A customer manager auxiliary operation system for new energy automobile consumption finance

Abstract

The application discloses a customer manager auxiliary operation system for new energy automobile consumption finance, and relates to the field of automobile consumption finance.The system comprises a model construction module, a business stage identification engine, a rule constraint processing module, a task generation module and a feedback optimization engine.The model construction module is used for establishing a unified structured model.The business stage identification engine is used for acquiring multi-source state information and identifying the current business stage of a customer.The rule constraint processing module is used for matching a material list and a service rule according to the business stage, and outputting via an artificial confirmation interface for confirmation.The task generation module is used for generating a customer manager auxiliary task according to the confirmed rule.The service trace module is used for recording a processing process and customer feedback and correlating to evidence records.The feedback optimization engine is used for updating a rule condition according to historical processing results and synchronizing to the model.The application can improve the standardization level and rule self-adaptive ability of customer manager auxiliary operation without intervening in credit approval and loan decision-making.

Description

Technical Field

[0001] This invention relates to the field of auto consumer finance, and in particular to a customer manager auxiliary operation system for new energy vehicle consumer finance. Background Technology

[0002] New energy vehicle consumer finance services involve complex collaboration among multiple entities, including OEMs, dealers, and stores. From placing an order, submitting documents, applying for installment plans, and tracking progress to delivery, customers require multiple rounds of cross-entity communication and information flow. With the continuous increase in the market penetration of new energy vehicles and the rapid growth of auto consumer finance business, banks and auto finance institutions are placing increasingly higher demands on the efficiency, multi-terminal collaboration capabilities, and consistency of service quality of account managers. Currently, the industry has begun to explore the introduction of artificial intelligence technology into the auto finance field. However, existing technical solutions mostly focus on intelligent responses, automated approvals, or end-to-end business decisions for end customers, lacking a technical framework specifically designed for account manager operations that integrates multi-source information in a structured manner and generates executable auxiliary tasks. Against this backdrop, how to provide standardized, traceable, and intelligent auxiliary operational capabilities for new energy vehicle consumer finance account managers without replacing core financial functions such as bank credit approval and loan disbursement decisions has become a pressing technical problem to be solved in this field.

[0003] Existing auxiliary tools and methods for auto finance customer managers have the following shortcomings: First, customer car purchase and installment services involve multiple business stages such as consultation, deposit placement, and document preparation. The required material lists, service scripts, and operating procedures vary at each stage, and existing tools lack the ability to automatically identify business stages based on the customer's real-time status and dynamically match corresponding operating rules. Second, in the new energy vehicle consumer finance scenario, information such as customer order status, document submission progress, and store collaboration matters are scattered across different systems. Existing tools lack a mechanism to unify and aggregate multi-source heterogeneous information and present it in a structured manner according to business stages, resulting in information inconsistency and delayed progress tracking. Third, existing systems lack the ability to automatically identify and proactively warn of abnormal states such as customer status timeouts, missing documents, and abnormal progress. The discovery of abnormal nodes still relies on manual inspections by customer managers, affecting the timeliness of service response and customer experience. Therefore, based on the above problems, this invention proposes an auxiliary operating system for customer managers in new energy vehicle consumer finance. Summary of the Invention

[0004] To address the aforementioned issues, the present invention aims to provide a customer manager auxiliary operation system for new energy vehicle consumer finance. This system can improve the customer manager's response efficiency at various business nodes, the consistency and standardization of service execution, enhance the controllability and traceability of multi-entity collaborative operations, and provide a data-driven closed-loop feedback basis for the continuous optimization of operation rules, without intervening in the bank's core financial functions such as credit approval, loan disbursement decisions, and risk pricing.

[0005] To achieve the above objectives, this invention provides a customer manager assistance system for new energy vehicle consumer finance. The system includes: a model building module for establishing a unified structured model; a business stage identification engine for acquiring multi-source status information and identifying the customer's current business stage; a rule constraint processing module for matching a materials list and service rules according to the business stage and outputting them for confirmation via a manual confirmation interface; a task generation module for generating customer manager assistance tasks based on the confirmed rules; a service traceability module for recording the processing procedure and customer feedback and linking it to evidence records; and a feedback optimization engine for updating rule conditions based on historical processing results and synchronizing them to the model.

[0006] In a first aspect, the present invention provides a customer manager auxiliary operation system for new energy vehicle consumer finance, comprising: The model building module is used to create a unified structured object model that includes role objects, process objects, rule objects, knowledge objects, task objects, interface objects, permission objects, and evidence record objects. The objects are associated with each other through predefined relationships. The business stage identification engine is used to acquire multi-source status information of customers in the process of new energy vehicle consumer finance services, and identify the current business stage of the customer based on the multi-source status information and the stage flow logic defined by the process object. The rule constraint processing module is used to match the corresponding bill of materials and service rules from the rule object according to the business stage, and output the matching result through the manual confirmation interface for the account manager to confirm and then take effect. The task generation module is used to generate customer manager auxiliary tasks based on the manually confirmed list of materials and service rules. The anomaly warning module is used to monitor at least one of the following: customer status, material submission status, progress status, and store collaboration status, and generate an warning task when an anomaly is detected. The service record module is used to associate the customer manager's handling process of the auxiliary task and the early warning task, the manual connection operation record and the customer feedback information with the evidence record object for storage; The feedback optimization engine is used to update the bill of materials matching conditions and / or anomaly judgment conditions in the rule object based on the historical processing results and customer feedback information stored in the evidence record object, and to synchronize the updated rule object to the structured object model.

