Processing method and device for coping with customer loss, equipment, medium and program product

Abstract

The invention provides a processing method and device for coping with customer loss, equipment, a medium and a program product, and relates to the technical field of artificial intelligence. The method comprises the steps that in response to a received loss early warning event of a target customer, data of the target customer is inquired from multiple internal systems to obtain a comprehensive data snapshot of the target customer, and different systems in the multiple internal systems are used for recording data of different dimensions of the customer; generating a first prompt word based on the comprehensive data snapshot of the target customer; and using the first prompt word to indicate the fine-tuned large language model to generate a risk abstract of the target customer. In this way, when customer loss early warning needs to be analyzed, the trouble of manually switching data checking, data arrangement and analysis among a plurality of systems can be avoided, and the analysis efficiency of customer loss is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more specifically to a method, apparatus, device, medium, and program product for dealing with customer churn. Background Technology

[0002] The application of artificial intelligence technology to customer churn early warning can effectively help companies identify potential customers in a timely manner and take appropriate measures. This can help companies improve their service capabilities and adapt to the market environment more effectively.

[0003] However, current AI-powered solutions for customer churn primarily focus on predicting churn probabilities. In this case, the final output of the alert is still a single value or a set of values ​​(such as a risk score) or some quantitative indicators (such as "the slope of the amount over the past 3 months -0.8"). While this output is accurate, it essentially only states the fact of the risk status without providing any information about the causes of the risk. This is abstract and difficult for non-technical business personnel to understand intuitively, making it difficult to efficiently translate alert information into concrete action recommendations. Summary of the Invention

[0004] The present invention provides a method, apparatus, device, medium, and program product for dealing with customer churn, which can at least partially solve the above-mentioned problems.

[0005] A first aspect of this invention provides a method for handling customer churn. The method includes: in response to receiving a churn warning event for a target customer, querying data of the target customer from multiple internal systems to obtain a comprehensive data snapshot of the target customer, wherein different systems within the multiple internal systems are used to record data of the customer in different dimensions; generating a first prompt word based on the comprehensive data snapshot of the target customer; and using the first prompt word to instruct a finely tuned large language model to generate a risk summary of the target customer.

[0006] According to an embodiment of the present invention, generating a first prompt word based on a comprehensive data snapshot of the target customer includes: obtaining a first prompt word template, the first prompt word template including multiple placeholders, the multiple placeholders corresponding to the dimensions of customer data recorded in the multiple internal systems; and filling the multiple placeholders according to the comprehensive data snapshot of the target customer to obtain the first prompt word.

[0007] According to an embodiment of the present invention, the method further includes: retrieving communication records related to the target customer from one or more internal communication channels; obtaining public information of the target customer from one or more external data sources; and generating an in-depth attribution analysis report based on the risk summary, the communication records, and the public information.

[0008] According to an embodiment of the present invention, generating a deep attribution analysis report based on the risk summary, the communication records, and the publicly available information includes: extracting a summary from the communication records using the large language model to generate an internal communication information summary; extracting a summary from the publicly available information using the large language model to generate an external publicly available information summary; identifying the semantic bias category of the publicly available information of the target customer using a classification model for performing semantic bias analysis tasks; generating a second prompt word based on the risk summary, the internal communication information summary, the external publicly available information summary, and the semantic bias category; and using the second prompt word to instruct the large language model to generate the deep attribution analysis report.

[0009] According to an embodiment of the present invention, the method further includes: generating a third cue word based on the risk summary and the deep attribution analysis report; using the third cue word to instruct the large language model to generate an intervention strategy, the intervention strategy including action suggestions for different staff members; and generating action plans for each staff member serving the target customer based on the intervention strategy.

[0010] According to an embodiment of the present invention, the method further includes: receiving execution feedback information of the action plan; performing structured processing and tagging processing on the execution feedback information, and storing it in the fine-tuning database of the large language model, wherein the fine-tuning database is used to fine-tune and optimize the large language model.

[0011] A second aspect of this invention provides a processing apparatus for addressing customer churn. The processing apparatus includes: an early warning event receiving and data integration module and a dynamic narrative report generation module. The early warning event receiving and data integration module is configured to, in response to receiving a churn early warning event for a target customer, query data of the target customer from multiple internal systems to obtain a comprehensive data snapshot of the target customer, wherein different systems within the multiple internal systems record data of the customer in different dimensions. The dynamic narrative report generation module is configured to generate a first prompt word based on the comprehensive data snapshot of the target customer; and to use the first prompt word to instruct a finely tuned large language model to generate a risk summary of the target customer.

[0012] A third aspect of the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0013] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0014] A fifth aspect of the present invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0015] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0016] Figure 1 A flowchart illustrating a method for handling customer churn according to an embodiment of the present invention is shown schematically;

[0017] Figure 2 The illustration depicts an application scenario of a method for handling customer churn according to an embodiment of the present invention.

