Intelligent customer life cycle management SCRM system and method

Abstract

The invention relates to the technical field of customer relationship management, and discloses an intelligent customer life cycle management SCRM system, which comprises a multi-source data acquisition and fusion engine; a dynamic customer portrait modeling engine; a customer life cycle state evaluation engine; a predictive intervention engine; an intelligent strategy decision engine; a task distribution and execution engine; and a feedback closed loop and learning engine. According to the intelligent customer life cycle management SCRM system and method, by setting a multi-modal decision engine, taking a customer state as input, learning an optimal strategy of maximizing a long term value (LTV) through a DQN / PPO algorithm, when a customer is in a high loss risk state, triggering an exclusive retention scheme for the customer, and generating a strategy according to a comprehensive conversion rate, cost and satisfaction target, so that the customer life cycle management is realized. Moreover, the personalized content template can be automatically generated, the optimal information channel is accurately sent to different clients, personalized matching is carried out for the clients, and the loss rate of the clients is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of customer relationship management, and in particular to an intelligent customer lifecycle management (SCRM) system and method. Background Art

[0002] Customer lifecycle management is the core link of customer relationship management (CRM), covering the entire process from potential customer identification, conversion, retention to churn recovery; with the rise of social marketing, social customer relationship management (SCRM) systems have become an important carrier for corporate customer operations.

[0003] According to a factory production customer lifecycle management system mentioned in the invention patent with Chinese patent application number 202210785466.1, the factory production customer lifecycle management system, when managing the customer lifecycle, is convenient for obtaining inspection reports from customer development to product testing, and conducting full-process control. It is open and transparent, which facilitates process control and customer review, and improves the credibility and ease of review during product testing. However, the marketing service strategies of the factory customer lifecycle management system when in use are mostly general templates and manually preset rules, which lack the ability to adapt to different customers in a targeted manner and cannot match the customer's real-time purchasing intentions, making the maintenance of the customer's life cycle relatively homogeneous. Therefore, it is necessary to propose an intelligent customer lifecycle management SCRM system and method to solve the above-mentioned problems. Summary of the Invention

[0004] (1) Technical problems solved

[0005] In response to the shortcomings of the existing technology, the present invention provides an intelligent customer lifecycle management (SCRM) system and method, which has the advantages of being able to achieve personalized adaptation capabilities for different customers, solving the problem of the background technology that cannot achieve personalized adaptation capabilities for different customers.

[0006] (2) Technical solution

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent customer lifecycle management (SCRM) system, comprising:

[0008] A multi-source data collection and fusion engine, which collects and fuses structured and unstructured data from multiple customer touchpoints (S1-S6) in real time, processes it using a streaming computing framework, and uses NLP and / or audio and video analysis technology (V) to extract deep semantic information and output a unified customer behavior event stream;

[0009] A dynamic customer profile modeling engine is used to continuously build and update customer profiles based on the customer behavior event stream, including basic attributes (PA), behavioral characteristics (BA), interest needs (IN), emotional satisfaction (SA), value indicators (VP), and real-time status and intention (SI); wherein the real-time status and intention (SI) are dynamically inferred in real time using sequence models such as LSTM and Transformer and / or pre-trained language models;

[0010] A customer lifecycle status assessment engine, configured to output a quantitative lifecycle stage (Q-LS) representing the customer's current specific stage, loyalty, churn risk, and other dimensions based on the dynamic customer profile using a trained machine learning model (M);

[0011] A predictive intervention engine, configured to predict the probability of a customer's future critical event (PE) based on the dynamic customer profile and historical data; and trigger an intervention signal when the predicted probability exceeds a threshold;

[0012] An intelligent strategy decision engine, configured to receive trigger signals from the quantitative lifecycle stage (Q-LS), the dynamic customer profile, and the predictive intervention engine, invoke a reinforcement learning model (R) and / or a multi-attribute decision model (MAUT / TOPSIS), and generate an optimal interaction strategy combination including a recommended action (Action), a content template (Content), a reach channel (Channel), an execution timing (Timing), and a personalized script (Script);

[0013] The task distribution and execution engine is used to convert the optimal interaction strategy combination into specific tasks, and intelligently match them based on the executor profile and task profile, and distribute them to the executor or automated program (AE) for execution;

[0014] The feedback loop and learning engine is used to collect customer feedback data on strategy execution, and use the feedback data (customer feedback data) to update the dynamic customer profile in real time, and trigger the update of the model (Model Update) in the customer lifecycle status assessment engine, intelligent strategy decision engine, and predictive intervention engine to achieve system adaptive optimization.

