Search

CN-121998649-A - Customer asset intelligent management system based on layered dynamic driving model

CN121998649ACN 121998649 ACN121998649 ACN 121998649ACN-121998649-A

Abstract

The invention discloses a customer asset intelligent management system based on a layered dynamic driving model, which relates to the technical field of business data processing and comprises a unified data center, a layered dynamic driving engine, a strategy driving unit, a feedback optimizing unit, a loss early warning and intervention unit and a quantitative stripping decision unit. The invention realizes the dynamic and automatic management of the customer assets, and the system can automatically execute the accurate strategy to perform intervention, upgrading or stripping, thereby remarkably improving the overall value and the operation efficiency of the customer assets and improving the management efficiency of the customer assets.

Inventors

  • ZHENG HAO

Assignees

  • 山东财经大学

Dates

Publication Date
20260508
Application Date
20260130

Claims (10)

  1. 1.A customer asset intelligent management system based on a layered dynamic driving model is characterized by comprising the following modules: the unified data center is used for accessing and fusing original customer data from a transaction system, an interaction platform and an external data source in real time, and outputting a dynamic feature vector which takes a unique customer identifier as a key value and comprises a time sequence behavior label and a statistical index by executing data cleaning, entity association and feature engineering; The hierarchical dynamic driving engine is used for carrying out dynamic value evaluation and hierarchical division on customers and predicting future state changes and comprises a cooperative computing unit, a dynamic mapping unit and a state deduction unit; The collaborative computing unit is used for parallelly executing customer lifetime value prediction and customer loyalty quantification based on the dynamic feature vector, wherein the customer lifetime value prediction calls different prediction models according to a customer behavior mode to calculate, and the customer loyalty quantification carries out fitting calculation on preset latent variable observation indexes through a structural equation model; the dynamic mapping unit is used for respectively taking the customer lifetime value prediction result and the customer loyalty quantization result as the input of a first coordinate axis and a second coordinate axis, and mapping the customer to a designated level in a layered architecture formed by a platinum layer, a gold layer, a steel layer and a heavy lead layer in real time through a preset two-dimensional threshold grid; The state deduction unit is used for deducting a state transition probability distribution vector of the customer from the current level to other levels in a future preset period through a time sequence classification model according to the historical level sequence of the customer and the behavior trend index in the dynamic characteristic vector; And the strategy driving unit is used for matching the corresponding automatic operation strategy from the strategy knowledge base according to the current level of the customer, the predicted state transition probability and the real-time triggering behavior event, converting the automatic operation strategy into an executable instruction and distributing the executable instruction to the downstream service system.
  2. 2. The intelligent customer asset management system based on a hierarchical dynamic driving model according to claim 1, wherein a first prediction model is configured in the collaborative computing unit, the first prediction model is a fusion algorithm fusing a BG/NBD model and a Markov chain model, and the specific implementation process is that the BG/NBD model is utilized to process the historical transaction time and transaction times data of customers in the dynamic feature vector, predict the transaction probability of the customers in a specific future period, dynamically construct the transaction probability sequence into a non-homogeneous state transition matrix in the Markov chain model, and output all the residual lifetime value of the customers by solving the steady expected benefits of the Markov chain.
  3. 3. The intelligent customer asset management system based on the hierarchical dynamic driving model according to claim 2, wherein the collaborative computing unit is specifically configured to execute model selection logic, to determine a customer purchase behavior model according to a customer dynamic feature vector, in particular, a variation coefficient of a historical transaction interval and a latest transaction time, acquired from the unified data center, to call the first prediction model to predict a customer lifetime value by using a fusion algorithm fusing a negative binomial distribution model and a Markov chain model if the intermittent purchase behavior is determined and the latest transaction time is less than or equal to a preset activity threshold, to use an improved Pareto/NBD model if the intermittent purchase behavior is determined and the latest transaction time is greater than the preset activity threshold, to predict by using a Shifted-Beta-geotric model if the intermittent purchase behavior is determined, and to estimate by using a mean initialization model based on a similar customer group for a new customer.
  4. 4. The intelligent customer asset management system based on the hierarchical dynamic driving model according to claim 1, wherein a trained recurrent neural network is adopted as the time series classification model in the state deduction unit, an input layer of the recurrent neural network receives the time series customer-level coding sequences provided by the dynamic mapping unit, and a customer lifetime value change rate and a behavior activity slope derived from the dynamic feature vector, and an output layer of the recurrent neural network provides the state transition probability distribution vector, wherein each element represents the probability of transition of a customer to a corresponding level.
  5. 5. The intelligent customer asset management system based on the hierarchical dynamic driving model of claim 1, further comprising a feedback optimization unit for collecting customer feedback and behavior change data returned by a downstream business system after executing the automated operation strategy as effect data, converting the effect data into a reward signal, and performing iterative optimization on parameters of a time-series classification model in the state deduction unit and policy matching logic of a policy knowledge base in the policy driving unit through a reinforcement learning algorithm.
  6. 6. The intelligent customer asset management system based on the hierarchical dynamic driving model of claim 1, wherein the unified data center further comprises a customer asset quality evaluation unit for modeling the hierarchical distribution evolution of the newly obtained customer group in the hierarchical architecture into a three-state bin model, wherein the three states correspond to a heavy iron steel layer, a gold layer and a platinum layer, fitting the time-varying transfer rate parameters of customers among states based on the real-time output of the collaborative computing unit and the dynamic mapping unit, predicting the long-term steady-state hierarchical composition ratio of the new customer group by solving the differential equation set corresponding to the three-state bin model, and generating a quantization score for the quality of a new customer channel or marketing activity based on the ratio.
  7. 7. The intelligent customer asset management system based on the hierarchical dynamic driving model of claim 1, further comprising a loss early warning and intervention unit, wherein for customers in the gold layer and the platinum layer, key covariates related to loss risks are continuously extracted from the dynamic feature vectors, the key covariates are input into a pre-trained Cox proportional risk regression model, individual loss risk indexes of the customers are calculated and updated, and when the individual loss risk indexes break through a threshold dynamically calculated according to historical data, high-risk early warning signals are generated and sent to a policy driving unit; the policy driving unit is also used for preferentially matching and activating the customer saving policy when the high-risk early warning signal is received.
  8. 8. The intelligent customer asset management system based on the hierarchical dynamic driving model as set forth in claim 1, further comprising a quantitative stripping decision unit for constructing a discrete state space according to the time interval of the last transaction of the customer and the transaction frequency in a preset time period, solving an optimal strategy for maximizing the long-term expected net benefit of the customer by combining a preset state transition probability matrix and a benefit matrix by using a Markov decision process model, wherein the optimal strategy designates an action of 'continue input' or 'stop input' for each state, and for a customer group corresponding to the state of which the optimal strategy indicates 'stop input', controlling the strategy driving unit not to match and distribute any resource input type operation strategy for the customer group.
  9. 9. The intelligent customer asset management system based on the hierarchical dynamic driving model according to claim 1, wherein the customer loyalty quantification in the collaborative computing unit is specifically that a structural equation measurement model comprising a plurality of latent variables of customer satisfaction, customer trust, relationship commitment and future loyalty is preset, each latent variable is associated with a plurality of observation indexes from a questionnaire table or a behavior log, observation index data generated by a table in the unified data are periodically input into the measurement model, model fitting and verification are performed by adopting a partial least square algorithm, a factor score of the latent variable of each customer is output, and the factor score of the future loyalty is used as the current loyalty quantification result.
  10. 10. The intelligent customer asset management system based on a hierarchical dynamic driving model of claim 1, wherein said policy driving unit comprises: The strategy knowledge base is pre-stored with association relations, wherein the association relations associate a hierarchical transition scene, a triggering condition and at least one automatic operation action; The strategy matcher is used for inquiring the strategy knowledge base according to the current level, the target transition level and the real-time triggered behavior event of the customer and matching out the corresponding target automatic operation action; And the instruction distributor is used for converting the target automatic operation action into an executable instruction and distributing the executable instruction to a downstream service execution system.