[0018] Figure 3 This illustration schematically shows a system architecture for applying a customer churn handling method according to an embodiment of the present invention;

[0019] Figure 4 The diagram illustrates the execution flowchart of the early warning event receiving and data integration module in one embodiment of the present invention.

[0020] Figure 5 This schematically illustrates the execution flowchart of a dynamic narrative report generation module in one embodiment of the present invention; and

[0021] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a customer churn handling method according to an embodiment of the present invention. Detailed Implementation

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0023] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0024] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0025] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0026] In the technical solution of this invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0027] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this embodiment of the invention all provide users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process proceeds to the expert decision-making process.

[0028] Figure 1 A flowchart illustrating a method for handling customer churn according to an embodiment of the present invention is shown.

[0029] like Figure 1 As shown, the method may include operations S110 to S130.

[0030] In operation S110, in response to receiving a churn warning event for a target customer, the system queries the target customer's data from multiple internal systems to obtain a comprehensive data snapshot of the target customer. The multiple internal systems are used to record different dimensions of customer data.

[0031] In operation S120, a first prompt word is generated based on a comprehensive data snapshot of the target customer.

[0032] In operation S130, the risk summary of the target customer is generated using the first prompt word as an instruction to a finely tuned large language model.

[0033] When a customer churn warning event occurs, this invention can automatically obtain relevant data about the customer from multiple internal systems. Then, through the natural language generation capabilities of a large language model, the obtained data is transformed into an easy-to-understand and practical risk summary, improving the readability of the warning information, providing users with in-depth risk cause analysis, and providing clear and specific guidance for subsequent decision-making and intervention measures.

[0034] In existing technologies, after receiving a customer churn warning, users typically need to log into multiple internal business systems (such as customer relationship management and credit systems), manually query and integrate various structured and unstructured information about the customer, and rely on personal experience and business understanding to piece together the full picture of the problem before deciding on the next steps for customer retention. This entire process is not only inefficient and time-consuming, but also highly dependent on the individual abilities of staff, resulting in inconsistent quality of customer retention services and an inability to develop standardized, high-level contingency plans.

[0035] In contrast, the embodiments of the present invention avoid the trouble and inefficiency of manually switching between multiple systems to retrieve data, organize information and perform analysis. Moreover, by designing prompt words, the format of the output risk summary can be controlled, achieving standardized and normalized analysis, and improving the quality and reliability of the analysis.

[0036] According to other embodiments of the present invention, after operation S130, further analysis can be conducted using a large language model, combining customer interaction records, publicly available customer information in external systems, etc., to further analyze and identify the underlying causes of customer churn. In other embodiments, based on these analyses, the large language model can also be used to generate intervention strategies, such as action suggestions for different staff members. This greatly improves the efficiency of addressing customer churn.

[0037] The following description uses a bank customer as an example to illustrate the embodiments of the present invention. It should be noted that the methods and apparatus for handling customer churn determined in the embodiments of the present invention can be used in the fintech field, or in any field other than fintech; the present invention does not limit the application field.

[0038] Figure 2 The illustration depicts an application scenario of a method for handling customer churn according to an embodiment of the present invention.

[0039] like Figure 2 As shown, application scenario 200 according to this embodiment may include a first terminal device 201, a second terminal device 202, a third terminal device 203, a network 204, and a server 205. The network 204 serves as a medium for providing a communication link between the first terminal device 201, the second terminal device 202, the third terminal device 203, and the server 205. The network 204 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0040] Users can use the first terminal device 201, the second terminal device 202, and the third terminal device 203 to interact with the server 205 via the network 204 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 201, the second terminal device 202, and the third terminal device 203, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0041] Server 205 can be a server that provides various services. Server 205 may have a large language model deployed in it, or server 205 may communicate with a device that has a large language model deployed in it. Server 205 may execute the method of the embodiments of the present invention and feed back the generated report to the user terminal (such as the first terminal device 201, the second terminal device 202, or the third terminal device 203).

[0042] It should be noted that the customer churn handling method provided in the embodiments of the present invention can generally be executed by server 205. Correspondingly, the customer churn handling device provided in the embodiments of the present invention can generally be located in server 205. The customer churn handling method provided in the embodiments of the present invention can also be executed by a server or server cluster that is different from server 205 and capable of communicating with the first terminal device 201, the second terminal device 202, the third terminal device 203, and / or server 205. Correspondingly, the customer churn handling device provided in the embodiments of the present invention can also be located in a server or server cluster that is different from server 205 and capable of communicating with the first terminal device 201, the second terminal device 202, the third terminal device 203, and / or server 205.

[0043] It should be understood that Figure 2 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0044] Figure 3 The diagram illustrates a system architecture for applying a customer churn handling method according to an embodiment of the present invention. The following is in conjunction with... Figure 3The system architecture of the present invention is described, and an intelligent intervention and interpretation system is introduced that can implement the customer churn handling method of the embodiments of the present invention.

