Enterprise information unified management system for multi-source heterogeneous data

Abstract

The invention discloses an enterprise information unified management system for multi-source heterogeneous data, and belongs to the field of data processing. Comprising a data processing module, a client matrix construction module, a model training and analysis output module, a retention strategy providing module, a closed-loop optimization module and a man-machine interaction module, all the modules work cooperatively, the crossing from static automation to dynamic intelligence is realized, and finally, the operation efficiency is improved while the real-time performance of the system is improved. And the accuracy of the retention strategy and the sustainability of the customer relationship are ensured.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of data processing, and particularly relates to an enterprise information unified management system for multi-source heterogeneous data. BACKGROUND

[0002] In the daily operation of enterprises, customer loss is a common challenge. However, the current way of many enterprises to deal with lost customers often excessively relies on the personal experience and relationship maintenance means of front-line responsible persons, and lacks systematic and intelligent analysis support. At the same time, the customer data accumulated in the enterprise is scattered in multiple heterogeneous systems (such as structured transaction databases, unstructured customer service work orders and communication records, real-time user behavior flow data, etc.), and these data islands lead to fragmented customer information and difficulty in association, significantly lengthening the query time and increasing the analysis cost, and easily causing data inconsistency. Finally, the management layer is difficult to accurately identify the deep causes and common rules of customer loss, and cannot effectively develop preventive strategies, but only can passively carry out post-repair, which needs to be improved. SUMMARY

[0003] Therefore, in view of the above problems, the present application provides an enterprise information unified management system for multi-source heterogeneous data.

[0004] In a first aspect, the present application provides an enterprise information unified management system for multi-source heterogeneous data, which comprises: Module one, a data processing module, for converting the collected multi-source heterogeneous original data related to customers into a standardized data set with customer ID as the core, unified and regular, which can be directly used for analysis, through standardization rules and processing technology; The data processing module comprises: A data collection submodule for collecting customer-related data; the related data includes internal structured data, internal unstructured data and external data; the external data includes external competitive intelligence, industry intelligence and third-party data; A unified standard submodule for defining the core entity with customer ID as the primary key and its basic attributes, clearly defining the calculation logic of analysis-specific indicators, and establishing mapping rules from cross-system fields to prediction models, to generate a standardized customer data model document that can be analyzed across systems; A data processing submodule for cleaning, labeling and calculating various types of original data to generate a standardized data set with customer ID as the core; Module two, a customer feature matrix construction module, for processing the standardized data set into a multi-dimensional, quantifiable and directly usable modeling customer feature matrix; Module three, model training and analysis output module, is used for customer churn prediction by using customer feature matrix, and global and personalized attribution explanation is generated; The model training and analysis output module comprises: A training prediction model submodule is configured to receive customer feature matrix data, and output a churn probability corresponding to each customer ID based on analysis; An explanation submodule is configured to receive the output result of the training prediction model, and derive a key driver factor ranking report applicable to all customers; and apply SHAP, LIME and other explainable AI technologies to generate a personalized and visual churn reason attribution report for each high-risk customer; Module four, retention strategy providing module, is used for converting churn prediction and attribution into executable, simulative and economically efficient personalized customer retention strategies; The retention strategy providing module comprises: A strategy rule base submodule is configured to store a mapping relationship between "strategy-churn reason", and match candidate strategies in the knowledge base according to the "churn reason" output by the explanation submodule; A strategy simulator submodule is configured to receive candidate strategies output from the strategy rule base, and predict the improvement effect of different strategies on the target customer retention probability and the expected cost benefit before actually inputting resources, and predict the potential return of different strategies; An intelligent strategy matching and output submodule is configured to sort all candidate strategies according to the effectiveness and cost benefit of the strategies, and output a priority strategy list with benefit prediction and a strategy execution mode; Module five, closed-loop optimization module, is used for collecting effect data and quantitatively evaluating, and realizing dynamic regulation and continuous optimization of the retention strategy system; The closed-loop optimization module comprises: A strategy effect collection submodule is configured to collect and integrate the execution results of the retention strategies; A strategy effect evaluation submodule is configured to analyze the fatigue attenuation curve of the strategy effect by combining the strategy effect success rate and the actual return, and give a quantitative strategy effect score, which is fed back to the "dynamic regulation" module and simultaneously copied to the human-computer interaction module; A dynamic regulation submodule is configured to receive the strategy effect score data from the strategy effect evaluation submodule, optimize the strategy rule base and the intelligent strategy output module, and retrain the prediction model; Module six, human-computer interaction module, is an interface for business experts or management personnel to interact with the AI system, allowing them to question or recognize the model features and retention strategies, and injecting the business expert experience into the closed-loop optimization process.