Description

Customer asset intelligent management system based on layered dynamic driving model Technical Field The invention relates to the technical field of business data processing, in particular to a customer asset intelligent management system based on a hierarchical dynamic driving model. Background In the digital economic age, customer relationships have become one of the most central strategic assets of enterprises. The traditional "customer relationship management" concept is evolving towards "customer asset management" which emphasizes quantitative assessment, active operation and dynamic optimization of customer relationships in terms of return on investment. With the maturity of big data, machine learning and cloud computing technologies, enterprises can acquire and store massive and multidimensional customer interaction data, which lays a technical foundation for constructing an intelligent and automatic customer asset management system. How to transform these data into accurate insights into the future value of customers, and drive efficient resource allocation and personalized interactions accordingly, becomes a key challenge for enterprises to promote core competitiveness and realize sustainable growth. Early customer asset management relied primarily on simple statistical analysis models based on historical transaction data. Most typically, the RFM model clusters customers through three static dimensions of customer last consumption, frequency of consumption, and amount of consumption. In addition, basic postamble models of Customer Life Value (CLV) are also widely used, which generally assume that customer behavior is stable, and future benefits can be simply predicted based on past means. These methods rely on manually running the analysis periodically, with the results presented in a static report. Marketing decisions are heavily dependent on the personal experience of the manager, have long cycle from analysis to execution, lag actions, and are difficult to achieve large-scale personalization. As technology advances, the prior art has evolved from traditional methods toward more complex models and partial automation. Students and business software began to introduce more sophisticated statistical models to predict CLV, such as Pareto/NBD models and their derivatives, such as BG/NBD models, for predicting the future transaction behavior of customers, and markov chain models were used to describe the state transitions of customers. These models improve the accuracy of value predictions, especially when dealing with non-contractual customer relationships. Some advanced CRM or customer data platforms begin dynamic customer clustering using clustering algorithms, such as K-Means, and attempt to predict customer chunking using machine learning classification models. However, in the prior art, the existing grouping model or prediction model usually operates at a fixed time point to generate a snapshot analysis result, in the interval between two analyses, the value state, loyalty and behavior intention of the customer may have changed significantly, but the system cannot sense and respond in real time, resulting in serious delay of intervention measures and missing of the optimal operation time, most of the prior art focuses on predicting a single index, lacks prediction capability of the customer on a dynamic transition path in the whole asset value hierarchy system, so that formulation of preventive maintenance and active upgrading strategies lacks accurate data support, and in CLV metering, the existing scheme often adopts a single model for all customers or needs to manually select models for different groups, so that complex mixed behavior modes such as intermittent purchase, regular purchase and the like in reality cannot be flexibly and accurately processed, the prediction error of the existing model is large, which leads to inaccurate asset estimation and influences the basic reliability of layering and resource allocation. Aiming at the technical scheme, the defects in the prior art lead to the problems that customer asset management is still in a relatively static, passive response and manual intervention stage, mismatching of marketing resources and aggravation of high-value customer loss risk are caused, and the management efficiency of customer assets is low. Disclosure of Invention The invention aims to provide a customer asset intelligent management system based on a hierarchical dynamic driving model, which aims to solve the problems in the background technology. The invention provides a customer asset intelligent management system based on a layered dynamic driving model, which adopts the following technical scheme to realize the aim of the invention: A customer asset intelligent management system based on a hierarchical dynamic driving model comprises the following modules: the unified data center is used for accessing and fusing original customer data from a transaction system, an inte