[0007] Furthermore, the model building module uses a metadata-driven approach to define the attribute fields of each object and the relationships between objects. The process object contains a reference identifier for the rule object and a trigger condition identifier for the task object. The task object contains an association identifier for the evidence record object.

[0008] Furthermore, the matching result output by the rule constraint processing module is transmitted to the account manager terminal via the manual confirmation interface. After receiving the confirmation instruction returned by the account manager terminal, the rule constraint processing module transmits the confirmed matching result to the task generation module. The rule constraint processing module also records the modification operations on the matching result during the manual confirmation process and associates the modification operation records with the evidence record object.

[0009] Furthermore, the anomaly warning module adopts a multi-threshold hierarchical warning mechanism, which is configured with different anomaly judgment logic and corresponding warning levels for different types of anomaly states. When an anomaly state is determined, the priority and method of pushing the warning task are determined according to the anomaly type and warning level.

[0010] Furthermore, the feedback optimization engine includes a rule update unit and a model synchronization unit. The rule update unit performs statistical analysis on the processing time data, customer response rate data, and anomaly detection rate data of multiple historical tasks stored in the evidence record object, and generates rule update parameters based on the statistical analysis results. The model synchronization unit maps the rule update parameters to attribute value adjustment instructions for corresponding rule entries in the rule object, and synchronously updates the adjusted rule object to the structured object model.

[0011] Furthermore, each rule entry in the rule object is configured with a variable weight value; the feedback optimization engine calculates the processing timeliness index, customer response effectiveness index, and abnormal triggering frequency index of the historical tasks associated with each rule entry based on the historical task processing records stored in the evidence record object, and performs weighted fusion after normalization of the three indicators to generate an updated weight value for each rule entry, and writes the updated weight value into the rule object to replace the original weight value; the task generation module determines the generation priority and push order of the generated auxiliary tasks based on the updated weight value, and the rule constraint processing module determines the display order of each material item in the material list based on the updated weight value.

[0012] Furthermore, the anomaly warning module is equipped with a dynamic threshold adjustment unit. The dynamic threshold adjustment unit extracts historical stage dwell time data of all customers currently in the same business stage from the evidence record object, divides the dwell time into several groups and calculates the proportion of the number of customers in each group to the total number of customers, and calculates the state transition entropy value of the business stage based on the proportion. The dynamic threshold adjustment unit tightens or loosens the time limit threshold currently used in the business stage according to the level of the state transition entropy value, and uses the corrected time limit threshold as the warning judgment benchmark for the business stage.

[0013] Furthermore, the rule constraint processing module is configured with a subset of business rules, a subset of compliance rules, a subset of product rules, and a subset of process rules. The subset of business rules defines the standard operating procedures for each business stage, the subset of compliance rules defines the compliance boundary conditions for financial services, the subset of product rules defines the service templates for different types of financial products, and the subset of process rules defines the transition conditions and sequence between business stages. The rule constraint processing module extracts the corresponding rule entries from each subset of rules according to the current business stage of the customer and combines them to form a set of rule constraints that apply to the current business stage.

[0014] Furthermore, the auxiliary tasks generated by the task generation module are configured with different execution templates according to the task type. The execution templates include task description templates, script templates, and operation guidance templates. After the task generation module pushes the generated auxiliary tasks to the customer manager terminal, it triggers the execution of the corresponding subsequent process nodes in the process object based on the task processing status change events recorded in the evidence record object.

[0015] Secondly, the present invention also provides a method for assisting customer managers in the operation of new energy vehicle consumer finance, the method being based on the system described in the first aspect above, comprising: Establish a unified structured object model that includes role objects, process objects, rule objects, knowledge objects, task objects, interface objects, permission objects, and evidence record objects, and define the relationships between these objects; Obtain multi-source status information of customers during the new energy vehicle consumer finance service process, and identify the current business stage of the customer based on the multi-source status information and the stage flow logic defined by the process object; Match the corresponding bill of materials and service rules from the rule object according to the business stage, and output the matching results through the manual confirmation interface for the account manager to confirm. Generate customer manager support tasks based on the manually confirmed materials list and service rules; Monitor at least one of the following: customer status, material submission status, progress status, and store collaboration status, and generate an alert task when an abnormal status is detected; The process of the account manager handling the auxiliary tasks and the early warning tasks, the records of manual follow-up operations, and the customer feedback information are associated with the evidence record object and stored. Based on the historical processing results and customer feedback information stored in the evidence record object, update the bill of materials matching conditions and / or anomaly judgment conditions in the rule object, and synchronize the updated rule object to the structured object model.