[0045] Specifically, this embodiment aims to solve at least some of the core technical problems in the prior art, such as poor interpretability of warning information, lack of in-depth attribution analysis, lack of intelligent decision support, and failure to effectively integrate internal and external unstructured data after accurate quantitative warning.

[0046] This embodiment provides a system that can seamlessly connect to upstream quantitative risk early warning systems. Leveraging the powerful natural language understanding, generation, reasoning, and knowledge integration capabilities of large language models, it automatically transforms abstract risk scores or data indicators into an "intelligent customer risk insight report" that is easily understood by non-professionals, rich in information, and includes attribution analysis and even specific action suggestions. This embodiment aims to revolutionize the customer retention workflow from the traditional model of "receiving churn warning events - manual analysis - passive decision-making" to a highly efficient and intelligent model of "receiving insights - instant understanding - proactive intervention," thereby improving, to at least some extent, the work efficiency, decision-making quality, and customer relationship management level of users (such as staff).

[0047] like Figure 3 As shown in the illustration, this embodiment demonstrates a customer churn early warning interpretation and intelligent intervention system 300 that can integrate a large language model. This system 300 can use risk early warning as a downstream enhancement module, working collaboratively with the risk early warning system. The risk early warning system is used to predict the probability of customer churn. When the system 300 receives a high-risk early warning event (also called a churn early warning event, risk early warning event, or early warning event) output by the risk early warning system, the system 300 can achieve in-depth interpretation, intelligent intervention, and knowledge accumulation of the early warning event through the collaborative work of the early warning event receiving and data integration module 301, the dynamic narrative report generation module 302, the multi-source information fusion and deep attribution module 303, the generative personalized intervention strategy recommendation module 304, and the closed-loop feedback and knowledge base construction module 305. The specific operation of each module is detailed below.

[0048] (I) Early Warning Event Receiving and Data Integration Module 301

[0049] The early warning event receiving and data integration module 301 is responsible for receiving early warning events in real time and integrating various structured and unstructured data to provide basic data support for subsequent risk analysis, generation of personalized intervention strategies and report generation. Figure 4 The diagram illustrates the execution flowchart of the early warning event receiving and data integration module in one embodiment of the present invention.

[0050] like Figure 4 As shown, the process may include steps S41 to S44.

[0051] Step S41: Receive high-risk warning events through event listening.

[0052] The early warning event receiving and data integration module 301 serves as the entry point for system 300, communicating in real-time with the upstream risk prediction system via an API interface. When the upstream risk prediction system outputs a customer's high-risk score, the early warning event can be sent to the early warning event receiving and data integration module 301 in the form of a structured JSON object. This early warning event contains key customer information and risk assessment results, ensuring that system 300 can accurately understand and process this data.

[0053] In one embodiment, the alert event may include, but is not limited to, the following information: customer ID, customer name, risk score (reflecting the degree of customer churn risk), customer importance level or priority information, and a list of key characteristics. In one embodiment, the list of key characteristics may include the top N (e.g., N=5) characteristics that affect the current customer risk score, along with their specific values ​​and contributions. These characteristics reflect the key factors influencing changes in customer risk.

[0054] Step S42: Analyze the warning event and obtain key risk characteristics.

[0055] To improve the interpretability of risk warnings, the warning event receiving and data integration module 301 introduces a feature contribution analysis method to further analyze the risk scores in warning events. In the warning event receiving and data integration module 301, feature contribution analysis helps identify the key features that contribute most to the current customer risk score and displays the values ​​and contribution levels of these features. For example, system 300 might display the following features and their contribution levels: Transaction amount slope of a certain business in the past 3 months: -0.8 (feature contribution level: 0.45), indicating that the decrease in the customer's transaction amount in this area is a significant factor leading to their high-risk score; Abnormal behavior score: 0.92 (feature contribution level: 0.38), indicating that the customer's recent behavioral patterns are abnormal, further increasing their churn risk.

[0056] Step S43: Concurrently query multiple internal systems to obtain real-time data snapshots of customers and integrate the real-time data snapshots.

[0057] Upon receiving an early warning event, the early warning event receiving and data integration module 301 can immediately initiate concurrent query requests to multiple internal systems (such as internal system 1, internal system 2, internal system 3, etc.) based on the customer ID to obtain a comprehensive snapshot of the customer's data. This process ensures that system 300 can integrate a large amount of relevant information in the shortest possible time, providing foundational data for subsequent risk analysis and action recommendations.

[0058] In one embodiment, the specific data source may include the following:

[0059] Customer Relationship Management System: Obtains basic customer information, such as industry category, company size, key contacts, and historical records. This information helps in understanding the customer's basic situation and business background.

[0060] Core business system: Allows querying customer business transaction data and providing the latest business status.

