0098] Reference Figure 4 As shown, after completing the intelligent classification of customers in the open market, the system continues with the customer allocation operation. The customer allocation operation includes the following sub-steps:
[0099] S33: Retrieve the customer's historical attribution trajectory recorded in the system.
[0100] The historical attribution trajectory includes every change in the customer's attribution status between the high seas and private seas, the corresponding triggering event, the follow-up result, and the follow-up duration.
[0101] The attribution trajectory recording unit maintains a complete historical attribution trajectory for each customer in the system. Attribution status includes three types: private sea customer (with a clearly defined owner), departmental public sea customer (belonging to a specific department's public sea pool), and company public sea customer (belonging to a company-wide public sea pool).
[0102] Whenever a customer's ownership status changes, the ownership trajectory recording unit automatically records a trajectory record, including the following fields: customer ID, status change timestamp, ownership status before change, ownership status after change, trigger event type (such as customer entry, successful order grabbing, proactive release, automatic release after expiration, association merging, etc.), involved user ID, follow-up result (if applicable), and follow-up duration. For example, the ownership trajectory of customer Schmidt Industrial GmbH is as follows: 2026-01-01: Entry → Assigned to the "European Sales Department" public pool (trigger event: customer entry); 2026-01-05: Department public pool → private pool (trigger event: successful order grabbing, operating user: Li Si); 2026-01-20: Private pool → department public pool (trigger event: follow-up failure release, operating user: Li Si, follow-up result: failure, follow-up duration: 15 days). The above trajectory completely records the entire process of this customer from entry to being grabbed and then released back to the public pool.
[0103] Suppose a customer's account is successfully acquired by a salesperson from the department's public account pool, and their ownership status changes from "department public account pool" to "private account pool." The triggering event is "acquisition successful," and the user involved is the salesperson who acquired the account. If the salesperson follows up for a period of time but fails to close the deal and releases the customer back to the public account pool, another record is made. The status before the change is "private account pool," the status after the change is either "department public account pool" or "company public account pool," the triggering event is either "active release" or "automatic release after expiration," the follow-up result is "failure," and the follow-up duration is the number of days from acquisition to release.
[0104] By maintaining a complete historical attribution trajectory, the system establishes a complete lifecycle profile for each customer from their entry into the system to their current status.
[0105] The information recorded by the attribution trajectory recording unit includes each change in attribution status (from what status to what status), the trigger event type (successful order grabbing, automatic release after expiration, active release, etc.), the follow-up result (success / failure / in progress), and the follow-up duration. These will serve as the data source for the order grabbing parameter adjustment unit in subsequent step S34. Specifically: by retrieving the number of records in a specific customer's attribution trajectory where "attribution status changes from public sea to private sea and the trigger event is successful order grabbing," the total number of times the customer's orders have been grabbed can be determined; by retrieving the number of records where "attribution status changes from private sea to public sea" (trigger events include active release, automatic release after expiration, release after follow-up failure, etc.), the number of times the customer has been released back to the public sea can be determined, i.e., the "number of times released after being grabbed" as described in Rule 1; by analyzing the trigger event type (automatic release after expiration or active release) and the corresponding follow-up result of each record when "attribution status changes from private sea to public sea," it can be distinguished whether the customer was automatically reclaimed by the system (long-term lack of follow-up) or actively released by the salesperson, and whether the follow-up failed or was not carried out. This data will be used in step S34 to adjust the differentiated order-grabbing parameters.
[0106] S34: When the customer is detected to have entered the order-grabbing state, adjust the order-grabbing permission determination parameters according to the historical attribution trajectory.
[0107] The order-grabbing parameter adjustment unit continuously monitors changes in customer ownership status. When a customer is detected to have entered an order-grabbing state (i.e., the customer is released back into the department's or company's public pool for sales staff to grab orders), this module reads the customer's historical ownership trajectory from the database and dynamically adjusts the order-grabbing permission determination parameters based on the trajectory information.
[0108] In this embodiment, the order-grabbing permission determination parameters specifically include a comprehensive score threshold for determining whether a candidate user is qualified to grab orders and a priority ranking weight for order-grabbing recommendations. The comprehensive score threshold is a dynamically adjustable threshold value, with an initial value preset by the system (e.g., 70 points), and can be decreased or increased according to rules 2 and 3 below. The priority ranking weight for order-grabbing recommendations controls the customer's display order position in the list of eligible customers. The following adjustment rules are independent of each other and can be applied simultaneously or selectively.