[0016] This invention constructs a unified structured object model to manage the associated elements of new energy vehicle consumer finance customer manager operations, including roles, processes, rules, knowledge, tasks, interfaces, permissions, and evidence records. Based on the customer's real-time status, it automatically identifies business stages and dynamically matches rule constraints to generate auxiliary tasks. Combined with a tiered early warning mechanism for multiple types of anomalies and full-process service traceability, it enables standardized output and precise task delivery for customer manager auxiliary operations, effectively reducing the risks of oversight caused by fragmented cross-system information and manual follow-up. The established manual confirmation interface and feedback optimization mechanism allow rule matching results to be manually verified before execution, and historical data after execution can drive adaptive adjustments to rule conditions. This improves the controllability, traceability, and objectivity of rule evolution in the service process while maintaining clear system boundaries.

[0017] Beneficial effects By implementing the customer manager auxiliary operation system for new energy vehicle consumer finance provided by the present invention, the following technical effects are achieved: (1) Define the attribute fields of objects such as roles, processes, rules, knowledge, tasks, interfaces, permissions and evidence records and the nested reference relationships between objects through metadata. This enables process objects to carry rule reference identifiers and task trigger condition identifiers, and task objects to carry evidence record association identifiers. This allows the data flow between modules during the execution of business processes to be traceable, improving the structured consistency and cross-module audit traceability efficiency after the convergence of multi-source heterogeneous information.

[0018] (2) By setting up a mandatory manual confirmation interface between the rule constraint processing module and the task generation module, the material list and service rule matching results output by the rule object must be verified and confirmed by the customer manager terminal before execution. At the same time, the system records the modification operations in the confirmation process differently and associates them with the evidence record object. Thus, while maintaining the automation of auxiliary operations, it ensures that the rule execution boundary is always within the range of human control, avoiding unclear service responsibility boundaries and financial compliance risks caused by simple automated output.

[0019] (3) By establishing a quantitative mapping relationship between the weight of rule items and the timeliness of historical task processing, customer response rate and frequency of abnormal triggering, and introducing an exponential smoothing update strategy, the priority order of each material item in the bill of materials and the task push order of the task generation module can continuously evolve with the accumulation of actual operation data, thereby improving the fit between rule configuration and actual operation efficiency, and enhancing the accuracy and timeliness orientation of the auxiliary tasks received by customer managers.

[0020] (4) By calculating the probability distribution entropy value of the customer’s stay time in a specific business stage, the regularity of group behavior is quantified, and the time limit threshold of the abnormal warning is dynamically adjusted according to the entropy value, so that the warning sensitivity can adapt to the changes in the dispersion of customer behavior patterns in different stages and time periods, effectively reducing the risk of missed reporting caused by the static threshold when the behavior regularity is strong and the redundancy of false reporting caused by the behavior dispersion is large, and improving the matching rationality between abnormal discovery and the actual operation rhythm. Attached Figure Description

[0021] To make the above-described customer manager auxiliary operation system for new energy vehicle consumer finance of the present invention more obvious and understandable, the drawings used in the specific embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the method of the present invention. Figure 2 This is a flowchart illustrating the process of business phase identification and rule confirmation. Figure 3 This is a flowchart illustrating the task generation and anomaly warning handling process. Figure 4 This represents a flowchart illustrating feedback optimization and rule adaptation. Detailed Implementation

[0023] Example 1: This embodiment provides a customer manager auxiliary operation system and method for new energy vehicle consumer finance. In actual deployment, it can be deployed privately, via a dedicated cloud, or in a controlled zone, depending on the internal system construction requirements of the financial institution. The system interacts with external banking business systems, customer management systems, process systems, knowledge base systems, intelligent agent platforms, or digital human platforms through controlled interfaces. These controlled interfaces are configured with data format conversion and permission verification functions to ensure the consistency of the exchanged data format and the legality of data access.

[0024] like Figure 1As shown, the overall system operation flow is as follows: First, the model building module establishes a unified structured object model containing role objects, process objects, rule objects, knowledge objects, task objects, interface objects, permission objects, and evidence record objects, and defines the relationships between these objects; then, the business stage identification engine identifies the customer's current business stage based on multi-source status information obtained from multiple channels, combined with the predefined stage flow logic in the process objects; the rule constraint processing module matches the corresponding material list and service rules from the rule objects according to the identified business stage, and the rules take effect after being verified and confirmed by the account manager through a manual confirmation interface; the task generation module generates the task based on the confirmed... The rules generate auxiliary tasks for account managers and push them to the account manager's terminal; the anomaly warning module continuously monitors customer status, material submission status, progress status, and store collaboration status during operation, and generates warning tasks when an anomaly is detected; the service traceability module associates the account manager's processing of auxiliary tasks and warning tasks, manual follow-up operation records, and customer feedback information with the evidence record object for storage; the feedback optimization engine updates the material list matching conditions and / or anomaly judgment conditions in the rule object based on the historical processing results and customer feedback information stored in the evidence record object, and synchronizes the updated rule object to the structured object model to form a closed-loop adaptive evolution.