[0061] System 300 retrieves recent customer data recorded in the internal system, which helps to understand business dealings with customers and better assess changes in their risks.

[0062] Through concurrent queries, this multi-dimensional data will be quickly integrated into a unified customer data object containing detailed context, and then passed to subsequent modules for processing.

[0063] Step S44: Pass the data to the subsequent modules.

[0064] After integration, various structured and unstructured data from customer objects can be organized into a structured format to form a comprehensive structured data snapshot, which is then passed to the downstream dynamic narrative report generation module 302. The dynamic narrative report generation module 302 can use this data to provide in-depth risk analysis reports, and in some embodiments, it can also generate specific action recommendations.

[0065] A comprehensive data snapshot can include not only basic customer data, but also historical customer behavior, potential reasons for churn, and personalized risk assessments based on customer importance levels. This integration process allows for the rapid acquisition of comprehensive customer background information upon receiving an alert, reducing the time spent manually switching between multiple systems, improving decision-making efficiency, and ensuring the comprehensiveness and accuracy of the information to a large extent through standardized process settings.

[0066] (ii) Dynamic Narrative Report Generation Module 302

[0067] The core task of the dynamic narrative report generation module 302 is to transform the comprehensive data snapshot obtained by the early warning event receiving and data integration module 301 into a natural language description, so as to provide relevant personnel with an easy-to-understand and actionable risk summary. Leveraging the powerful natural language processing capabilities of the large language model, the dynamic narrative report generation module 302 can process structured data through language processing to generate a logical and hierarchical risk analysis report that meets business needs.

[0068] Figure 5The diagram illustrates the execution flowchart of the dynamic narrative report generation module 302 in one embodiment of the present invention. (See reference...) Figure 5 The execution flow of the dynamic narrative report generation module 302 can include three main steps S51 to S53.

[0069] Step S51: Construct structured prompts through structured prompt word engineering.

[0070] In the dynamic narrative report generation module 302, the core of generating natural language reports lies in constructing an efficient and structured prompt word template. The purpose of the prompt word design is to enable the large language model to generate natural language text that meets business requirements based on the input structured data. This prompt word template can contain multiple placeholders, and the dynamic narrative report generation module 302 can dynamically fill these placeholders with comprehensive data snapshots to form personalized prompt words. This can largely ensure the standardization and accuracy of the prompt words, so that the generated report content has both data-driven accuracy and maintains a fluent and easy-to-understand language structure.

[0071] In one embodiment, the prompt word template can be designed as follows:

[0072] "Role: You are a senior risk analyst for xx business;

[0073] Task: Based on the structured data provided below, write a risk profile summary for the account manager regarding customer "{customer_name}". The summary must be objective, concise, data-driven, and highlight the most important risk points.

[0074] Customer basic information:

[0075] -Customer Name: {customer_name}

[0076] - Current churn risk score: {risk_score} / 100

[0077] Core risk indicators (ranked by importance):

[0078] {key_features_list}

[0079] Supplementary data:

[0080] - Data from the past 6 months: {};

[0081] Please generate a risk summary based on the information above.

[0082] Step S52, data population and large language model invocation.

[0083] After the prompt template design is complete, the dynamic narrative report generation module 302 can extract relevant information from the structured data object obtained from the previous module and populate the placeholders in the prompt template. For example, the customer's name, risk score, level, and recent key risk characteristics can be filled into the corresponding positions in the template. In one embodiment, the core risk indicators section can be formatted as an easy-to-read, importance-sorted list to more clearly display each risk factor. After filling in the information, the complete prompt text can be sent to the pre-configured large language model. The large language model can generate a fluent, accurate, and data-driven natural language summary based on the input prompts.

[0084] Step S53, Result Analysis and Output

[0085] After receiving the prompt words, the large language model performs complex data analysis and language generation, and the final output can include a concise, clear, and key-point risk summary. This risk summary can clearly reflect the customer's current risk status, and in some embodiments, it can also highlight the core factors affecting the risk score, which can help relevant staff quickly understand the root causes of the current customer churn risk.

[0086] In one embodiment, the generated natural language text is shown below:

[0087] High-risk warning: XX Technology Company

[0088] Please note that the churn risk score for XX Technology Company, a strategic client managed by your organization, has risen sharply to 85 points. System analysis shows that the risk is primarily driven by the following factors:

[0089] 1. First, its transaction volume over the past three months has shown a significant and rapid downward trend. Specifically, this is manifested in…

[0090] 2. Secondly, a recent behavioral pattern has significantly deviated from its historical norms. The behavioral abnormality score is 0.92, indicating a high risk of abnormal behavior.

[0091] 3. Finally, the frequency of business transactions also decreased significantly. Specifically, this manifested as…

[0092] Based on the above analysis, it is recommended to communicate with the client's senior management as soon as possible to assess their cash flow situation and arrange countermeasures.