[0109] Rule 1 (Reduce Recommendation Priority When the Number of Releases After Being Bid Exceeds the Threshold): When the number of times a customer's order is released after being bid on exceeds the preset release number threshold in the historical attribution trajectory, the customer's order-grabbing recommendation priority is reduced by a preset step size (for example, every time the customer is released back into the public sea after being bid on, their recommendation priority score is reduced by 10 points). When the system displays the list of customers eligible to bid on orders to the salesperson, it sorts them from high to low recommendation priority. Reducing the recommendation priority means that the customer is displayed later in the list, avoiding low-quality or high-difficulty customers frequently occupying the salesperson's attention. It should be noted that the order-grabbing recommendation priority is an auxiliary filtering condition. When multiple customers enter the order-grabbing state at the same time, the customer with the higher priority is displayed first, but the priority itself does not change the user's judgment of the customer's eligibility to bid on orders. The eligibility judgment is still determined by the comprehensive score threshold. The system relies on the statistics of the number of records in the trajectory where "attribution status changes from private sea to public sea" (triggering events include active release, automatic release after expiration, and release after failed follow-up).
[0110] Rule 2 (Lowering the Threshold for Order-Grabbing Qualification When Released Due to Long-Term Lack of Follow-up): When a customer is released from order-grabbing status due to long-term lack of follow-up in the historical tracking, the threshold for order-grabbing qualification will be lowered by a first preset value (e.g., 2-6 points). The first preset value is positive. Release due to long-term lack of follow-up means that after a salesperson grabs an order, the customer has no follow-up record within the specified follow-up period and is automatically reclaimed by the system. This indicates that the customer's value may have been underestimated, and lowering the threshold for order-grabbing can encourage more salespeople to try to follow up. The record in the dependency tracking showing "Trigger Event: Automatic Release Due to Expiration" directly indicates that the customer was automatically reclaimed by the system due to long-term lack of follow-up.
[0111] Rule 3 (Increase the threshold for order-grabbing qualification when a customer is released due to follow-up failure): When a customer is released due to follow-up failure in the historical attribution trajectory, the threshold for order-grabbing qualification will be increased by a second preset value (e.g., 2-6 points). The second preset value is a positive value. Release due to follow-up failure means that the salesperson followed up but ultimately failed to close the deal, resulting in active or passive release. This indicates that the customer is difficult to follow up with, and increasing the order-grabbing threshold ensures that only more capable salespeople can grab the order. The record in the dependency trajectory where the "trigger event is active release and the follow-up result is failure" indicates that the customer was actively released by the salesperson and the follow-up failed.
[0112] Rule 4 (Same Success Experience Earns Matching Bonus): When the system searches a candidate user's historical follow-up records and finds that among the clients they have followed up with and successfully converted, there are records of clients with the same source channel, industry, or release pattern as the current client waiting to place an order, the candidate user's user-client matching score receives additional matching bonus points. Here, the release pattern refers to the combination of the release trigger event type and the pre-release ownership status, including at least the following categories: automatic release after expiration (release from private pool to public pool), proactive release - follow-up failure (release from private pool to public pool), proactive release - no follow-up required (release from private pool to public pool), and inter-departmental transfer (transfer from one department's public pool to another department's public pool). For example, if client A comes from the "Munich Exhibition in Germany" channel, and candidate user A successfully followed up with a client also from the "Munich Exhibition in Germany" channel three months ago and ultimately closed the deal, then user A's matching score for client A receives additional bonus points. The system relies on the "successful follow-up result" records in the candidate user's historical follow-up history. It then links these records to the corresponding customer information to obtain the source channel, industry, and release pattern from these customers' historical attribution trajectories, and matches them with the corresponding information of the currently waiting-to-order customers. It should be noted that the data source for Rule 4 is the candidate user's own historical follow-up records, which belongs to a different data dimension than the customer historical attribution trajectories relied upon by Rules 1-3.