[0005] Further, the model training and analysis output module further comprises a feedback submodule, configured to receive the attribution report output by the explanation submodule and the feedback data of the human-computer interaction module, analyze the negative feedback, obtain a labeled feedback mechanism data set, identify and generate a system blind spot group label and a high misjudgment feature from the labeled feedback mechanism data set, and feed back the system blind spot group label and the high misjudgment feature to the training prediction model submodule and the explanation submodule.

[0006] Further, the customer feature matrix construction module further comprises a feature optimization submodule, configured to receive the blind spot group label and the high misjudgment feature fed back by the feedback submodule, and optimize the customer feature matrix model.

[0007] Further, the retention strategy providing module further comprises a competitor response submodule, configured to collect the probability of competitors taking targeted measures and the time data of the competitors taking the targeted measures under a preset probability, and feed back the data to the strategy simulator submodule, so as to optimize the prediction method of the strategy simulator submodule.

[0008] Further, the strategy rule base submodule is further configured to receive the labeled feedback record data set of the feedback submodule, and adjust the mapping relationship between the strategy and the loss reason.

[0009] Further, the closed-loop optimization module further comprises a fatigue monitoring submodule, configured to monitor and quantify the contact frequency and response state of customers to different strategies in real time, and feed the data to the strategy effect collection submodule, so as to construct a strategy fatigue dynamic evaluation index.

[0010] Further, the closed-loop optimization module further comprises a touch attribution model submodule, configured to apply an attribution model to quantify the contribution of each touch to the final retention success, and output the data to the strategy effect evaluation submodule.

[0011] Further, the dynamic regulation submodule is further configured to receive the fatigue data of the fatigue monitoring submodule, and automatically perform disabling or downgrading operations on a strategy with low efficiency or excessive touch.

[0012] Further, the human-computer interaction module is further configured to synchronously copy the evaluation report of the strategy effect evaluation module to a business expert or a management personnel, receive “question” or “acknowledgment” information proposed by the business expert or the management personnel, and feed back the question information to the feedback submodule.

[0013] Further, the human-computer interaction module, the business expert or the management personnel proposes modification or deletion of an unreasonable strategy in the strategy rule base submodule, and improves the output loss prediction accuracy.

[0014] Compared with the prior art, the present application has the following beneficial effects: 1. The present application constructs a feedback submodule, according to the churn attribution report and the negative feedback output by the human-computer interaction module, outputs the blind spot population and high misjudgment characteristics to the customer characteristic matrix model, so that the customer characteristics output by the customer characteristic matrix model are more accurate, thereby indirectly improving the output accuracy of the prediction model and the pertinence and success rate of the retention strategy; in addition, the feedback submodule analyzes the high misjudgment characteristics and blind spot population data output by the human-computer interaction module, feeds back the data information related to model calculation to the prediction model and the strategy rule library, optimizes the calculation weight of the prediction model and the rule library from multiple dimensions, realizes the full-link and multiple closed-loop optimization from the characteristic matrix to the strategy decision.

[0015] 2. The present application constructs a dynamic regulation submodule, comprehensively uses the execution results of the retention strategy and the negative feedback data of the fatigue monitoring to correct the error loss reasons of the explanation submodule, iteratively update the strategy rule library and adjust the execution mode, effectively optimizes the adaptability and accuracy of the two core modules of model training and retention strategy; the fatigue monitoring module not only directly affects the dynamic regulation module, but also indirectly affects the dynamic regulation module by inputting the customer fatigue degree into the strategy collection module and comprehensively using other customer behavior score results through the strategy effect evaluation module, which enables the system to continuously adapt to the changes of customer behavior patterns and market dynamics, and the prediction accuracy of the model and the effectiveness of the strategy are continuously improved over time.