[0015] Preferably, the state space of the reinforcement learning model (R) in the intelligent policy decision engine is constructed based on the dynamic customer portrait and Q-LS, the action space is a set of optional interactive strategies, and the reward function is designed based on the customer's subsequent feedback behavior (customer feedback data).

[0016] Preferably, the intelligent policy decision engine also includes a module for evaluating policy effectiveness based on an AB testing framework and using the evaluation results for model updating.

[0017] Preferably, the predictive intervention engine uses a survival analysis model and / or a classification model to predict the probability of key events such as customer churn, purchase, and upgrade.

[0018] Preferably, the dynamic customer profile modeling engine supports dynamic adjustment of the weight of each feature in the profile according to the importance of the time and type of behavior.

[0019] Preferably, the output Q-LS of the customer lifecycle status assessment engine is a probability or score vector containing multiple dimensions (stage, risk, value).

[0020] An intelligent customer lifecycle management (SCRM) method includes any one of the intelligent customer lifecycle management (SCRM) systems described above, including:

[0021] S1: Real-time collection and integration of data from multiple customer touchpoints (S1-S6);

[0022] S2: Based on the integrated data, dynamic customer profiles are constructed and updated in real time;

[0023] S3: Based on dynamic customer profiles, evaluate and output quantitative life cycle stages (Q-LS);

[0024] S4: Based on dynamic customer profiles and historical data, predict the probability of critical events (PE) and trigger intervention if the threshold is exceeded;

[0025] S5: Combine Q-LS, dynamic customer profiles, and intervention signals to generate the optimal interaction strategy combination;

[0026] S6: Convert the optimal interaction strategy combination into tasks, distribute them intelligently and execute them;

[0027] S7: Collect customer feedback data on strategy implementation;

[0028] S8: Use customer feedback data to update dynamic customer profiles;

[0029] S9: Based on customer feedback data and system status, trigger relevant model updates, complete the closed loop, and return to S1.

[0030] Preferably, when generating the optimal interaction strategy combination in step S5, reinforcement learning strategy optimization or multi-attribute strategy sorting is selectively applied according to the decision context.

[0031] Preferably, the task distribution in step S6 is performed based on the matching degree between the executor's ability, experience, historical load performance, task complexity, and customer value.

[0032] Preferably, the model update in step S9 includes online learning or batch processing updates to the parameters or structures of the customer lifecycle status evaluation model, the policy decision model, and the prediction model.

[0033] (3) Beneficial effects

[0034] Compared with the existing technology, the present invention provides an intelligent customer lifecycle management (SCRM) system and method, which has the following beneficial effects:

[0035] 1. This intelligent customer lifecycle management (SCRM) system and method builds customer portraits in real time, uses machine learning models to accurately determine the customer's current lifecycle stage and pipe characteristics in real time, and uses streaming computing frameworks (such as Flink / Kafka) to fuse multi-source data. Combined with LSTM / Transformer sequence models, it dynamically captures customer status evolution (such as "active → silent → churn warning") and uses BERT-type models to analyze intent in real time (such as "price sensitivity" and "functional consultation"). The portrait update delay is reduced to seconds, and dynamic images can analyze customer intent in real time.

[0036] 2. This intelligent customer lifecycle management (SCRM) system and method sets up a multimodal decision engine, takes customer status as input, and learns the optimal strategy for maximizing long-term value (LTV) through the DQN / PPO algorithm. When a customer is at a high risk of churn, it triggers an exclusive retention plan for the customer and generates strategies based on comprehensive conversion rate, cost, and satisfaction targets. It can also automatically generate personalized content templates, accurately send the optimal information channels to different customers, perform personalized matching for customers, and reduce customer churn rates.