[0113] The user-customer matching score is calculated by weighting industry matching, regional matching, and historical experience with similar customers. Each sub-dimensional is normalized to the 0-100 score range and then summed according to the following weights: industry matching weight 0.4, regional matching weight 0.3, historical experience with similar customers weight 0.3, and the sum of all weights is 1. The user capability assessment score is calculated by weighting the following: conversion rate (number of orders completed in the past 6 months divided by the number of customers followed up, normalized to the 0-100 score range), positive review rate (average historical satisfaction rating, normalized to the 0-100 score range), response time (average customer message reply time, normalized to 0-100 points, the faster the reply, the higher the score), business proficiency (product knowledge assessment score or experience in following up with similar customers, normalized to the 0-100 score range), and reputation score (records of illegal order grabbing and malicious operation deductions, normalized to the 0-100 score range after deducting the base of 100 points). The matching scoring mechanism is triggered by the order-grabbing parameter adjustment unit during the order-grabbing parameter adjustment phase (S34), but the specific numerical accumulation is uniformly completed in the user-customer matching degree evaluation score during the target user determination phase (S35). The corrected final matching degree evaluation score is then weighted and summed with the user's ability evaluation score (e.g., ability score × 0.4 + matching degree score × 0.6) to obtain the comprehensive order-grabbing score for each candidate user.
[0114] Through this step S34, the order-grabbing permission determination no longer uses fixed parameters, but is personalized based on each customer's unique "experience," realizing the historical perception capability and differentiated adaptation capability of order-grabbing decision-making.
[0115] For example, after salesperson A successfully bids for a customer, "Schmidt Industrial GmbH," in the system, the system automatically releases the customer back to the open market because salesperson A has not followed up for two consecutive weeks, recording "Release Reason: Long-term Lack of Follow-up" in the customer's tracking. When the customer re-enters the bidding state, the bidding parameter adjustment unit reads the customer's tracking and lowers the bidding qualification threshold by 5 points from the default 70 to 65. Salesperson B has a capability score of 68 and a matching score of 62, for a total score of 64.4 (68 × 0.4 + 62 × 0.6). Salesperson C has a capability score of 72 and a matching score of 60, for a total score of 64.8 (72 × 0.4 + 60 × 0.6). Salesperson D has a capability score of 75 and a matching score of 65, for a total score of 69 (75 × 0.4 + 65 × 0.6). At the adjusted threshold of 65, salespersons B and C are not qualified to bid, while salesperson D is qualified. If the customer is released due to a failed follow-up, the system will raise the threshold by 5 points to 75 points. Ding's 69 points will not be enough to reach the threshold, and a more capable salesperson is needed to grab the order.
[0116] S35: Based on the adjusted order-grabbing permission determination parameters, combined with the user capability assessment score and the user-customer matching degree assessment score, determine the target users who are qualified to grab orders, and assign the customers to the target users.
[0117] After the order-grabbing parameter adjustment unit completes parameter adjustments, the target user determination unit acquires all available candidate users (salespeople) in the system and calculates their user capability assessment score and user-customer matching score for each candidate user. Once the candidate user set is determined, if multiple customers awaiting allocation exist simultaneously (i.e., specific customer entities currently in the public pool that have triggered order-grabbing permission adjustments and allocation actions), the system sorts each customer according to their order-grabbing recommendation priority from highest to lowest, and sequentially assigns the customer with the highest recommendation priority to the user with the highest overall order-grabbing score among the remaining candidate users, until all customers awaiting allocation have been assigned or there are no available candidate users qualified to grab orders.
[0118] The user capability assessment score integrates five dimensions: salesperson's historical closing rate, customer satisfaction rate, response time, business proficiency, and reputation score. The capability score ranges from 0 to 100 after weighted summation of each dimension. The raw data for each sub-dimension are uniformly converted to a score range of 0-100 as follows: Conversion rate is calculated by dividing the number of completed orders in the past 6 months by the number of customers followed up, then multiplying the result by 100 (e.g., a conversion rate of 45% is recorded as 45 points); Positive review rate is calculated by dividing the historical average satisfaction rating (0-5 points) by 5, then multiplying by 100 to convert to a percentage; Response time is calculated based on the average customer message response time, converted using the formula max(0, 100 - average response minutes / 2) (a response time exceeding 200 minutes is recorded as 0 points); Business proficiency is based on product knowledge assessment score or experience score in following up with similar customers, directly expressed as a percentage. The system has a pre-set product knowledge question bank, and salespersons complete an online assessment once a month. The assessment score (out of 100) is directly used as the product knowledge assessment score; Simultaneously, the system counts the number of similar customers (customers in the same third-level industry as the current customer) followed up by the salesperson historically and their conversion rate, calculated using the formula: "Similar customer follow-up experience score = min(100, number of customers followed up × 5)". The final business proficiency score is calculated as follows: (Number of completed orders × 10) + (0.5 × Product knowledge assessment score + 0.5 × Experience in following up with similar customers). The reputation score is based on 100 points; each violation of order-grabbing rules or malicious operation deducts 10 points, and if the score is below 0 after deduction, it is recorded as 0. All the above sub-dimensions are standardized to a range of 0-100 points and then weighted and summed according to the following weights: completion rate 30%, positive review rate 25%, response time 20%, business proficiency 15%, and reputation score 10%. The completion rate has the highest weight (30%) because it directly reflects the salesperson's actual conversion ability and is the core indicator for evaluating the salesperson's value; positive review rate (25%) and response time (20%) reflect service quality and customer experience; business proficiency (15%) and reputation score (10%) are auxiliary evaluation dimensions with relatively lower weights.