[0016] 3. The present application introduces a human-computer interaction module, provides visual attribution explanation and strategy intervention channels, constructs a human-computer collaborative decision mechanism, improves the credibility and decision transparency of complex system models, and promotes the precise intervention and global control of management personnel on key decisions. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the data processing module of the present application; Figure 2 The figure is a schematic diagram of the customer characteristic matrix model construction module of the present application; Figure 3 The figure is a schematic diagram of the model training and analysis output module of the present application; Figure 4 The figure is a schematic diagram of the retention strategy providing module of the present application; Figure 5 The figure is a schematic diagram of the closed-loop optimization module of the present application; Figure 6 The figure is a schematic diagram of the human-computer interaction module of the present application; Figure 7 The figure is a schematic diagram of the feedback submodule of the present application; Figure 8 The figure is a dynamic regulation module of the present application; Figure 9A schematic diagram of an enterprise management system for multi-source heterogeneous data according to the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0019] It can be understood that the terms "first", "second" and the like used herein can be used to describe various elements, but unless specifically stated, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of the present application, the first xx script can be referred to as the second xx script, and similarly, the second xx script can be referred to as the first xx script.

[0020] Embodiment

[0021] One kind of enterprise information unified management system for multi-source heterogeneous data, comprising: Module 1, data processing module, such as Figure 1 , comprising (1) data acquisition submodule, comprehensively, automatically and in real time collect all data related to enterprise customers; the data includes internal structured data, internal unstructured data and external data; the external data includes external competitive intelligence, industry intelligence and third-party data; (2) unified standard submodule, for defining the core entity (customer, product, etc.) with customer ID as the primary key and its basic attributes (such as customer basic information, product type), clearly defining the calculation logic of analysis special indicators (for example, the active decay rate = the number of login times in the last 30 days / the average monthly login times in history), and establishing the mapping rules of cross-system fields to prediction models (such as mapping the "complaint level" of the customer service system to the "negative interaction intensity value"), generating standardized, cross-system analysis customer data model documents (including entity relationship diagram and mapping rules); (3) data processing submodule, for structured data (such as transaction records), cleaning missing values (focus on completing missing customer ID) and unifying date format, generating structured data sets with customer ID as the primary key; for unstructured data (such as customer service work order), apply NLP (natural language processing) to extract key topic labels (such as "response delay", "product quality") and sentiment orientation, and generate label data sets with customer ID as the key after associating customer ID; for real-time data (such as behavior log), aggregate and calculate basic time series indicators (such as the frequency of function use rate in the last 7 days) according to customer ID, and generate real-time indicator data sets with customer ID as the key; Module 2, customer feature matrix construction module, such asFigure 2 , comprising (1) Customer feature matrix model, based on structured data set, label data set, real-time index data set, calculate basic features (such as renewal rate, monthly average ticket number), time series / behavior features (such as active degree week change rate), interaction / emotion features (such as negative emotion ticket proportion) and cross / compound features (such as complaint active degree decline value = average active degree before complaint- average active degree after complaint, service risk score is obtained by negative emotion ticket proportion and response overtime number in the past 3 months) fused with multi-source information, using the standardized data set output by the data processing module and the optimized data output by the feature optimization sub-module, a structured customer-feature matrix is constructed; (2) Feature optimization sub-module, used to receive feedback information from feedback sub-module in module 3, and adjust the feature matrix, for example: feedback "high misjudgment feature" list and "potential blind spot population" label, create new compound features, introduce external data sources, and engineer existing features; Module 3, model training and analysis output module, such as Figure 3 , comprising (1) Training prediction model Receive customer feature matrix data, based on analysis of customer feature matrix, output the churn probability corresponding to each customer ID; at the same time, receive data information related to model calculation fed back by feedback sub-module, real-time optimize training prediction model, for example, according to the analysis results and attribution report of "potential blind spot population", reduce the weight of this population in the churn probability calculation model; (2) Explanation sub-module Receive the output results of the