[0037] 3. This intelligent customer lifecycle management (SCRM) system and method uses the Cox model to predict the customer churn time window through survival analysis and early warning, triggers intervention in customer churn in advance, and improves customer retention rate. It also predicts the probability of key behaviors such as purchases and upgrades through GBDT / neural networks, and prioritizes sales resources for high-potential customers who are able to purchase, facilitating the allocation of pre-positioned customer resources and thereby improving transaction rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a structural flow chart of the customer lifecycle management (SCRM) method of the present invention;

[0039] Figure 2 This is a schematic diagram of the structure of the customer lifecycle management SCRM system of the present invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] See also Figure 1-2 , an intelligent customer lifecycle management (SCRM) system, including:

[0042] A multi-source data collection and fusion engine, which collects and fuses structured and unstructured data from multiple customer touchpoints (S1-S6) in real time, processes it using a streaming computing framework, and uses NLP and / or audio and video analysis technology (V) to extract deep semantic information and output a unified customer behavior event stream;

[0043] A dynamic customer profile modeling engine is used to continuously build and update customer profiles based on the customer behavior event stream, including basic attributes (PA), behavioral characteristics (BA), interest needs (IN), emotional satisfaction (SA), value indicators (VP), and real-time status and intention (SI); wherein the real-time status and intention (SI) are dynamically inferred in real time using sequence models such as LSTM and Transformer and / or pre-trained language models;

[0044] A customer lifecycle status assessment engine, configured to output a quantitative lifecycle stage (Q-LS) representing the customer's current specific stage, loyalty, churn risk, and other dimensions based on the dynamic customer profile using a trained machine learning model (M);

[0045] A predictive intervention engine, configured to predict the probability of a customer's future critical event (PE) based on the dynamic customer profile and historical data; and trigger an intervention signal when the predicted probability exceeds a threshold;

[0046] An intelligent strategy decision engine, configured to receive trigger signals from the quantitative lifecycle stage (Q-LS), the dynamic customer profile, and the predictive intervention engine, invoke a reinforcement learning model (R) and / or a multi-attribute decision model (MAUT / TOPSIS), and generate an optimal interaction strategy combination including a recommended action (Action), a content template (Content), a reach channel (Channel), an execution timing (Timing), and a personalized script (Script);

[0047] The task distribution and execution engine is used to convert the optimal interaction strategy combination into specific tasks, and intelligently match them based on the executor profile and task profile, and distribute them to the executor or automated program (AE) for execution;

[0048] Feedback closed loop and learning engine, used to collect customer feedback data on strategy execution, and use the feedback data (customer feedback data) to update the dynamic customer profile in real time, and trigger the update of the model (Model Update) in the customer life cycle status assessment engine, intelligent strategy decision engine, and predictive intervention engine to achieve system adaptive optimization

[0049] An intelligent customer lifecycle management (SCRM) method includes any one of the intelligent customer lifecycle management (SCRM) systems described above, including:

[0050] S1: Real-time collection and integration of data from multiple customer touchpoints (S1-S6);

[0051] S2: Based on the integrated data, dynamic customer profiles are constructed and updated in real time;

[0052] S3: Based on dynamic customer profiles, evaluate and output quantitative life cycle stages (Q-LS);

[0053] S4: Based on dynamic customer profiles and historical data, predict the probability of critical events (PE) and trigger intervention if the threshold is exceeded;

[0054] S5: Combine Q-LS, dynamic customer profiles, and intervention signals to generate the optimal interaction strategy combination;

[0055] S6: Convert the optimal interaction strategy combination into tasks, distribute them intelligently and execute them;

[0056] S7: Collect customer feedback data on strategy implementation;

[0057] S8: Use customer feedback data to update dynamic customer profiles;

[0058] S9: Based on customer feedback data and system status, trigger relevant model updates, complete the closed loop, and return to S1.

[0059] In the case implementation, stream processing platforms such as Kafka / Flink can be used to access multi-source data, Spark can be used for batch processing, and NLP cloud services (such as AWS Comprehend, Azure Text Analytics) or self-built models can be combined to analyze text / speech, realize data fusion, and build a unified customer ID system.