[0119] The user-customer matching score comprehensively considers the industry matching degree between the candidate user and the current customer (Jaccard similarity between the user's historical customers' industries and the current customer's industries, normalized to the 0-100 score range; the current customer's third-level industries and their corresponding first- and second-level industries form the industry set Tcur; the user's historical customers' third-level industries and their corresponding first- and second-level industries form the industry set Tuser; industry matching degree = |Tcur ∩ Tuser| / |Tcur ∪ Tuser| × 100), regional matching degree (the degree of matching between the user's responsible region and the customer's region: perfect match 100 points, same region 50 points, different regions 0 points), and historical experience with similar customers (whether the user has followed up with similar customers and the success rate of such follow-ups, normalized to the 0-100 score range; the number of customers N and the number of orders M that the user has followed up with in the past and belong to the same second-level industry category as the current customer are counted; if N=0, the experience score = 0; if N>0, the experience score = min(100, (M / N)×80 + ...). The minimum score is (min(N, 10) × 2), with a maximum order quality score of 80 points and a maximum experience breadth score of 20 points. Each follow-up with a client adds 2 points. "10" represents the statistically minimum effective sample size (following up with 10 similar clients). The weighted sum of each dimension yields a matching score from 0 to 100. After normalizing each sub-dimension to the 0-100 range, the following weights are applied for weighted summation: industry matching score 0.4, region matching score 0.3, and historical similar client experience score 0.3, with the sum of all weights being 1. Industry matching score has the highest weight (0.4) because matching industry knowledge is a prerequisite for effective communication and professional service, directly impacting customer conversion rates. Region matching score is next (0.3) because matching geographical location affects the feasibility and efficiency of follow-ups. Historical similar client experience has the lowest weight (0.3) because while historical experience is valuable, it is not the core factor determining the current client matching score. The basic matching score is obtained by weighted summation of the above industry matching score, region matching score, and historical similar client experience. In Rule 4 of step S4 above, if a candidate user's historical follow-up records show successful follow-up records from the same source channel, industry, or release mode as the current customer, then an additional matching bonus is added to the candidate user's base matching score (e.g., 5 points for each match, the specific bonus value is configurable, multiple matches can be stacked, but the total score is capped at 100 points). This additional bonus does not change the definition and calculation method of the base matching score, but rather modifies and adjusts it to reflect the candidate user's experience advantage with specific customer types.
[0120] The specific logic for determining target users qualified to bid includes: a weighted sum of the user's capability assessment score and the user-customer matching score (the weights for the weighted sum can be referenced in the previous example; the system administrator can dynamically adjust these weights through the management backend based on actual business data) to obtain the overall bid-bidding score for each candidate user; using a bid-bidding qualification threshold adjusted based on historical attribution trajectories, comparing the overall bid-bidding score of each candidate user with this threshold, and identifying candidates who reach or exceed this threshold as the set of qualified bid-bidding users; when this set contains only one candidate user, that candidate user is identified as the target user; when this set contains multiple candidate users, the user with the highest overall bid-bidding score is selected as the target user; when this set is empty, the customer remains in the public customer pool, and the bid-bidding qualification determination is retried after a preset retry interval (e.g., 24 hours), with the re-determination based on whether new candidate users have joined within this interval (e.g., newly hired salespersons or changes in salesperson capability scores). If the candidate user set is still empty after N consecutive re-evaluations (e.g., 7 times), the customer is marked as a "low-match customer" and the system administrator will manually intervene to handle it.
