Marketing intervention method and system based on customer churn prediction

Abstract

The application relates to the technical field of loss prediction, and discloses a marketing intervention method and system based on customer loss prediction, which comprises the following steps: performing abnormal fluctuation detection on the interactive behavior data of customers to obtain loss sign data; performing state evolution analysis on the loss sign data to obtain active decline characteristics; performing time sequence trajectory fitting on the loss sign data to obtain an active decline trajectory graph; matching the active decline trajectory graph with the life cycle of the customers, and fusing the static attribute data of the customers to obtain a loss risk multidimensional portrait; performing decision factor analysis on the loss risk driving factors of the customers, comprehensively judging the root cause characteristics after the analysis, and obtaining a loss risk grade; mapping the root cause characteristics and the loss risk grade to a preset strategy intervention library, combining the static attribute data, and generating an individualized intervention strategy; and the application can improve the efficiency of marketing intervention based on customer loss prediction.

Description

Technical Field

[0001] This invention relates to the field of churn prediction technology, and in particular to a marketing intervention method and system based on customer churn prediction. Background Technology

[0002] Existing customer churn prediction technologies have limitations at the data processing level. They lack a systematic approach to detecting abnormal fluctuations in customer interaction behavior data and fail to construct accurate baselines based on historical behavior data. This makes it difficult to effectively filter out routine behavior data and accurately extract churn warning signs. Furthermore, the analysis of the evolution of churn-related data is not in-depth enough, remaining at the level of simple extraction of surface features. It fails to comprehensively capture the decline trends across multiple dimensions such as frequency, interval, and channel, resulting in an incomplete characterization of activity decline characteristics and impacting the accuracy of subsequent predictions.

[0003] In the risk assessment and intervention strategy generation stages, existing technologies have failed to effectively correlate and match customer activity decline trajectories with lifecycle stages. Furthermore, when constructing churn risk profiles, the integration of static attribute data and dynamic decline characteristics is insufficient, resulting in a lack of scientific rigor and accuracy in risk level assessment. In addition, existing intervention strategies are mostly general designs, failing to be tailored to the root causes of churn and individual customer attributes, leading to insufficient targeting of marketing interventions and difficulty in effectively curbing customer churn. Therefore, improving the accuracy of customer churn prediction and the targeting and effectiveness of marketing interventions has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a marketing intervention method and system based on customer churn prediction to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a marketing intervention method based on customer churn prediction, comprising:

[0006] S1. Detect abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data;

[0007] S2. Perform state evolution analysis on the churn symptom data to obtain the activity decay characteristics of the churn symptom data;

[0008] S3. Based on the activity decline characteristics, perform time-series trajectory fitting on the churn symptom data to obtain the customer's activity decline trajectory map;

[0009] S4. Match the activity decline trajectory map with the customer's life cycle stages and integrate the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk.

[0010] S5. Based on the multi-dimensional profile of churn risk, perform decision factor analysis on the driving factors of customer churn risk, and conduct a comprehensive risk assessment on the root cause characteristics after analysis to obtain the churn risk level of the customer.

[0011] S6. Map the root cause characteristics and the churn risk level to a preset strategy intervention library, and combine the static attribute data to generate a personalized intervention strategy for the customer.

[0012] In a preferred embodiment, the step of detecting abnormal fluctuations in customer interaction behavior data to obtain customer churn indicator data includes:

[0013] Obtain customers' historical behavior data and interaction behavior data;

[0014] The historical behavior data is analyzed for routine behavior to obtain the customer's historical statistical baseline;

[0015] Based on the historical statistical baseline, a difference comparison analysis is performed on the interaction behavior data to obtain the behavior deviation index of the interaction behavior data;

[0016] Based on the behavioral deviation index, routine behavioral data is removed from the interaction behavior data to obtain the customer churn symptom data.