training prediction model, and obtain the key driver factor ranking report applicable to all customers; apply SHAP, LIME and other explainable AI technologies to generate individualized and visualized churn reason attribution report for each high-risk customer; at the same time, receive feedback information from feedback sub-module and customer behavior data fed back by dynamic control module, real-time optimize the explanation model of explanation sub-module, for example, if the score of the evaluation report of the retention strategy selected based on a certain reason is < 50, reduce the weight or delete the reason in the explanation model; (3) Feedback sub-module (such as Figure 7 ), Receive feedback data of attribution report from human-computer interaction module, preliminarily and automatically analyze the collected negative feedback, for example: cluster analysis of customers receiving negative feedback to quickly identify potential population blind spots; analyze which model features are marked as key driver factors but denied by business in negative feedback to find out "high misjudgment features"; And the feedback to the information related to the operation of each module is synchronized to optimize the interpretation sub-module, train the prediction model, the feature optimization sub-module and the strategy rule base; Module 4, retention strategy providing module, such as Figure 4 , including (1) Strategy rule base sub-module According to the "churn reason" output by the interpretation sub-module, match the candidate strategy in the knowledge base; the strategy rule base receives the data related to the strategy feedback by the dynamic regulation module, the human-computer interaction module and the feedback sub-module, constantly optimizes the mapping relationship of "strategy-churn reason" and adjusts the strategy weight coefficient and updates the strategy rule base; (2) Competitor response sub-module Through network crawler and API interface, continuously obtain competitor / competitor dynamics (such as product update, pricing adjustment), and use knowledge graph technology to build competitor relationship network, collect the probability and time of competitors to launch targeted measures, and feed back the intelligence data to the strategy simulator; (3) Strategy simulator sub-module Receive the candidate strategy output from the strategy rule base and the data output from the competitor response module, before real investment, use causal inference method to predict the promotion effect of different strategies on target customer retention probability and expected cost benefit, and predict the potential return on investment (ROI) of different strategies; (4) Intelligent strategy matching and output sub-module Integrate the effectiveness (Uplift) and cost benefit (ROI) of the strategy, as well as the feedback data of the strategy execution mode output by the dynamic regulation module, sort all candidate strategies and output a priority strategy list with benefit prediction and strategy execution mode; Module 5, closed-loop optimization module, such as Figure 5 , including (1) Fatigue monitoring sub-module Used to increase the contact frequency of retention strategy and track the history response record, real-time update "customer strategy contact frequency" and "negative feedback record", for example: when detecting strategy fatigue (such as multiple similar contacts in a short period of time and no positive feedback), automatically suspend the invalid strategy, trigger alternative solution (such as switching to more gentle artificial care follow-up), and output data to the strategy effect collection module and "dynamic regulation" module; (2) Strategy effect collection module Used to collect and integrate the execution results of retention strategy and customer strategy contact frequency; (3) Touch attribution model sub-module Apply attribution model to quantify the contribution of each touch to the final retention success, and output data to the strategy effect evaluation module; (4) Strategy Effect Evaluation Module In combination with the Strategy Effect Collection Module and the Contact Attribution Model, the success rate, actual return on investment (ROI) of the strategy is calculated, the fatigue decay curve of the strategy effect is analyzed, and a quantitative strategy effect score is given. The score is fed back to the Dynamic Regulation Module and simultaneously copied to the Human-Machine Interaction Module; (5) Dynamic Regulation Sub-Module (such as Figure 8 ) Receiving the strategy effect score and strategy fatigue feedback data from Module 5, the strategy weight coefficient of the strategy rule base and the execution mode of the intelligent strategy matching and output module are adjusted, and the error loss reason is helped to be eliminated. When the effectiveness score of a certain strategy continuously falls below the threshold value or the fatigue degree rises sharply, the system can automatically disable it, automatically reduce the weight or exclude similar strategies, and also change the strategy execution mode; Module 6, Human-Machine Interaction Module, such as Figure 6 Receiving the evaluation report of the Strategy Effect Evaluation Module, a decision-making interface is provided for business personnel. For example, business personnel can provide 'approval' or 'question' feedback on the attribution report generated by the Model Training and Analysis Output Module on this interface (this feedback is received and processed by the feedback sub-module of Module 3), and can perform operations such as auditing, approving, modifying, or vetoing on the retention strategies recommended by the strategy rule base in the Retention Strategy Providing Module.