[0060] In the case implementation, TensorFlow / PyTorch was used to build LSTM / GRU / Transformer models to process behavior sequence prediction states; BERT-type models were used for intent recognition, and Redis / Dynamo DB was used to store and quickly access real-time portraits.

[0061] In case implementation, XG Boost / Light GBM can be used for stage classification and risk scoring; feature engineering needs to integrate basic attributes, behavioral characteristics, sentiment scores, interaction frequency, response time, etc.

[0062] In the case implementation, the policy decision engine:

[0063] (1) Reinforcement learning: Use frameworks such as TensorFlow Agents and RayRLlib to design the state space (portrait features + stages), action space (strategy options), and reward function (purchase amount is positive, churn cost is negative); use DeepQ-Network (DQN) or Proximal Policy Optimization (PPO) algorithm; (2) Multi-attribute decision-making: Use rule-based or model-based methods to calculate the utility value of each strategy in terms of conversion rate, satisfaction improvement, and cost, combine the entropy weight method to determine the weight, and select the TOPSIS method for sorting.

[0064] In the case implementation, prediction model: use the survival analysis model to predict the churn time, and use the classification model to predict the purchase / upgrade probability; task distribution matching: establish personnel portraits (skill matrix, efficiency indicators, current load), and use greedy algorithms, Hungarian algorithms or simple rules to maximize cost-benefit matching; closed-loop learning: regularly collect feedback data, use incremental learning frameworks or trigger regular full-scale model retraining, and use A / B testing platforms to verify the effectiveness of strategies; system deployment: based on microservice architecture, deployed on cloud platforms (such as AWS, Azure, GCP), core AI modules can be deployed on GPU-accelerated instances, and the front-end uses React / Vue combined with the mobile terminal.

[0065] In summary, the intelligent customer lifecycle management SCRM system and method, through setting up a real-time customer portrait construction, uses a machine learning model to accurately determine the customer's current life cycle stage and pipe characteristics in real time, and uses a streaming computing framework (such as Flink / Kafka) to fuse multi-source data, combined with the LSTM / Transformer sequence model to dynamically capture the evolution of customer status (such as "active→silent→churn warning"), and uses a BERT-type model to analyze intentions in real time (such as "price sensitivity" and "functional consultation"), and the portrait update delay is reduced to seconds. Dynamic images can analyze customer intentions in real time.

[0066] In addition, by setting up a multimodal decision-making engine, taking customer status as input, and learning the optimal strategy to maximize long-term value (LTV) through the DQN / PPO algorithm, when the customer is at high risk of churn, a dedicated retention plan is triggered for the customer, and strategies are generated based on comprehensive conversion rate, cost and satisfaction goals. It can also automatically generate personalized content templates, accurately send the optimal information channel to different customers, perform personalized matching for customers, and reduce customer churn rate.

[0067] In addition, by conducting survival analysis and early warning on customers and using the Cox model to predict the customer churn time window, we can trigger intervention in customer churn in advance and improve customer retention rate. By predicting the probability of key behaviors such as purchases and upgrades through GBDT / neural networks, we can prioritize sales resources for high-potential customers who are able to purchase, facilitate the allocation of front-end customer resources, and thus improve transaction rates.