[0121] For example, in step S34 above, after the customer was released due to prolonged lack of follow-up, the threshold for determining eligibility to bid for orders has been adjusted to 65 points. The system calculates the combined bid-grabbing scores of the three candidate users: Salesperson A has a capability score of 88 points and a matching score of 70 points, with a combined score of 88 × 0.4 + 70 × 0.6 = 77.2 points; Salesperson B has a capability score of 75 points and a matching score of 80 points, with a combined score of 75 × 0.4 + 80 × 0.6 = 78 points; Salesperson C has a capability score of 60 points and a matching score of 55 points, with a combined score of 60 × 0.4 + 55 × 0.6 = 57 points. Among the three candidate users, A and B both have combined scores exceeding 65 points and are included in the candidate user set; C's combined score of 57 points is below 65 points, and he is not eligible to bid for orders. Since the candidate user set includes both A and B, the system selects B (78 points), who has the highest combined score, as the target user and assigns the customer to B.
[0122] After customer allocation, to establish a feedback loop from order tracking results back to the plagiarism detection model for continuous optimization, this invention also includes a backpropagation optimization step for order tracking efficiency: collecting order tracking efficiency indicators of customers assigned to target users, including order grabbing response speed, follow-up initiation speed, stage progress speed, final follow-up result, and order amount; periodically conducting retrospective analysis on customers whose order tracking cycles have ended, and performing correlation analysis between the customer's order tracking efficiency indicators and the customer's matching confidence sub-scores and confidence levels in each dimension during the plagiarism detection process, classification decisions in the public sea classification process, and historical attribution trajectories in the order grabbing state; identifying plagiarism detection patterns associated with high order tracking efficiency based on the retrospective analysis results, and adjusting the weights of each dimension and / or the thresholds of each confidence level based on the identified plagiarism detection patterns. The plagiarism detection pattern refers to the level combination of matching confidence sub-scores of the customer to be plagiarized in each dimension, with the matching confidence sub-scores of each dimension divided into three levels: high matching, medium matching, and low matching according to a preset threshold.
[0123] Specifically, the system uses the matching confidence sub-scores of each dimension as feature variables and the order follow-up efficiency indicators (such as final follow-up results and order amount) as target variables. The system discretizes the matching confidence sub-scores (out of 100, ranging from 0 to 100) of the six dimensions into three levels: "high matching" (sub-score ≥ 80), "medium matching" (30 ≤ sub-score < 80), and "low matching / no matching" (sub-score < 30). The levels of each dimension are combined as a sample of duplicate detection patterns. By statistically analyzing the proportion of customers with high order follow-up efficiency in each sample of duplicate detection patterns, the system identifies dimension combination patterns with a significantly higher order conversion rate than the average. For example, the combination pattern of "high matching in company name dimension and high matching in social media dimension" has a high order conversion rate, and this pattern is marked as a duplicate detection pattern associated with high order follow-up efficiency. Conversely, if the order conversion rate of a certain pattern is significantly lower than the average level, it is marked as a duplicate detection pattern associated with low order follow-up efficiency. During subsequent deduplication, the system can appropriately increase the recommendation priority or lower the order-grabbing threshold for customers with high follow-up efficiency correlation patterns during the order-grabbing stage; conversely, it can correspondingly decrease the recommendation priority or raise the order-grabbing threshold for customers with low follow-up efficiency correlation patterns. The correlation analysis employs the common CART decision tree regression model (Classification and Regression Tree), using matching confidence sub-scores (six dimensions, values 0-100), confidence levels (1 / 2 / 3), classification decision results (department pool / company pool), and historical attribution trajectory characteristics (number of releases, release reason category) as feature variables, and follow-up efficiency indicators (successful order amount, continuous values) as the target variable. Model training uses the least squares splitting criterion and 10-fold cross-validation to prevent overfitting. The training dataset consists of customer records from the past 6 months that have completed their follow-up cycles (sample size no less than 500 records; if insufficient, the dataset will be accumulated until the required sample size is met before execution). By calculating the feature importance score of each feature during the decision tree splitting process (normalized based on the reduction in mean squared error, a dimensionless relative proportion value between 0 and 1), feature combinations that significantly impact order tracking efficiency are identified. When the normalized importance score of a feature exceeds 0.3, the feature is determined to be a highly relevant feature, and the deduplication weights are adjusted accordingly: the weight of highly relevant features is increased by 10%, and the weight of low-relevance features is decreased by 5%. After adjustment, the weights need to be renormalized to ensure that the sum of the six dimensions of weights is 1.