[0022] II. Role of System Modules 1. Data Processing Module The data collection module is responsible for comprehensively, automatically, and in real time collecting all data related to enterprise customers, ensuring the comprehensiveness and timeliness of the information source. Its data range covers structured data (such as transaction records, user information) and unstructured data (such as customer service work orders, communication records) within the enterprise, as well as external data such as external competitive intelligence, industry intelligence, and third-party data, building this comprehensive data collection network.

[0023] The unified standard module aims to define a unified language and rules for customer-related data from different systems (such as customer service, product, and transaction systems) within the enterprise. By enforcing the use of customer ID as the primary key, it ensures that all data can be accurately associated with specific customers. It defines core entities and basic attributes (such as customer basic information and product types) to provide a basic framework. It clearly defines the calculation logic of analysis-specific indicators (such as the active decay rate) to ensure consistent and comparable indicator calculations. It establishes cross-system field mapping rules (such as mapping "complaint level" to "negative interaction intensity value") to convert raw data from different systems into standardized numerical features that can be understood by prediction models. The final generated customer data model document (including entity relationship diagrams and mapping rules) is the standard basis for data integration and analysis throughout the system, ensuring data consistency and interpretability in subsequent content.

[0024] The data processing module is to realize the unification, standardization and availability of multi-source heterogeneous data. The core goal is to process various data types existing in reality (structured transaction records, unstructured customer service work orders, real-time behavior logs), and convert them into a format specification dataset with customer ID as the unified key required for subsequent construction of feature matrix. For structured data, the focus is on cleaning (completing key missing values such as customer ID) and standardization (unifying date format, etc.). For unstructured data, NLP is applied to extract valuable and standardized information (topic labels, sentiment orientation), and convert it into structured label data. For real-time data, preliminary time window-based aggregation calculations (such as 7-day usage frequency decline rate) are performed to generate lightweight time series indicators. The key goal is to convert raw, chaotic data into usable, clean, customer-centric structured / semi-structured data.

[0025] 2. Customer Feature Matrix Construction Module The customer feature matrix construction module converts various standardized datasets into interpretable and usable customer features for prediction. The customer feature matrix model constructs a structured matrix through feature engineering. The design purposes include: feature enrichment, not only calculating basic features (such as renewal rate), but also deeply mining time series behavior features (such as week-on-week change), interaction sentiment features (such as negative work order proportion), and especially emphasizing the fusion of cross / compound features of multi-source information (such as "complaint post activity decline value", "service risk score"). These compound features can more deeply and comprehensively reveal customer status and potential churn drivers. At the same time, it receives feedback data from the feature optimizer to create new compound features, adjusts the matrix, prepares for the input of the prediction model, and generates a matrix with clear structure, complete features, and numerical values.

[0026] In addition, the customer feature matrix construction module also includes a feature optimization submodule, which receives analysis results from the feedback submodule, including a "high misjudgment feature" list and a "potential blind spot population" label. Based on these feedback, it dynamically adjusts and optimizes the feature matrix. For example, for "high misjudgment features" frequently questioned by business personnel, the submodule will try to create new, more business-meaningful compound features to replace or supplement them; for the identified "potential blind spot population", it will introduce new external data sources or engineer existing features (for example, build exclusive feature derivation rules for specific populations) to fill the cognitive gaps of the model in that group. Through this continuous optimization, the feature system of the system can more and more accurately reflect the real business logic.

[0027] 3. Model Training and Analysis Output Module The model training and analysis output module predicts customer churn risk based on feature data and reveals the key reasons. The model training and analysis output module includes a training prediction model, an explanation submodule, and a feedback submodule. The module trains the prediction model using the customer-feature matrix constructed by the customer-feature matrix construction module. Based on the predicted churn probability value, the risk is quantified, and a quantified churn risk score (probability) is provided for each customer. The explanation submodule realizes explainability-driven key driver factor ranking reports to inform business personnel of the reasons for customer churn. The explanation submodule inputs the attribution report to the feedback submodule, which provides "approval" or "question" feedback to the explanation submodule, the training prediction model, the feature optimization submodule of the customer-feature matrix construction module, and the strategy rule base in the retention strategy module.