[0025] During system operation, the business stage identification engine first receives information from multiple channels, including customer car purchase order status, installment application status, document submission status, store service status, and bank process status. Based on this multi-source status information and its internally predefined stage transition logic, the identification module determines the customer's current business stage. Business stages include consultation, deposit placement, document preparation, installment application, progress tracking, benefits explanation, and service delivery. Each stage has transition conditions and trigger events. When the received status information meets the transition conditions from the current stage to the next, the stage identification module automatically updates the customer's corresponding stage field.

[0026] like Figure 2As shown, the specific process for business stage identification and rule confirmation is as follows: After the system obtains information such as customer order status, installment application status, document submission status, store service status, and bank process status from multiple sources, the business stage identification engine determines the customer's current business stage based on this multi-source status information and the predefined stage flow logic in the process object. After identifying the customer's current business stage, the rule constraint processing module extracts the corresponding material list and service rules from the rule base according to the stage. The rule base consists of business rules, compliance rules, product rules, and process rules. Different rule subsets define the standard operating procedures for each stage, the compliance boundaries of financial services, service templates for different financial product types, and the flow order between business stages. The rule constraint processing module also selects a matching set from multiple alternative templates under the same business stage as the final output material list based on the financial product type or customer type selected by the customer. After matching, the matching result is not executed automatically but is output to the customer manager's terminal via a manual confirmation interface. The customer manager verifies and confirms the material list and service rules. The customer manager can modify the matching result according to the actual situation, and both modification operations and confirmation instructions are recorded by the system. Only after receiving confirmation from the account manager will the matching result be passed to the subsequent modules to take effect. The system also records every modification operation on the matching result during the manual confirmation process and associates it with the corresponding evidence record object to ensure that the rule execution boundary is always within the range of human control.

[0027] The confirmed materials list and service rules are sent to the task generation module. Based on this, the task generation module generates auxiliary tasks for account managers. These auxiliary tasks include customer reminders, document supplementation, store collaboration, progress tracking, and exception handling. For different task types, the system is configured with different execution templates, including task description templates, script templates, and operation guidance templates. The task generation module pushes the generated auxiliary tasks to the account manager's terminal device and, based on the task processing status change events recorded by the service tracking module, triggers the execution of the corresponding subsequent process nodes in the process object.

[0028] like Figure 3As shown, the specific process for task generation and anomaly warning handling is as follows: After the confirmed rule constraints are sent to the task generation module, the system matches the corresponding execution template according to the task type and generates an auxiliary task, which is then pushed to the account manager's terminal. During system operation, the anomaly warning module monitors the customer status, material submission status, progress status, and store collaboration status in real time. Monitoring dimensions include: whether the customer's dwell time in the current business stage exceeds the preset time limit threshold, whether the submitted materials meet the preset material integrity conditions, whether the bank process status is consistent with the customer stage, and whether store collaboration matters have been effectively responded to. For different types of abnormal states, the anomaly warning module adopts a multi-threshold hierarchical warning mechanism, configured with different anomaly judgment logic and corresponding warning levels. The judgment logic includes time limit deviation judgment, material integrity judgment, and status consistency judgment. When an anomaly is detected, the anomaly warning module determines the push priority and push method of the warning task according to the anomaly type and warning level, generates the corresponding warning task, and pushes it to the account manager.

[0029] The service logging module records the entire process of account managers handling auxiliary and early warning tasks. Records include task completion status, account manager's manual follow-up operations, customer feedback, anomaly handling results, and changes in business metrics. All records are associated with corresponding customer and task identifiers and stored in the system as evidence for subsequent auditing, tracing, and rule optimization.

[0030] like Figure 4 As shown, the specific process of feedback optimization and rule adaptation is as follows: After the service trace module associates the processing process, operation records, and customer feedback with the evidence record object, the feedback optimization engine periodically extracts the historical processing results and customer feedback information stored in the evidence record object from the service trace module. Based on this data, the feedback optimization engine optimizes and adjusts the material list matching conditions used by the rule constraint processing module and the anomaly judgment conditions used by the anomaly warning module. Specifically, the feedback optimization engine statistically analyzes indicators such as the average processing time, customer response rate, and anomaly detection rate of historical tasks under each business stage, and updates the attribute values ​​of the corresponding rule entries in the rule base accordingly. For example, for the material list of a certain business stage, if a certain material item frequently causes customer re-parts or timeouts in historical tasks, the system can adjust the priority of that material item in the list or the re-parts reminder information; for the time limit threshold of the anomaly warning, if the historical average dwell time of a certain stage changes significantly, the system adjusts the timeout judgment threshold of that stage accordingly. The update operation of the feedback optimization engine can be expressed in the following general form: for any adjustable parameter in the rule object Its updated value From the current value In addition to historical feedback data Correction items Composition, that is ,in The preset learning rate coefficient, For historical datasets The reflected error or deviation statistics. This correction item can be calculated using different statistical methods such as mean deviation, median offset, or percentile change, depending on the actual business scenario. The updated rule parameters are synchronized back to the rule objects in the structured object model, enabling the system rule configuration to continuously evolve with the accumulation of work data, thereby achieving a closed-loop feedback of execution feedback, abnormal states, and manual processing results in the method process.