[0093] The dynamic narrative report generation module 302 utilizes the natural language generation capabilities of a large language model to transform complex structured data into easily understandable and practical risk summaries. Through standardized prompt word template design, precise data filling, and efficient large language model invocation, the dynamic narrative report generation module 302 not only improves the readability of warning information but also enhances the report's operability. It provides relevant personnel with in-depth risk causal analysis and offers clear and specific guidance for subsequent decision-making and intervention measures.

[0094] (III) Multi-source information fusion and deep attribution module 303

[0095] The core objective of the Multi-Source Information Fusion and Deep Attribution Module 303 is to deeply analyze and explain the root causes of customer risk, answering the question, "Why did this happen?" By fusing various unstructured information from internal and external sources and combining the reasoning capabilities of a large language model, it can provide in-depth attribution analysis of customer churn risk. Compared to traditional simple data analysis, the Multi-Source Information Fusion and Deep Attribution Module 303 can integrate multi-dimensional information and reveal potential, deep-seated business risk factors, thereby helping staff make more accurate and forward-looking decisions. The implementation process is detailed below.

[0096] First, extract and process internal unstructured communication records and other information.

[0097] Specifically, the multi-source information fusion and deep attribution module 303 can first access multiple internal data sources to extract unstructured information such as customer-related communication records. Through direct connection with internal systems' application programming interfaces (APIs) or databases, it can obtain customer communication records, email archives, and call logs from various internal communication channels such as customer relationship management systems, email servers, and customer service centers. Before using these customer-related communication records, email archives, and call logs, customer authorization is required. For example, when signing a contract with a customer, their opinion must be clearly solicited and the specific purpose explained. If the customer does not agree, these customer-related communication records cannot be used; if the customer agrees, these customer-related communication records can be obtained. This information may contain customer attitudes, cooperation intentions, and potential business issues. Specifically, after obtaining customer authorization, relevant text records can be retrieved using the customer ID and a set time window (e.g., the past 90 days). After preprocessing, the retrieved text data can be summarized through a single call to a large language model, extracting key information such as customer attitudes, cooperation intentions, complaints, executive changes, and operational difficulties. For example, it can extract information such as "the customer showed strong interest in a certain product" from the visit notes of relevant staff or identify "the customer complained about the low efficiency of our service" from email records.

[0098] Secondly, it can acquire and process publicly available information from external sources.

[0099] While acquiring internal unstructured information, the multi-source information fusion and deep attribution module 303 also collects publicly available information related to the client by integrating APIs from external data sources. These external information sources may include, for example, business news aggregation platforms and financial data providers. Collecting external information helps understand the dynamics of the client's industry, the external challenges the client company faces, and significant news events that may affect its risk profile. For example, key information such as the client's official name and stock code can be used to retrieve relevant news, announcements, and industry updates from the past 30 days. This process allows for real-time understanding of the client's external environment and assessment of potential external risk factors, such as changes in industry policies, company financial problems, or other negative events. After summarizing the retrieved external information, semantic analysis tasks can be performed to determine the semantic bias of these news items or reports (e.g., positive, neutral, or negative). For example, external news might mention "the company is facing liquidity problems," and analysis can determine whether this information is negative news, thus providing valuable clues for subsequent attribution analysis.

[0100] Secondly, large language models can be used for attribution analysis.

[0101] Attribution analysis is a core component of the multi-source information fusion and deep attribution module 303. It aims to infer the root causes of customer churn risk by deeply mining internal and external information. At this stage, a complex cue word template can be constructed to integrate all information obtained from internal and external data sources, which is then passed to a large language model for inference and analysis.

[0102] This prompt template can include not only a summary of the customer's risk, but also summaries of internal communications and external public information. This allows the large language model to obtain more complete background information, enabling it to make more accurate analyses and generate in-depth attribution analysis reports.

[0103] In one embodiment, the content of the prompt word template can be as follows:

[0104] "Role: You are a top business investigation consultant and analyst, skilled at finding the root cause of events from a wealth of complex information."

[0105] Task: Based on the comprehensive information about customer {customer_name} below, analyze and infer the possible reasons for their current high churn risk. Please provide 2-3 of the most likely causes and your supporting analysis.

[0106] Risk profile summary (generated from the first round of analysis):

[0107] {narrative_summary_from_front module}

[0108] Summary of internal communication information:

[0109] {internal_text_summary}

[0110] Summary of publicly available external information:

[0111] {external_text_summary}

[0112] Please generate an in-depth attribution analysis report.

[0113] After the prompt words generated based on the prompt word template are provided to the large language model, the large language model can integrate all the information to reason and identify the core reasons for customer churn and its possible underlying background.

[0114] Next, the results will be analyzed and output.