[0028] The prediction model uses a supervised learning classification algorithm (such as XGBoost, LightGBM, or random forest) to use whether or not to churn in customer historical data as a label (binary classification target variable). By cross-validation, the hyperparameters (such as tree depth, learning rate) are optimized, and the prediction model learns the complex nonlinear relationship between features and churn risk.

[0029] The explanation submodule deeply excavates the logic behind the prediction based on the model output churn probability. It first derives a key driver factor ranking report that is generally applicable to all customer groups, identifying the main factors affecting churn from a holistic perspective. More importantly, it applies SHAP, LIME, and other explainable AI technologies to generate personalized, visualized churn reason attribution reports for each identified high-risk customer.

[0030] The feedback submodule receives feedback data from the human-computer interaction module and performs automated analysis to quickly locate system blind spots and feed the analysis results related to model calculation data to the prediction model and strategy rule base. For example, it performs cluster analysis on customers who have received negative feedback to quickly identify "potential blind spot groups" that the model fails to understand well. At the same time, it analyzes which model features are marked as key drivers but are explicitly denied by business personnel, thereby screening out a "high misjudgment feature" list.

[0031] In the explainable AI technology, SHAP (SHapley Additive exPlanations) value calculation: based on the trained prediction model, SHAP algorithm is run for each high-risk customer to quantify the contribution value of each feature to the current churn probability (positive value increases churn risk, negative value reduces risk).

[0032] LIME (Local Interpretable Model-agnostic Explanations) (Optional Supplement): For complex cases, LIME is used to build a local linear model around customer characteristics to explain the causal relationship between characteristics and churn probability (e.g., "When response latency > 3 hours, churn probability increases by 35%").

[0033] Generate an attribution report, sort key factors by absolute SHAP value, and extract the top 3-5 core features that positively drive churn (such as "response delay contribution +42%)".

[0034] The prediction model outputs two results: Churn probability prediction: The prediction model outputs a churn probability value between 0 and 1 for each customer (e.g., "Customer A001 churn probability: 0.85"). Key driver factor analysis utilizes interpretability techniques (such as SHAP values ​​and feature importance ranking) to analyze the decision logic of the predictive model.

[0035] 4. Retention Strategy Provision Module The retention strategy module analyzes churn attribution reports and key driver ranking reports to develop and push customized retention strategies (e.g., "Customer ID: A001, follow up within 24 hours, suggested solution: free extended warranty"). This module consists of the following core sub-modules: (1) Strategy rule base: As the core strategy knowledge base of the system, it maintains a dynamic mapping relationship between "churn characteristics and response strategies". It intelligently matches the initial strategy set based on the attribution results output by the model, and continuously iterates and optimizes the mapping rules based on the feedback data of the feedback sub-module, the closed-loop dynamic control module and the human-computer interaction module, so as to realize the self-evolution of strategy rules; (2) Competitor Response Submodule: Based on competitive intelligence and historical interaction data, simulate the targeted countermeasures that major competitors may take, introduce competitive environment variables into the strategy simulator, and transform external market competitive intelligence into quantifiable risk warning and strategy adjustment signals within the enterprise, thereby realizing competitive risk management from passive defense to proactive response. (3) Strategy Simulator: Simulates and pre-evaluates intervention measures before strategy deployment. Applying causal inference models, it quantifies and predicts the uplift effect and expected return on investment (ROI) of different strategies on the retention probability of target customers, providing forward-looking data support for resource allocation; (4) Intelligent strategy matching and output module: serving as the strategy selection and decision engine. It integrates the expected effectiveness and cost-effectiveness of the strategy, performs multi-objective optimization and ranking of candidate strategies, and generates a priority strategy list with quantitative profit prediction to support refined decision-making; 5. Closed-loop optimization module The closed-loop optimization module is used to collect and integrate the execution results (success / failure) of retention strategies and subsequent key behavior indicators (such as repeat purchase rate after retention, change in activity), analyze the performance differences of prediction models for different customer groups, and use these feedback data to optimize retention strategy rules (such as strategy fatigue threshold, touchpoint combination weight), execution strategy, and prediction models.