[0031] Throughout the system, the processing results of all sensitive matters involving customer information, approval results, credit limits, risk pricing, and the suitability of financial products are not directly output as the final business conclusion. The system sets boundaries for manual confirmation or result citation at the corresponding stages to ensure that the final business conclusion is based on the results of the financial institution's authorized system or manual confirmation. The system does not participate in credit approval, loan disbursement decisions, or risk pricing.

[0032] Example 2: In the process of assisting customer managers in new energy vehicle consumer finance operations, the rule constraint processing module extracts corresponding material items and service rules from the rule object based on the business stage. However, the impact of different material items on the timeliness of task completion and customer conversion rate varies significantly under different business stages, customer types, or time windows. Existing technologies typically use static rule configuration, lacking the ability to dynamically adjust the weight of each rule item based on historical task performance. This embodiment constructs a weight vector for rule items and calculates the contribution index of each rule item to task performance based on the historical task processing results recorded by the service trace module, thereby dynamically adjusting the weight coefficient of the corresponding rule item in the rule object. Specifically, the system assigns a variable weight value to each rule item, which is calculated by weighting the normalized indicators of three dimensions: the average processing time of the historical tasks associated with the rule item, the customer response rate, and the frequency of abnormal triggering. The system periodically recalculates the weight vector and synchronizes the updated weight value to the rule object in the structured object model, so that the priority order of each material item in the bill of materials, the trigger sensitivity of reminder tasks, and the sorting basis when the task generation module generates auxiliary tasks can all adaptively optimize with the accumulation of historical task data.

[0033] During the initialization phase, the system configures an initial weight value for each rule entry in the rule object. These initial weight values ​​are set by business experts based on historical experience, or assigned default values ​​by the system based on predefined business priorities. Once the system has run and accumulated a certain number of historical task processing records, the feedback optimization engine initiates an adaptive weight adjustment process.

[0034] First, the feedback optimization engine extracts a set of historical task records associated with the current rule entry to be optimized from the evidence record object of the service trace module. Each historical record contains the following fields: task type identifier, rule entry identifier, task generation timestamp, task completion timestamp, customer feedback status, and exception flag.

[0035] Secondly, the feedback optimization engine calculates performance metrics across the following three dimensions for each rule entry: The first dimension is the average response time metric, which characterizes the statistical features of the time interval between the generation of the task associated with the rule entry and the first processing by the account manager. The system uses exponential smoothing to calculate the average processing time of all associated historical tasks, giving higher weight to recent data so that the metric can more sensitively reflect the current operational status.

[0036] The second dimension is the customer response effectiveness index, which represents the proportion of customers who provide effective responses in the tasks associated with this rule entry. The system calculates the ratio of the number of tasks with a customer feedback status of "responded" to the total number of tasks in the associated historical tasks.

[0037] The third dimension is the anomaly avoidance contribution index, which characterizes the frequency with which the task associated with the rule entry triggers anomaly warnings during execution. The system calculates the ratio of the number of tasks with anomaly tags in the associated historical tasks to the total number of tasks. The lower this ratio, the greater the contribution of the rule entry to anomaly avoidance.

[0038] Subsequently, the feedback optimization engine normalizes the indicator values ​​of the three dimensions mentioned above, mapping them to a unified numerical range. The normalization process uses a min-max standardization method, taking the minimum and maximum values ​​of all rule entries in the same dimension as the standardization benchmark.

[0039] After normalization, the feedback optimization engine calculates the updated weight value for each rule entry using the following formula:

[0040] In the formula, For rule entries In the next update cycle, the updated weight value is adaptively adjusted and written into the rule object in the structured object model. This value is used to determine the priority ranking and material list display order when generating subsequent tasks. The preset smoothing coefficient has a value range of [value range missing]. This is used to control the degree of trade-off between historical weight values ​​and current performance feedback values. , , The preset dimension weight coefficients correspond to the relative importance of the average response time dimension, the customer response effectiveness dimension, and the contribution of unusual avoidance dimension, respectively, and the three satisfy the following: The normalization constraint has a specific value that is configured during system initialization based on the business scenario. For the rule entries within the current period The normalized value of the average processing time of multiple associated historical tasks; the normalization uses the min-max normalization method, with the benchmark taken from the minimum and maximum values ​​of the average processing time dimension of all rule entries participating in the weight calculation within the current update cycle, and the output mapped to... The interval represents the processing timeliness of the task associated with the rule item. For the rule entries within the current period The normalized value of customer response effectiveness for multiple associated historical tasks; the original value is the ratio of the number of tasks with a customer feedback status of "responded" to the total number of tasks in the associated historical tasks. Normalization uses a min-max normalization method, with the benchmark taken from the minimum and maximum values ​​of all rule entries involved in the weight calculation within the current update cycle in the customer response effectiveness dimension. The output is mapped to... Interval. For the rule entries within the current period The normalized value of the anomaly trigger frequency of multiple associated historical tasks; the original value is the ratio of the number of tasks with anomaly tags to the total number of tasks in the associated historical tasks. Normalization uses a min-max normalization method, with the benchmark taken from the minimum and maximum values ​​of all rule entries participating in the weight calculation within the current update cycle in the anomaly trigger frequency dimension. The output is mapped to... Interval. For rule entries The original weight value in the current update cycle, that is, the weight retained after the adjustment in the previous cycle.