[0115] After reasoning and analysis using the large language model, the multi-source information fusion and deep attribution module 303 can output a deep attribution analysis report. This report can include several of the most likely reasons for increased customer churn risk and provide detailed reasoning. The report's content can cover information extracted from various data sources and can also combine semantic analysis and the impact of changes in the external environment to comprehensively analyze the causes of customer risk from multiple perspectives. The final deep attribution analysis report can serve as the second part of the customer risk report, providing relevant personnel with a more specific and in-depth explanation of the causes of risk.

[0116] For example, the generated report might look like this:

[0117] Deep attribution analysis:

[0118] 1. Core reason inference: The company is facing some kind of pressure; the basis for this is…;

[0119] 2. Possible reason two: Instability within the core team; the basis for this is...

[0120] Through the multi-source information fusion and deep attribution module 303, system 300 can deeply analyze the root causes of customer churn risk. System 300 first acquires data from multiple internal and external information sources, extracts key information through text summarization and semantic analysis, and then uses a large language model for reasoning to generate a deep attribution analysis report. The multi-source information fusion and deep attribution module 303 combines scattered and complex unstructured information with data-driven risk assessment, revealing the deep-seated causes of customer risk through intelligent analysis, thereby providing stronger support for customer retention and decision-making.

[0121] (iv) Regarding the generative personalized intervention strategy recommendation module 304

[0122] The core objective of the generative personalized intervention strategy recommendation module 304 is to answer the question "What should we do?" by providing staff with a set of specific and actionable customer retention action suggestions. After customer churn risk is identified and deeply analyzed, simple early warning and attribution analysis are insufficient to drive actual customer intervention and relationship maintenance. The generative personalized intervention strategy recommendation module 304 can generate personalized intervention strategies through a large language model, providing relevant staff with detailed action plans to help them take effective measures to address customer churn risk. In one embodiment, the generative personalized intervention strategy recommendation module 304 can intelligently generate short-term and medium-term action strategies and customize specific support suggestions based on customer level and needs, thereby improving the targeting and effectiveness of customer management.

[0123] First, corresponding prompts can be generated through a strategy-based prompt word generation project.

[0124] To generate personalized customer retention strategies, the generative personalized intervention strategy recommendation module 304 can construct a complex cue word template. In one embodiment, this cue word template can require a large language model to act as a "decision advisor," generating actionable suggestions based on the customer's risk profile, in-depth attribution analysis, and specific circumstances. In another embodiment, the cue word template design can be based on multiple requirements, such as accurately describing the customer's background, risk status, and the role requirements of relevant staff. The cue word template can structure customer information and clearly define the specific needs of staff at different levels.

[0125] In one embodiment, the prompt word template is as follows:

[0126] Role: You are a seasoned customer relationship management expert, proficient in customer retention and crisis management.

[0127] Task: Based on the risk profile and underlying causes of customer "{customer_name}", please provide a set of specific, personalized, and actionable customer retention action recommendations for staff at different levels.

[0128] Client Profile:

[0129] -Customer Name: {customer_name}

[0130] -Industry type: {industry_type}

[0131] - In the collaborative products: {product_list}

[0132] Risk Summary: {}

[0133] Deep attribution analysis: {}

[0134] Please provide the following:

[0135] 1. Short-term action recommendations for staff (to be implemented within the next 1-3 days).

[0136] 2. Mid-term follow-up strategy for staff (within the next 1-2 weeks).

[0137] 3. Suggestions for collaborative support to business leaders.

[0138] Then, the large language model is invoked to generate the results.

[0139] After receiving the complete contextual input above, the large language model can generate a highly customized set of customer retention action suggestions based on the customer's risk summary, in-depth attribution analysis, and customer profile. The generative personalized intervention strategy recommendation module 304 automatically generates specific action plans by calling the large language model and leveraging its powerful natural language generation capabilities.

[0140] In one embodiment, the generated strategy may not only encompass short- and medium-term action plans but also provide specific support recommendations to staff at different levels. For example, for account managers, short-term action recommendations might include phone calls with the client's finance manager, while medium-term action recommendations might involve arranging meetings with company executives. For business leaders, collaborative support recommendations could be provided.

[0141] Next, the results will be analyzed and output.

[0142] After the large language model returns the results, the system will parse the generated strategies and deliver them to account managers and relevant personnel. These strategy recommendations will serve as actionable plans, providing account managers with clear guidelines for action and helping them effectively address the risk of customer churn.

[0143] Each action plan will specify the timeframe for implementation (such as short-term and medium-term) and the content of implementation (such as telephone communication, face-to-face meeting arrangements, etc.) to ensure that the recommendations are both actionable and can be quickly implemented in practice.