[0036] The closed-loop optimization module, as the core of the system's self-evolution, realizes continuous optimization through the following mechanisms: (1) Fatigue monitoring submodule: used to track the increase in retention strategy contact frequency and historical response records. When it is detected that a customer has received multiple similar retention strategies in a short period of time (such as 3 times of coupon push in 30 days) and has no positive feedback (such as not triggering target conversion behavior, not generating positive touch interaction, or subsequent customer service communication / comment analysis showing no attitude improvement), the system automatically suspends the strategy type and marks strategy fatigue, triggering alternative solutions (such as switching to manual care follow-up), to avoid over-disturbing and causing customer experience deterioration; (2) Strategy effect collection module: systematically collects and integrates direct results (such as conversion status) and long-term aftereffect indicators (such as repeat purchase rate) of strategy execution, building a complete strategy effect tracking dataset to provide a data basis for effect analysis; (3) Touchpoint attribution model submodule: used to track customer interaction paths at different retention touchpoints (such as push notifications, outbound calls, marketing emails, APP pop-ups) when executing retention strategies (by embedding UTM parameters and other methods), and apply attribution models (such as Markov chain model) to quantify the contribution of each touchpoint to the final retention success, thereby continuously optimizing touchpoint combination strategies (for example, analyzing that the combination strategy of "sending a warning SMS and then making a dedicated customer service outbound call within 48 hours" has the most significant success rate improvement); (4) Strategy effect evaluation module: performs multi-dimensional performance analysis on collected strategy data, accurately calculates key indicators such as strategy success rate and actual ROI, models the decay law of strategy effectiveness, and outputs a quantitative strategy effect score, providing a core basis for iterative optimization of the strategy library; (5) Dynamic regulation submodule: realizes real-time monitoring and self-adaptive regulation of strategy during execution. By continuously receiving strategy effectiveness scores and fatigue indicators, the system automatically performs disabling, downgrading, and other operations on strategies with low effectiveness or excessive touch, ensuring the agility and accuracy of the strategy library.

[0037] Human-computer interaction module Based on the analysis of the effect evaluation report output by the strategy effect evaluation module, the attribution report of the feedback submodule is questioned or recognized, so that it is fed back to the customer feature matrix model with high misjudgment characteristics and blind spot population, and unreasonable strategies in the strategy rule library in the retention strategy providing module are modified or deleted, ensuring that the flow loss prediction and intervention strategy output by the system not only meet statistical significance, but also have business feasibility and execution efficiency.

[0038] The system realizes intelligent operation of the whole process from customer insight, strategy generation, simulation deduction, intelligent matching to effect evaluation and continuous optimization through deep cooperation between modules and data closed loop, ensuring that the retention strategy always maintains the best adaptability and effectiveness.

[0039] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise stated herein, the execution of these steps has no strict sequence limitation, and these steps can be executed in other order. Moreover, at least part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or sub-steps or stages of other steps.

[0040] The technical features of the above-described embodiments can be combined in any way. In order to make the description concise, all possible combinations of the technical features in the above-described embodiments are not described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0041] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the present application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

[0042] The above-described only the preferred embodiments of the present application, and does not limit the present application, any modification, equivalent replacement and improvement within the spirit and principles of the present application, should be included in the protection scope of the present application.

[0043] Furthermore, it should be understood that although the specification is described in terms of embodiments, not every embodiment includes every feature described. The specification can include implicit combinations of explicitly mentioned features and / or explicit combinations of implicitely mentioned features. Each embodiment depends on the explicit combinations of features and / or the implicit combinations of features made specifically within that embodiment, and each such embodiment can be combined with every other such embodiment to create further embodiments.