[0041] The feedback optimization engine will calculate the updated weight values. The rule object is written into the structured object model, replacing the original weight value of the rule entry. When generating auxiliary tasks, the task generation module determines the generation priority and push order of each task item based on the updated weight values: material prompts or customer reminders corresponding to rule entries with higher weight values ​​will be generated and pushed to the account manager's terminal first. When matching the material list, the rule constraint processing module sorts and displays material items according to their weight values, with material items with higher weight values ​​appearing earlier in the list to guide account managers to prioritize and process them.

[0042] The aforementioned adaptive weight adjustment process is automatically triggered and executed at a preset time period, forming a closed-loop mechanism in which the rule weights continuously evolve with the operational efficiency data.

[0043] A comparative experiment was conducted between the mechanism described in this embodiment and a conventional method using static weight configuration. The comparison conditions were set as follows: both methods used the same initial set of rule entries and the same task generation logic; the experiment selected the same batch of customer orders from the same type of customer served by on-duty account managers; and the business scenarios, personnel configurations, financial product types, and operating periods of the two groups were ensured to be completely consistent. The difference was that the rule weights in the experimental group were adaptively updated every 24 hours, while the weights in the control group remained fixed. The experiment was run in parallel under the same new energy vehicle consumer finance business scenario, collecting full task processing data for 30 consecutive natural days as a statistical sample. The comparison results showed that the rank correlation coefficient between the rule entry weights in the experimental group and the actual task processing time was approximately 22% to 28% higher than that in the control group. This range represents the fluctuation range of the increase in different statistical periods within the experimental period, indicating a significant improvement in the consistency between weight ranking and actual operational efficiency. The average submission time for the top three priority material items in the material list was shortened by approximately 15% to 20% compared to the control group. This range represents the time reduction range under different business sub-stages, reflecting the enhanced effectiveness of high-weight rule items in guiding account manager behavior. This effect is attributed to the weight update formula... and The synergistic effect of the two factors leads to a continuous increase in the weight of rule items with fast processing speed and low anomaly rate, which in turn forms a positive reinforcement cycle through the adjustment of task push ranking.

[0044] Example 3: In the auxiliary operation system for customer managers in new energy vehicle consumer finance, the anomaly warning module determines whether a customer's status is abnormal based on preset time limit thresholds and material integrity conditions. However, the distribution of normal operation time varies significantly across different business stages, customer groups, and time periods, and a fixed threshold often leads to a high false alarm or false negative rate. This embodiment introduces the concept of state transition entropy to construct an uncertainty measure of customer transitions between different business stages, and dynamically adjusts the time limit threshold for anomaly warnings based on this. Specifically, the system models the historical state transition sequence of each customer, calculates the probability distribution characteristics of their dwell time in the current business stage, and determines the dynamic threshold based on the quantiles of this distribution. Simultaneously, the system introduces a state transition entropy factor to characterize the degree of certainty of a customer's transition from the current stage to the next stage: when a large number of homogeneous customers complete stage transitions within similar time windows, the state transition entropy value is low, indicating that the normal operation time in this stage has high regularity, and the system will tighten the warning threshold to improve the sensitivity of anomaly detection; when customer behavior patterns are more dispersed, the state transition entropy value is high, and the system will appropriately relax the threshold to avoid false alarms.

[0045] During operation, the system's business phase identification engine continuously tracks the entry and exit times of each customer in each business phase and records phase transition events to the evidence record object. Each phase transition record contains at least the following fields: customer identifier, source phase identifier, target phase identifier, phase entry timestamp, phase exit timestamp, and phase stay duration.

[0046] When the anomaly warning module initiates the adaptive threshold adjustment process, it first extracts historical dwell time data for all customers currently in the same business stage from the evidence record object, constructing an empirical distribution of dwell time under that stage. Let the current business stage be... The system extracts all data from the most recent preset time window, from each stage. Transfer customer records to the next stage to obtain a set of dwell time. .

[0047] Secondly, the anomaly warning module calculates the statistical characteristics of the set of dwell times, including the median, quartiles, and upper and lower quartile distances. Based on these statistical characteristics, the system determines a reference range for the normal dwell time during this period. The upper limit of the reference range is initially set to the upper quartile plus a preset multiple of the quartile distance.

[0048] At the same time, the anomaly warning module calculates the state transition entropy value. The system will consider the same business phase... All customers are divided into several equally wide intervals based on their dwell time. The number of customers falling into each group is counted, and the probability distribution of dwell time in that interval is calculated. The formula for calculating state transition entropy is as follows:

[0049] In the formula, For business phase The state transition entropy value represents the degree of dispersion or uncertainty of the duration of all customers in this stage across multiple groups. This entropy value is used to dynamically adjust the abnormal warning time limit threshold for this business stage. This represents the total number of equal-width interval groups used to group the duration of stay. In the business phase Below, the duration of stay falls into the first category. The proportion of customers in each group to the total number of customers in that phase. This is a preset small regularization value used to avoid logarithmic inconsistencies when the probability is 0. It is a positive real number much smaller than any non-zero probability value, with a typical value range of [range missing]. The regularization parameter is determined based on the principle of not changing the relative magnitude of the entropy values ​​of each group, and is determined according to the sample size and probability distribution accuracy of the corresponding business stage. Different business stages can adopt a unified default value, or they can be configured independently according to the degree of dispersion of the dwell time distribution of each stage.