[0144] In existing technologies, upstream risk warning systems cease their mission after issuing a customer churn risk warning, offering no guidance on customer retention actions. This leaves relevant personnel aware of the churn risk but unable to make informed decisions on how to maintain relationships, relying entirely on individual judgment. This not only increases the difficulty and uncertainty of decision-making but also prevents companies from scaling up effective customer retention practices. In this embodiment of the invention, the generative personalized intervention strategy recommendation module 304 can combine internal and external data to deeply attribute customer churn risk, and further, based on the needs of different staff members and medium- to long-term requirements, provide targeted action strategies.

[0145] (V) Regarding the closed-loop feedback and knowledge base construction module 305

[0146] The core objective of the closed-loop feedback and knowledge base construction module 305 is to enable the learning and evolution capabilities of various modules within System 300, as well as the large language model it invokes. By collecting feedback from staff during actual operations, System 300 can continuously accumulate experience, optimize the decision-making process, and improve the quality of future intervention recommendations. The closed-loop feedback and knowledge base construction module 305 not only enhances the long-term intelligent capabilities of System 300 but also ensures that customer retention strategies remain adaptable to constantly changing market and customer needs. Through the storage and application of structured and vectorized feedback information, System 300 can learn independently, continuously optimize, and provide banks with more precise customer risk intervention solutions.

[0147] First, obtain records of intervention behaviors and outcomes.

[0148] The closed-loop feedback and knowledge base construction module 305 can record feedback received by staff after implementing intervention measures. Intelligent risk insight reports can be automatically pushed to the relevant staff's workspace interface, which provides a structured feedback form. After intervention, staff can fill out this form, recording in detail the implemented intervention measures, customer feedback, action results, and key success / failure factor analysis. Specifically, staff can choose preset action options (such as "telephone communication" or "face-to-face meeting") or freely enter text descriptions of customer feedback (such as "the customer expressed satisfaction and is willing to continue cooperation" or "the customer believes the solution is unsuitable"). Simultaneously, staff need to evaluate the action results, including assessment items such as "risk mitigation," "risk persistence," or "customer churn," and conduct corresponding key success or failure factor analysis.

[0149] Secondly, the feedback information is structured and vectorized.

[0150] After collecting unstructured feedback text from staff, System 300 can parse this information and transform it into structured data tags. To ensure the accuracy and standardization of the information, System 300 can call a large language model to parse the feedback text, automatically identifying and tagging key elements within it.

[0151] Building upon this foundation, successful intervention cases—comprehensive case texts outlining the "risk scenario - attribution analysis - intervention strategy - positive customer feedback"—can be converted into high-dimensional vectors using a text embedding model. This vectorization process transforms the semantic information of the text into a computer-processable vector form for similarity matching and retrieval. Through vectorization, System 300 can quickly match similar cases in future warning events, providing precise contextual references for subsequent intervention strategy generation.

[0152] Next, we will discuss the construction and application of the knowledge base.

[0153] To ensure that System 300 can continuously learn and improve from past intervention cases, the closed-loop feedback and knowledge base construction module 305 stores all structured feedback data in a relational database and establishes a vector-based knowledge base. Through this structured database, System 300 can perform statistical analysis at any time to identify which intervention strategies are most effective in specific industries, customer types, or risk scenarios.

[0154] Simultaneously, the text vectors of intervention cases are stored in a vector database. When generating intervention strategies in the future, System 300 first converts the descriptive text of the current situation into vectors and performs a similarity search in the vector database to find the N most similar historical success cases. These historical success cases can serve as examples for "contextual learning," assisting in generating more precise intervention recommendations.

[0155] Finally, we will continue to optimize.

[0156] By continuously accumulating successful and unsuccessful intervention cases, System 300 can generate a large amount of high-quality labeled data. This data can be used to fine-tune the large language model. The fine-tuned large language model can be more adapted to specific business scenarios, especially in customer retention strategies, providing more accurate and effective intervention suggestions based on different industries, customer levels, and their risk status. Through fine-tuning, System 300 can learn itself and continuously optimize its decision-making process, improving customer retention efficiency.

[0157] Furthermore, the fine-tuned large language model can be adjusted based on feedback from historical cases to improve the model's learning strategy, enabling it to automatically select more effective action suggestions in future intervention strategy generation.

[0158] As can be seen from the above description, compared with the prior art, the method of the embodiments of the present invention has the following beneficial effects:

[0159] First, it can transform early warning information from "data" to "intelligence". Through the application of large language models, abstract risk scores (or churn probability) and related data indicators can be transformed into easy-to-understand narrative reports and attribution analyses, which greatly improves the interpretability, readability and insight of early warning information.

[0160] Secondly, the method in this invention can provide personalized and intelligent decision support. For example, it can dynamically generate corresponding intervention strategies based on each customer's unique risk situation, significantly improving the scientific nature of decision-making and the effectiveness of customer retention actions.

[0161] Furthermore, the method of this invention can improve the efficiency of customer relationship maintenance. Specifically, by automating the integration and in-depth analysis of internal and external multi-source information, it can avoid tedious and repetitive data queries and manual analysis by staff, thereby improving response speed and processing efficiency.