[0017] In a preferred embodiment, the step of performing state evolution analysis on the churn symptom data to obtain the activity decay characteristics of the churn symptom data includes:

[0018] Heterogeneous feature extraction is performed on the churn symptom data to obtain the frequency characteristics, interval characteristics, and channel diversity characteristics of the churn symptom data;

[0019] The frequency feature, the interval feature, and the channel diversity feature are sorted in a multidimensional order according to time sequence to obtain the customer's feature time sequence.

[0020] The change trend analysis of the characteristic time series is performed to obtain the frequency decay rate, interval decay rate and channel decay rate of the characteristic time series.

[0021] The frequency decay rate, the interval decay rate, and the channel decay rate are integrated into the activity decay feature of the churn symptom data.

[0022] In a preferred embodiment, the step of fitting a time-series trajectory of the churn symptom data based on the activity decline characteristics to obtain the customer's activity decline trajectory map includes:

[0023] A multidimensional coordinate system for the customer is constructed with time as the horizontal axis and the multidimensional feature vector of the activity decline feature as the vertical axis.

[0024] The activity decay feature is mapped to the multidimensional coordinate system to obtain the discrete time-series state points of the activity decay feature;

[0025] Nonlinear fitting is performed on the discrete time-series state points to obtain the continuous decay trajectory curve of the discrete time-series state points;

[0026] The smoothness of the continuous decline trajectory curve is optimized, and key information is annotated on the optimized continuous decline trajectory curve to obtain the customer's activity decline trajectory map.

[0027] In a preferred embodiment, matching the activity decline trajectory with the customer's lifecycle stages includes:

[0028] Based on the preset enterprise customer lifecycle relationship, the customer lifecycle is divided into the introduction stage, growth stage, maturity stage and decline stage;

[0029] The trend of the activity decline trajectory is analyzed to identify the key inflection points in the activity decline trajectory.

[0030] Based on the timing and magnitude of the key inflection points, the activity decline trajectory is compared with the introduction, growth, maturity and decline phases to confirm the customer's current lifecycle stage.

[0031] In a preferred embodiment, the process of fusing the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk includes:

[0032] Obtain the industry attributes, years of cooperation, customer level, and historical value contribution data of the aforementioned customers;

[0033] The static attribute feature vector of the customer is obtained by synthesizing the industry attributes, the years of cooperation, the customer level and the historical value contribution data.

[0034] Multi-source feature fusion is performed on the static attribute feature vector and the stage feature vector of the current life cycle stage to obtain the customer's fused feature vector;

[0035] Based on the fused feature vector, the activity decline trajectory map is reconstructed into a multi-dimensional view to obtain a multi-dimensional profile of the customer's churn risk.

[0036] In a preferred embodiment, the step of performing decision factor analysis on the customer's churn risk drivers based on the multi-dimensional churn risk profile, and conducting a comprehensive risk assessment on the analyzed root cause characteristics to obtain the customer's churn risk level, includes:

[0037] Based on a pre-defined decision factor weight rule base, an importance analysis is performed on the multi-dimensional features of the multi-dimensional profile of churn risk to obtain the churn risk contribution of the multi-dimensional features.

[0038] Based on the contribution of the churn risk, a dominant factor analysis is performed on the churn risk drivers of the customer to obtain the root cause characteristics of the customer. The churn risk drivers are extracted from the churn risk multidimensional profile and include static attribute characteristics of the customer and dynamic characteristics of customer activity decline.

[0039] The risk quantification assessment of the root cause characteristics is performed to obtain the loss risk degree of the root cause characteristics;

[0040] The customer's churn risk level is obtained by associating and matching the churn risk level with a preset risk level rule base.

[0041] In a preferred embodiment, the formula for calculating the churn risk level is as follows: ;

[0042] In the formula, Indicates the first The risk of loss of each root cause characteristic Indicates the first The contribution of each root cause characteristic to the risk of loss. This represents the total number of the root cause features. Represents the natural logarithm function. This represents the square root operation. This indicates a summation operation.