[0050] The principle behind the above formula is that when the dwell time of all customers is highly concentrated in a few groups, the probability distribution... The concentration is relatively high, and the state transition entropy is high. A lower value indicates that the customer's work time in this stage is more regular and predictable; conversely, when the customer's stay time is dispersed across multiple groups, the probability distribution is more uniform, and the state transition entropy is higher, indicating that there are greater individual differences in customer behavior and weaker regularity in stage duration.

[0051] Subsequently, the anomaly warning module uses the state transition entropy value. The upper limit of the aforementioned reference interval is dynamically adjusted. The adjustment rule is: when When the time spent in this stage is below the preset first entropy threshold, it indicates that the duration of the stay is highly regular. The system will then tighten the upper limit of the reference interval downwards to improve the sensitivity of anomaly detection. When the value exceeds the preset second entropy threshold, it indicates a high degree of dispersion in the duration of stay during this stage. The system will then widen the upper limit of the reference interval upwards to reduce the false alarm rate. When the value is between the two thresholds, the system maintains the upper limit of the reference interval unchanged. The magnitude of tightening or loosening is related to... The degree of deviation from the threshold is positively correlated. The first entropy threshold and the second entropy threshold are preset classification judgment values, with the first entropy threshold being less than the second entropy threshold; typical values ​​can be set to 1.0 bits and 2.0 bits respectively, specifically calibrated according to the degree of dispersion of historical behavior in the corresponding business stage.

[0052] The anomaly warning module uses the dynamically adjusted upper limit of the reference range as the warning time limit threshold for the current period of this business phase, and writes it into the rule object in the structured object model, replacing the original static threshold. During subsequent anomaly monitoring, when a customer's dwell time in the current phase exceeds this dynamic threshold, the anomaly warning module determines that a timeout anomaly has occurred and generates a corresponding warning task.

[0053] The aforementioned dynamic threshold adjustment process is automatically triggered at a preset time period to ensure that the warning threshold always adapts to the actual situation of recent business operations. Simultaneously, the system executes the above adjustment process independently for different business stages to accommodate the differences in the distribution characteristics of work duration across each stage.

[0054] The mechanism described in this embodiment was compared with a conventional early warning scheme using a fixed time limit threshold. The comparison conditions were set as follows: both methods used the same customer dwell time dataset and the same anomaly detection logic (i.e., both triggered an early warning when the dwell time exceeded the threshold). The difference was that the threshold in the experimental group was dynamically adjusted every 6 hours based on the state transition entropy, while the threshold in the control group was fixed at a preset value (the initial value determined by the interquartile range method). The experiment selected the "data preparation stage," which exhibited a strong regularity in dwell time distribution, and the "progress tracking stage," which showed greater dispersion, for comparison. A concurrent parallel control design was used, ensuring that irrelevant variables such as customer group characteristics, business volume, and operating time periods remained consistent between the two groups; only the generation method of the early warning threshold was a single variable. The comparative results show that during the data preparation phase, the experimental group's false alarm rate decreased by approximately 12% to 15% compared to the control group, the false alarm rate decreased by approximately 6% to 8%, and the overall F1 score improved by approximately 0.18 to 0.22. During the progress tracking phase, the experimental group's false alarm rate decreased by approximately 18% to 22% compared to the control group, the false alarm rate remained essentially the same, and the overall F1 score improved by approximately 0.20 to 0.25. These results are based on statistical analysis of full customer stage transition data over 30 consecutive calendar days. The effective stage transition sample size included in each business stage met the minimum sample requirement for statistical testing. A paired-samples t-test showed that the difference in F1 scores between the two groups was statistically significant (p < 0.05). This effect stems from the quantitative expression ability of state transition entropy to represent the regularity of stage behavior: when the entropy value is low, the system automatically tightens the threshold to reduce false alarms; when the entropy value is high, the system automatically relaxes the threshold to avoid false alarms, ensuring that the warning sensitivity always matches the dynamic distribution of the actual operational rhythm.

Claims

1. A customer manager auxiliary operation system for new energy vehicle consumer finance, characterized in that, include: The model building module is used to create a unified structured object model that includes role objects, process objects, rule objects, knowledge objects, task objects, interface objects, permission objects, and evidence record objects. The objects are associated with each other through predefined relationships. The business stage identification engine is used to acquire multi-source status information of customers in the process of new energy vehicle consumer finance services, and identify the current business stage of the customer based on the multi-source status information and the stage flow logic defined by the process object. The rule constraint processing module is used to match the corresponding bill of materials and service rules from the rule object according to the business stage, and output the matching result through the manual confirmation interface for the account manager to confirm and then take effect. The task generation module is used to generate customer manager auxiliary tasks based on the manually confirmed list of materials and service rules. The anomaly warning module is used to monitor at least one of the following: customer status, material submission status, progress status, and store collaboration status, and generate an warning task when an anomaly is detected. The service record module is used to associate the customer manager's handling process of the auxiliary task and the early warning task, the manual connection operation record and the customer feedback information with the evidence record object for storage; The feedback optimization engine is used to update the bill of materials matching conditions and / or anomaly judgment conditions in the rule object based on the historical processing results and customer feedback information stored in the evidence record object, and to synchronize the updated rule object to the structured object model.