[0124] During the order follow-up performance indicator collection phase, the system continuously tracks the entire order follow-up process for each customer assigned to a salesperson, regardless of the customer's confidence level. Order grabbing response speed refers to the time interval from when a customer enters the order grabbing state to when an order is grabbed; follow-up initiation speed refers to the time interval from when an order is grabbed to when the first follow-up record is generated; stage progression speed refers to the time interval between each stage of the customer's journey from "initial contact" to "potential customer" to "quoting" to "sold"; the final follow-up result includes closed, failed, or in progress; and the closed order amount is the transaction amount for customers who have closed orders.
[0125] During the retrospective analysis phase, the system periodically performs retrospective analysis on customers whose follow-up cycles have ended (either completed or confirmed). For each customer whose follow-up cycle has ended, the system extracts their complete data chain: sub-scores and overall confidence scores for each dimension during the deduplication phase, and confidence levels; classification decisions during the categorization phase (which department or company's pool they were placed in); historical attribution trajectory during the order-grabbing phase (who grabbed the order, number of releases, etc.); and performance indicators during the follow-up phase. The system links these data together to form a complete causal chain from deduplication to follow-up. The aforementioned weight and threshold adjustments based on the retrospective analysis results take effect in real time after each analysis and are used for deduplication and categorization processing of all newly entered customers. The adjusted parameters are persistently stored in the system configuration until the next retrospective analysis triggers a new adjustment. Through this continuous iterative feedback mechanism, the system continuously optimizes the accuracy of deduplication and categorization as its runtime increases. It is important to note that for customers identified as having the highest confidence level and directly associated with existing clients during the plagiarism detection process, they do not actually enter the public customer pool and therefore do not generate their own independent public customer attribution trajectory. When conducting order tracking performance retrospective analysis, the system uses the historical attribution trajectory of the existing clients associated with the customer to be checked as its attribution data source. That is, the system obtains the complete attribution status change record of the existing client within the system and uses it as the 'virtual attribution trajectory' of the newly associated client for correlation analysis with subsequent order tracking performance indicators. This approach ensures that all customers at all confidence levels (including directly associated clients) are included in the closed-loop system of order tracking performance backpropagation optimization.
[0126] During the pattern recognition and parameter adjustment phase, the system uses statistical analysis to identify plagiarism detection patterns associated with high order follow-up efficiency. For example, if the analysis reveals that plagiarism detection results with "medium to high company name similarity and matching social media accounts" have a significantly higher conversion rate among associated clients than those with "company email only," the system automatically increases the weight of the "company name + social media" combination and relatively decreases the weight of "company email only." Similarly, if the analysis finds that the actual conversion rate among clients in the second confidence level is much higher than expected, the system appropriately lowers the second threshold for level determination, allowing more clients to enter the second confidence level and gain the advantage of weighted classification through plagiarism detection guidance.
[0127] In a specific application scenario, a foreign trade CRM system accumulated sufficient order tracking data after three months of operation. During backtracking analysis, the system discovered that a higher proportion of customers with a medium-to-high similarity in company name and perfectly matching social media accounts ultimately closed deals. Conversely, customers with perfectly matching company email addresses but no match in other dimensions had a relatively lower conversion rate. Based on this finding, the system automatically increased the overall contribution weight of the "company name + social media" combination while relatively decreasing the weight of the "company email" dimension alone. Simultaneously, the system found that the actual conversion rate of customers at the second confidence level was higher than expected, so it automatically lowered the second threshold for level determination, allowing more customers to enter the second confidence level. Through this backpropagation optimization of order tracking efficiency, the deduplication model continuously learns from actual business results, enabling the system's intelligence level to continuously improve over time.
[0128] Example 3
[0129] Combination Figure 4 As shown in the illustration, this disclosure also provides an electronic device for an intelligent management method for foreign trade customers, including a processor 500 and a memory 501. Optionally, the system may further include a communication interface 502 and a bus 503. The processor 500, communication interface 502, and memory 501 can communicate with each other via the bus 503. The communication interface 502 can be used for information transmission. The processor 500 can call logical instructions in the memory 501 to execute the intelligent management method for foreign trade customers described in the above embodiment.
[0130] Furthermore, the logic instructions in the aforementioned memory 501 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.
[0131] The memory 501, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 500 executes functional applications and data processing by running the program instructions / modules stored in the memory 501, thereby realizing the intelligent management method for foreign trade customers in the above embodiments.
[0132] The memory 501 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 501 may include high-speed random access memory and may also include non-volatile memory.