Claims

1. A multi-source heterogeneous data-oriented enterprise information unified management system, characterized in that, The application relates to a customer churn prediction and retention strategy system, which comprises the following modules: Module 1, a data processing module, is used for converting collected multi-source heterogeneous original data related to customers into a standardized data set with a customer ID as a core, a unified structure and direct analysis by using standardization rules and processing techniques; Module 2, a customer feature matrix construction module, is used for processing the standardized data set into a customer feature matrix with multiple dimensions, quantification and direct modeling; Module 3, a model training and analysis output module, is used for customer churn prediction by using the customer feature matrix and generating global and personalized attribution explanations; The model training and analysis output module comprises the following: A training prediction model submodule is used for receiving customer feature matrix data and outputting a churn probability corresponding to each customer ID; An explanation submodule is used for receiving the output result of the training prediction model, obtaining a key driving factor ranking report applicable to all customers and generating a personalized and visual churn reason attribution report for each high-risk customer; Module 4, a retention strategy providing module, is used for converting churn prediction and attribution into executable, simulative and economically efficient personalized customer retention strategies; The retention strategy providing module comprises the following: A strategy rule base submodule is used for storing a mapping relationship between a strategy and a churn reason and matching candidate strategies in a knowledge base according to the churn reason output by the explanation submodule; A strategy simulator submodule is used for receiving the candidate strategies output from the strategy rule base, predicting the potential return of different strategies; An intelligent strategy matching and output submodule is used for comprehensively considering the effectiveness and cost-effectiveness of the strategies, ranking all candidate strategies and outputting a priority strategy list with benefit prediction and a strategy execution mode; Module 5, a closed-loop optimization module, is used for collecting effect data and quantitatively evaluating, realizing dynamic regulation and continuous optimization of the retention strategy system; The closed-loop optimization module comprises the following: A strategy effect collection submodule is used for collecting and integrating the execution results of the retention strategies; A strategy effect evaluation submodule is used for combining the strategy effect success rate and actual return, giving a quantitative strategy effect score, feeding back the score to the dynamic regulation module and synchronously copying the score to the human-computer interaction module; A dynamic regulation submodule is used for receiving the strategy effect score data from the strategy effect evaluation submodule, optimizing the strategy rule base and the intelligent strategy output module and retraining the prediction model; Module 6, a human-computer interaction module, is used as an interface for business experts or managers to interact with the AI system, allows the business experts to question or recognize the model features and the retention strategies, and injects the business expert experience into the closed-loop optimization process.

2. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The model training and analysis output module further comprises a feedback submodule, which is used for receiving the attribution report output by the explanation submodule and the feedback data of the human-computer interaction module, analyzing the negative feedback, obtaining a marked feedback mechanism data set, identifying and generating system blind spot population labels and high misjudgment features, and feeding back the data information related to model calculation to the training prediction model submodule and the explanation submodule. 3.The enterprise information unified management system facing multi-source heterogeneous data according to claim 2, characterized in that, The customer feature matrix construction module further comprises a feature optimization submodule for receiving the blind spot population label and high misjudgment feature fed back by the feedback submodule, and optimizing the customer feature matrix model.

4. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The retention strategy providing module further comprises a competitor response submodule for collecting the probability of competitors taking targeted measures and the time data of pushing out the targeted measures under the preset probability, and feeding the data to the strategy simulator submodule to optimize the prediction method of the strategy simulator submodule.

5. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The strategy rule base submodule is further configured to receive the marked data information related to the strategy calculation model from the feedback submodule, and adjust the mapping relationship of "strategy-churn reason".

6. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The closed-loop optimization module further comprises a fatigue monitoring submodule for monitoring and quantifying the contact frequency and response state of customers to different strategies in real time, and feeding the data to the strategy effect collection submodule to construct a dynamic evaluation index of strategy fatigue.

7. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The closed-loop optimization module further comprises a touch attribution model submodule for applying an attribution model to quantify the contribution of each touch to the final retention success, and outputting the data to the strategy effect evaluation submodule.

8. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The dynamic regulation submodule is further configured to receive the fatigue data of the fatigue monitoring submodule, and automatically perform disabling or downgrading operation on the strategies with low efficiency or excessive touch.

9. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The human-computer interaction module is further configured to synchronize the evaluation report of the strategy effect evaluation module to business experts or management personnel, receive the "question" or "approval" information proposed by the business experts or management personnel, and feed the question information to the feedback submodule.

10. The enterprise information unified management system for multi-source heterogeneous data according to claim 1, characterized in that, The human-computer interaction module is further configured to enable the business experts or management personnel to modify or delete the unreasonable strategies in the strategy rule base submodule, and improve the output accuracy of the churn prediction.

Images