2. The system according to claim 1, characterized in that: The model building module uses a metadata-driven approach to define the attribute fields of each object and the relationships between objects. The process object contains nested reference identifiers for rule objects and trigger condition identifiers for task objects. The task object contains nested association identifiers for evidence record objects.

3. The system according to claim 1, characterized in that: The matching results output by the rule constraint processing module are transmitted to the account manager terminal via the manual confirmation interface. After receiving the confirmation instruction returned by the account manager terminal, the rule constraint processing module transmits the confirmed matching results to the task generation module. The rule constraint processing module also records the modification operations on the matching results during the manual confirmation process and associates the modification operation records with the evidence record object.

4. The system according to claim 1, characterized in that: The anomaly warning module adopts a multi-threshold hierarchical warning mechanism, which is configured with different anomaly judgment logic and corresponding warning levels for different types of anomaly states. When an anomaly state is determined, the priority and method of pushing the warning task are determined according to the anomaly type and warning level.

5. The system according to claim 1, characterized in that: The feedback optimization engine includes a rule update unit and a model synchronization unit. The rule update unit performs statistical analysis on the processing time data, customer response rate data, and anomaly detection rate data of multiple historical tasks stored in the evidence record object, and generates rule update parameters based on the statistical analysis results. The model synchronization unit maps the rule update parameters to attribute value adjustment instructions for corresponding rule entries in the rule object, and synchronously updates the adjusted rule object to the structured object model.

6. The system according to claim 1, characterized in that: Each rule entry in the rule object is configured with a variable weight value; the feedback optimization engine calculates the processing timeliness index, customer response effectiveness index, and abnormal trigger frequency index of the historical tasks associated with each rule entry based on the historical task processing records stored in the evidence record object, and performs weighted fusion on the three indicators after normalization to generate the updated weight value of each rule entry, and writes the updated weight value into the rule object to replace the original weight value. The task generation module determines the generation priority and push order of the generated auxiliary tasks based on the updated weight value, and the rule constraint processing module determines the display order of each material item in the material list based on the updated weight value.

7. The system according to claim 1, characterized in that: The anomaly warning module is equipped with a dynamic threshold adjustment unit. The dynamic threshold adjustment unit extracts the historical stage dwell time data of all customers currently in the same business stage from the evidence record object, divides the dwell time into several groups and calculates the proportion of the number of customers in each group to the total number of customers, and calculates the state transition entropy value of the business stage based on the proportion. The dynamic threshold adjustment unit tightens or loosens the time limit threshold currently used in the business stage according to the level of the state transition entropy value, and uses the corrected time limit threshold as the warning judgment benchmark for the business stage.

8. The system according to claim 1, characterized in that: The rule constraint processing module is configured with a subset of business rules, a subset of compliance rules, a subset of product rules, and a subset of process rules. The subset of business rules defines the standard operating procedures for each business stage. The subset of compliance rules defines the compliance boundary conditions for financial services. The subset of product rules defines the service templates for different types of financial products. The subset of process rules defines the flow conditions and flow order between business stages. The rule constraint processing module extracts the corresponding rule entries from each subset of rules according to the current business stage of the customer and combines them to form a set of rule constraints that apply to the current business stage.

9. The system according to claim 1, characterized in that: The auxiliary tasks generated by the task generation module are configured with different execution templates according to the task type. The execution templates include task description templates, script templates, and operation guidance templates. After the task generation module pushes the generated auxiliary tasks to the customer manager terminal, it triggers the execution of the corresponding subsequent process nodes in the process object based on the task processing status change events recorded in the evidence recording object.

10. A method for assisting account managers in new energy vehicle consumer finance, characterized in that, The method is implemented based on the system described in any one of claims 1 to 9, and includes: Establish a unified structured object model that includes role objects, process objects, rule objects, knowledge objects, task objects, interface objects, permission objects, and evidence record objects, and define the relationships between these objects; Obtain multi-source status information of customers during the new energy vehicle consumer finance service process, and identify the current business stage of the customer based on the multi-source status information and the stage flow logic defined by the process object; Match the corresponding bill of materials and service rules from the rule object according to the business stage, and output the matching results through the manual confirmation interface for the account manager to confirm. Generate customer manager support tasks based on the manually confirmed materials list and service rules; Monitor at least one of the following: customer status, material submission status, progress status, and store collaboration status, and generate an alert task when an abnormal status is detected; The process of the account manager handling the auxiliary tasks and the early warning tasks, the records of manual follow-up operations, and the customer feedback information are associated with the evidence record object and stored. Based on the historical processing results and customer feedback information stored in the evidence record object, update the bill of materials matching conditions and / or anomaly judgment conditions in the rule object, and synchronize the updated rule object to the structured object model.

Images