[0133] The foregoing description and accompanying drawings fully illustrate embodiments of the present disclosure to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in the embodiments of this disclosure is for descriptive purposes only and is not intended to limit the claims. As used in the description of the embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or” as used in the embodiments of this disclosure refers to any and all possible combinations of one or more of the associated listed elements. Additionally, when used in embodiments of this disclosure, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., disclosed in the embodiments, if they correspond to the method section disclosed in the embodiments, then the relevant parts can be referred to the description of the method section.
[0134] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
Claims
1. A method for intelligent management of foreign trade customers, characterized in that, The method includes: Obtain customer information from multiple dimensions of the customer to be checked for duplicates, perform database duplicate checks on each dimension, identify existing customers that match each dimension, calculate the matching confidence sub-score for each dimension, and weight and fuse the matching confidence sub-scores of each dimension that match the same existing customer, and use the highest weighted fusion score as the comprehensive confidence score. The plagiarism detection results are divided into multiple confidence levels based on the comprehensive confidence score, and the customers to be checked are classified into the public customer pool or associated with the existing customer status according to the preset customer classification trigger strategy corresponding to each confidence level. In response to the customer to be deduplicated being categorized into the public customer pool, the public customer intelligent classification processing strategy is executed, and a customer allocation operation is performed. The customer allocation operation includes: obtaining the customer's historical attribution trajectory; when the customer is detected to be in an order-grabbing state, adjusting the order-grabbing permission judgment parameters according to the historical attribution trajectory; based on the adjusted order-grabbing permission judgment parameters, combined with the user ability assessment score and the user-customer matching degree assessment score, determining the target user with order-grabbing qualifications, and allocating the customer to the target user.
2. The method according to claim 1, characterized in that, The plagiarism detection results are divided into multiple confidence levels based on the comprehensive confidence score, and a preset customer classification trigger strategy is triggered according to each confidence level, including: When the overall confidence score is 0, the customer to be checked for duplicates is determined to be a brand new customer and is classified into the public customer pool. When the overall confidence score is greater than 0, the confidence level is determined based on the relationship between the overall confidence score and a preset first threshold and a second threshold: a score higher than the first threshold is the first confidence level, a score between the second threshold and the first threshold is the second confidence level, and a score lower than the second threshold is the third confidence level, wherein the first threshold is greater than the second threshold; wherein... When the comprehensive confidence score belongs to the first confidence level, it is determined that the customer to be checked for duplicates and the existing customer are the same customer, and the customer to be checked for duplicates is associated with the ownership status of the existing customer. When the comprehensive confidence score belongs to the second confidence level, the public sea customer intelligent classification processing is performed, and the department to which the existing customer belongs, which is matched by deduplication, is used as a weighting factor in the department matching degree calculation; When the overall confidence score belongs to the third confidence level, the public sea customer intelligent classification processing is performed without adding any weighting factors related to deduplication.
3. The method according to claim 2, characterized in that, The intelligent classification and processing of offshore customers includes: Extract customer feature vectors for customers to be classified and department feature vectors for each department in the system. The customer feature vectors include source channels, inquiry quality scores, customer value levels, industry, and region. The department feature vectors include department business scope, responsible area, responsible product lines, and current customer load. For each department, a matching score is calculated with the customers to be classified. The matching score includes at least a weighted sum of regional matching score, industry matching score, product line matching score, channel source matching score, and departmental load factor score. If the highest matching score exceeds the classification threshold, the customer will be classified into the department's public pool corresponding to that department; if the matching scores of all departments are lower than the classification threshold, the customer will be classified into the company's public pool.
4. The method according to claim 3, characterized in that, When a customer is categorized into a department's public pool based on the aforementioned public pool customer intelligent classification processing strategy, the process also includes a multi-department information attribution step: Filter out customers who have been assigned to specific users from the department's public customer pool, and extract their set of user IDs; By batch querying the user-department relationship table, we can obtain all department IDs corresponding to each user ID and retrieve the names of multiple departments to which each user belongs. The multiple department names to which each user belongs are aggregated into a list of department names to serve as the complete department affiliation information for that user among the department's public sea of customers.
5. The method according to claim 2, characterized in that, The method further includes a threshold dynamic adjustment step: periodically adjusting the first threshold and / or the second threshold based on the customer's subsequent order conversion rate at the second and / or third confidence levels; and, at the second confidence level, the increase in the matching score of the department to which the existing customer belongs after deduplication is Δ=α·C total , where α is a preset coefficient, with a value of 0.1 < α < 0.
6.