[0162] Finally, the method of this invention can construct a knowledge accumulation system for sustainable learning and evolution. Through a closed-loop learning mechanism of "early warning-intervention-feedback," every customer retention practice can be transformed into reusable and learnable knowledge assets. This not only continuously optimizes the accuracy of its own recommendation strategy but also establishes a dynamically updated and shared practice knowledge base, thereby achieving the continuous evolution of organizational capabilities.

[0163] According to embodiments of the present invention, the system 300 described above, in whole or in part, can constitute a customer churn handling apparatus according to embodiments of the present invention. For example, in one embodiment, the customer churn handling apparatus according to embodiments of the present invention may include only an early warning event receiving and data integration module 301 and a dynamic narrative report generation module 302.

[0164] According to embodiments of the present invention, any and multiple modules among the early warning event receiving and data integration module 301, dynamic narrative report generation module 302, multi-source information fusion and deep attribution module 303, generative personalized intervention strategy recommendation module 304, and closed-loop feedback and knowledge base construction module 305 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of the present invention, at least one of the following modules—early warning event receiving and data integration module 301, dynamic narrative report generation module 302, multi-source information fusion and deep attribution module 303, generative personalized intervention strategy recommendation module 304, and closed-loop feedback and knowledge base construction module 305—can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the following modules—early warning event receiving and data integration module 301, dynamic narrative report generation module 302, multi-source information fusion and deep attribution module 303, generative personalized intervention strategy recommendation module 304, and closed-loop feedback and knowledge base construction module 305—can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0165] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a customer churn handling method according to an embodiment of the present invention.

[0166] like Figure 6 As shown, an electronic device 900 according to an embodiment of the present invention includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0167] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0168] According to an embodiment of the present invention, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0169] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0170] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0171] Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the customer churn handling method provided in the embodiments of the present invention.

[0172] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0173] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0174] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this embodiment of the invention. According to embodiments of the invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0175] According to embodiments of the present invention, program code for executing the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0176] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0177] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

Claims

1. A method for handling customer churn, comprising: in response to receiving a churn warning event of a target customer, querying data of the target customer from a plurality of internal systems, different systems of the plurality of internal systems being configured to record data of customers in different dimensions, to obtain a comprehensive data snapshot of the target customer; generating a first prompt based on the comprehensive data snapshot of the target customer; and indicating a fine-tuned large language model to generate a risk summary of the target customer using the first prompt. The generating of the first prompt based on the comprehensive data snapshot of the target customer comprises:

2. The method of claim 1, wherein, obtaining a first prompt template, the first prompt template comprising a plurality of placeholders corresponding to the dimensions in which the plurality of internal systems record data of customers; and filling the plurality of placeholders according to the comprehensive data snapshot of the target customer to obtain the first prompt. The method further comprises:

3. The method of claim 1, wherein, retrieving communication records related to the target customer from one or more internal communication channels; obtaining public information of the target customer from one or more external data sources; and generating a deep attribution analysis report based on the risk summary, the communication records, and the public information. The generating of the deep attribution analysis report based on the risk summary, the communication records, and the public information comprises:

4. The method of claim 3, wherein, extracting summaries from the communication records using the large language model to generate internal communication information summaries; extracting summaries from the public information using the large language model to generate external public information summaries; identifying semantic bias categories of the public information of the target customer using a classification model configured to perform a semantic bias analysis task; generating a second prompt based on the risk summary, the internal communication information summaries, the external public information summaries, and the semantic bias categories; and indicating the large language model to generate the deep attribution analysis report using the second prompt. The method further comprises:

5. The method of claim 3 or 4, wherein, generating a third prompt based on the risk summary and the deep attribution analysis report; indicating the large language model to generate an intervention strategy using the third prompt, the intervention strategy comprising action recommendations for different staff members; and generating action plans for different staff members serving the target customer based on the intervention strategy. The method further comprises:

6. The method of claim 5, wherein, receiving execution feedback information of the action plans; after structuring and tagging the execution feedback information, storing the execution feedback information in a fine-tuning database of the large language model, the fine-tuning database being configured to fine-tune the large language model.

7. An apparatus for handling customer churn, comprising: a warning event receiving and data integrating module configured to, in response to receiving a churn warning event of a target customer, query data of the target customer from a plurality of internal systems, different systems of the plurality of internal systems being configured to record data of customers in different dimensions, to obtain a comprehensive data snapshot of the target customer; and ​ A dynamic narrative report generation module configured to: generate a first prompt based on the comprehensive data snapshot of the target customer; and instruct a fine-tuned large language model to generate a risk summary of the target customer using the first prompt. 8.An electronic device comprising: one or more processors; memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program or instructions, wherein, The computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-6.

10. A computer program product comprising computer programs or instructions, wherein, The computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-6. The computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-6.

Images