[0068] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0069] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent customer lifecycle management (SCRM) system, characterized by: include: A multi-source data collection and fusion engine, which collects and fuses structured and unstructured data from multiple customer touchpoints (S1-S6) in real time, processes it using a streaming computing framework, and uses NLP and / or audio and video analysis technology (V) to extract deep semantic information and output a unified customer behavior event stream; A dynamic customer profile modeling engine is used to continuously build and update customer profiles based on the customer behavior event stream, including basic attributes (PA), behavioral characteristics (BA), interest needs (IN), emotional satisfaction (SA), value indicators (VP), and real-time status and intention (SI); wherein the real-time status and intention (SI) are dynamically inferred in real time using sequence models such as LSTM and Transformer and / or pre-trained language models; A customer lifecycle status assessment engine, configured to output a quantitative lifecycle stage (Q-LS) representing the customer's current specific stage, loyalty, churn risk, and other dimensions based on the dynamic customer profile using a trained machine learning model (M); A predictive intervention engine, configured to predict the probability of a customer's future critical event (PE) based on the dynamic customer profile and historical data; and trigger an intervention signal when the predicted probability exceeds a threshold; An intelligent strategy decision engine, configured to receive trigger signals from the quantitative lifecycle stage (Q-LS), the dynamic customer profile, and the predictive intervention engine, invoke a reinforcement learning model (R) and / or a multi-attribute decision model (MAUT / TOPSIS), and generate an optimal interaction strategy combination including a recommended action (Action), a content template (Content), a reach channel (Channel), an execution timing (Timing), and a personalized script (Script); The task distribution and execution engine is used to convert the optimal interaction strategy combination into specific tasks, and intelligently match them based on the executor profile and task profile, and distribute them to the executor or automated program (AE) for execution; The feedback loop and learning engine is used to collect customer feedback data on strategy execution, and use the feedback data (customer feedback data) to update the dynamic customer profile in real time, and trigger the update of the model (Model Update) in the customer lifecycle status assessment engine, intelligent strategy decision engine, and predictive intervention engine to achieve system adaptive optimization.

2. The intelligent customer lifecycle management (SCRM) system according to claim 1, characterized in that: The state space of the reinforcement learning model (R) in the intelligent strategy decision engine is constructed based on the dynamic customer profile and Q-LS, the action space is a set of optional interactive strategies, and the reward function is designed based on the customer's subsequent feedback behavior (customer feedback data).

3. The intelligent customer lifecycle management (SCRM) system according to claim 1, characterized in that: The intelligent strategy decision engine also includes a module for evaluating strategy effectiveness based on an AB testing framework and using the evaluation results for model updating.

4. The intelligent customer lifecycle management (SCRM) system according to claim 1, characterized in that: The predictive intervention engine uses survival analysis models and / or classification models to predict the probability of key events such as customer churn, purchase, and upgrade.

5. The intelligent customer lifecycle management (SCRM) system according to claim 1, characterized in that: The dynamic customer profile modeling engine supports dynamic adjustment of the weights of each feature in the profile based on the importance of the time and type of behavior.

6. The intelligent customer lifecycle management (SCRM) system according to claim 1, characterized in that: The output Q-LS of the customer lifecycle status assessment engine is a probability or score vector containing multiple dimensions (stage, risk, value).

7. An intelligent customer lifecycle management (SCRM) method, comprising the intelligent customer lifecycle management (SCRM) system according to any one of claims 1 to 6, characterized in that: include: S1: Real-time collection and integration of data from multiple customer touchpoints (S1-S6); S2: Based on the integrated data, dynamic customer profiles are constructed and updated in real time; S3: Based on dynamic customer profiles, evaluate and output quantitative life cycle stages (Q-LS); S4: Based on dynamic customer profiles and historical data, predict the probability of critical events (PE) and trigger intervention if the threshold is exceeded; S5: Combine Q-LS, dynamic customer profiles, and intervention signals to generate the optimal interaction strategy combination; S6: Convert the optimal interaction strategy combination into tasks, distribute them intelligently and execute them; S7: Collect customer feedback data on strategy implementation; S8: Use customer feedback data to update dynamic customer profiles; S9: Based on customer feedback data and system status, trigger relevant model updates, complete the closed loop, and return to S1.

8. The intelligent customer lifecycle management (SCRM) method according to claim 7, characterized in that: When generating the optimal interaction strategy combination in step S5, reinforcement learning strategy optimization or multi-attribute strategy sorting is selectively applied according to the decision context.

9. The intelligent customer lifecycle management (SCRM) method according to claim 7, characterized in that: In step S6, the task distribution is carried out based on the ability, experience, historical load performance of the executors, and the matching degree of task complexity and customer value.

10. The intelligent customer lifecycle management (SCRM) method according to claim 7, characterized in that: The model update in step S9 includes online learning or batch processing updates of the parameters or structures of the customer lifecycle status evaluation model, the policy decision model, and the prediction model.

Images