6. The method according to claim 2, characterized in that, The multiple dimensions include six dimensions: company name, company phone number, contact person's phone number, company email address, contact person's email address, and social media accounts; the comprehensive confidence score is calculated according to the following formula: S1 to S6 correspond to the matching confidence sub-scores of the above six dimensions, and W1 to W6 are the corresponding weights, with the sum of each weight being 1.
7. The method according to claim 1, characterized in that, The historical attribution trajectory includes every change in the customer's attribution status between the high seas and private seas, the corresponding triggering events, follow-up results, and follow-up duration. Adjusting the order-grabbing permission determination parameters based on the historical attribution trajectory includes: When the number of times a customer's orders are released after being bid on exceeds a preset release number threshold in the historical attribution trajectory, the customer's order-bid recommendation priority is reduced. When a customer is released from the historical attribution trajectory due to long-term lack of follow-up, the comprehensive score threshold used to determine whether the customer is eligible to bid will be reduced by a first preset value. When a customer is released due to follow-up failure in the historical attribution trajectory, the comprehensive score threshold is increased by a second preset value; When a candidate user's historical follow-up records contain successful follow-up records with customers who share the same source channel, industry, or release pattern as the current customer waiting to place an order, an additional matching score is added when calculating the candidate user's user-customer matching score. The user-customer matching score is calculated by weighting industry matching, regional matching, and historical experience with similar customers.
8. The method according to claim 7, characterized in that, The method for determining target users with order-grabbing privileges based on the adjusted order-grabbing permission judgment parameters, combined with user capability assessment scores and user-customer matching degree assessment scores, includes: The user capability assessment score and the user-customer matching score are weighted and summed to obtain the overall order-grabbing score for each candidate user; the user capability assessment score is calculated by weighting the order completion rate, positive review rate, response time, business proficiency, and reputation score. Using the comprehensive score threshold adjusted by the historical attribution trajectory, the comprehensive score of each candidate user for order grabbing is compared with the comprehensive score threshold, and the candidate users who reach or exceed the comprehensive score threshold are determined as the set of candidate users who are qualified to grab orders. When the candidate user set contains only one candidate user, that candidate user is determined as the target user; when the candidate user set contains multiple candidate users, the user with the highest overall order-grabbing score is selected as the target user.
9. The method according to any one of claims 1 to 8, characterized in that, After assigning the customer to the target user, the process also includes a backpropagation optimization step for order tracking efficiency: Collect the order follow-up efficiency indicators of customers after they are assigned to target users. The order follow-up efficiency indicators include order grabbing response speed, follow-up initiation speed, stage progress speed, final follow-up result and order amount. Regularly conduct retrospective analysis on customers whose order tracking cycle has ended, and perform correlation analysis on the customer's order tracking efficiency indicators with the customer's confidence sub-scores and confidence levels in various dimensions during the deduplication process, the classification decision in the intelligent classification and processing of public customers, and the historical affiliation trajectory in the state of being able to grab orders. Based on the backtracking analysis results, identify the plagiarism detection patterns associated with high follow-up efficiency, and adjust the weights of each dimension and / or the thresholds of each confidence level based on the identified plagiarism detection patterns.
10. A foreign trade customer intelligent management system, characterized in that, include: The deduplication module is used to obtain customer information from multiple dimensions of the customer to be deduplicated, perform database deduplication on each dimension, determine the existing customers matched by each dimension and calculate the matching confidence sub-score for each dimension, and weight and fuse the matching confidence sub-scores of each dimension that match the same existing customer, and use the highest weighted fusion as the comprehensive confidence score. The confidence level and classification module is used to divide the deduplication results into multiple confidence levels according to the comprehensive confidence score, and classify the deduplication customers to be deduplicated into the public customer pool or associate the deduplication customers to be deduplicated with the existing customer ownership status according to the preset customer classification trigger strategy corresponding to each confidence level. The public sea classification processing module is used to execute the public sea customer intelligent classification processing strategy in response to the fact that the customer to be deduplicated is classified into the public sea customer pool; The customer allocation module is used to perform a customer allocation operation in response to the customer to be deduplicated being classified into the public customer pool. The customer allocation operation includes: The attribution trajectory recording unit is used to obtain the customer's historical attribution trajectory; The order-grabbing parameter adjustment unit is used to adjust the order-grabbing permission determination parameters according to the historical attribution trajectory when the customer is detected to have entered the order-grabbing state. The target user determination unit is used to determine target users who are qualified to bid for orders based on the adjusted order-grabbing permission judgment parameters, combined with user ability assessment scores and user-customer matching degree assessment scores, and to assign the customers to the target users.