[0043] In a preferred embodiment, mapping the root cause characteristics and the churn risk level to a preset strategy intervention library, and combining this with the static attribute data to generate a personalized intervention strategy for the customer, includes:

[0044] Logical association parsing is performed on the preset strategy intervention library to obtain the intervention matching rules of the strategy intervention library;

[0045] Based on the intervention matching rules, the root cause characteristics and the churn risk level are correlated and mapped to obtain the basic intervention strategy for the customer.

[0046] Based on the customer level and industry attributes in the static attribute data, the key parameters of the basic intervention strategy are corrected to obtain the personalized intervention strategy for the customer.

[0047] To address the aforementioned problems, the present invention also provides a marketing intervention system based on customer churn prediction, the system comprising:

[0048] The abnormal fluctuation detection module is used to detect abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data.

[0049] The state evolution analysis module is used to perform state evolution analysis on the churn symptom data to obtain the activity decline characteristics of the churn symptom data.

[0050] The time-series trajectory fitting module is used to perform time-series trajectory fitting on the churn symptom data based on the activity decline characteristics to obtain the customer's activity decline trajectory map.

[0051] The risk profile building module is used to match the activity decline trajectory map with the customer's life cycle stages and integrate the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk.

[0052] The risk assessment and rating module is used to analyze the decision factors driving the customer's churn risk based on the multi-dimensional profile of churn risk, and to conduct a comprehensive risk assessment on the root cause characteristics after analysis to obtain the customer's churn risk level.

[0053] The intervention strategy generation module is used to map the root cause characteristics and the churn risk level to a preset strategy intervention library, and combine the static attribute data to generate a personalized intervention strategy for the customer.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. This invention accurately extracts customer churn signs by systematically detecting abnormal fluctuations, comprehensively captures activity decline characteristics through state evolution analysis, constructs a clear decline trajectory map by combining time-series trajectory fitting, and integrates customer lifecycle stages and static attribute data to form a multi-dimensional profile of churn risk, significantly improving the accuracy and comprehensiveness of churn risk identification and providing solid data support for risk assessment.

[0056] 2. This invention clarifies the root causes of customer churn through decision factor analysis, scientifically assesses the risk level of churn using quantitative formulas, and then generates personalized intervention strategies by linking them to a strategy intervention library and combining them with static attribute data. This ensures the objectivity of risk assessment and the pertinence of intervention strategies, effectively improving the efficiency of marketing interventions and helping to accurately prevent customer churn. Attached Figure Description

[0057] Figure 1A flowchart illustrating a marketing intervention method based on customer churn prediction provided in an embodiment of the present invention;

[0058] Figure 2 A functional block diagram of a marketing intervention system based on customer churn prediction provided in an embodiment of the present invention;

[0059] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0060] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0061] This application provides a marketing intervention method based on customer churn prediction. The executing entity of the marketing intervention method based on customer churn prediction includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the marketing intervention method based on customer churn prediction can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0062] Reference Figure 1 The diagram shown is a flowchart illustrating a marketing intervention method based on customer churn prediction according to an embodiment of the present invention. In this embodiment, the marketing intervention method based on customer churn prediction includes:

[0063] S1. Detect abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data;

[0064] In this embodiment of the invention, the step of detecting abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data includes:

[0065] Obtain customers' historical behavior data and interaction behavior data;

[0066] The historical behavior data is analyzed for routine behavior to obtain the customer's historical statistical baseline;

[0067] Based on the historical statistical baseline, a difference comparison analysis is performed on the interaction behavior data to obtain the behavior deviation index of the interaction behavior data;

[0068] Based on the behavioral deviation index, routine behavioral data is removed from the interaction behavior data to obtain the customer churn symptom data.

[0069] By using the customer data collection system deployed within the enterprise, all interaction records between customers and the enterprise in the past twelve months are comprehensively extracted as historical behavioral data. These records cover information such as the specific time of the interaction, the business content involved, the business process involved, and the interaction channels used. At the same time, all records of customer interactions with the enterprise in the past month through various channels such as online platforms, offline stores, and customer service hotlines are collected in real time as interactive behavioral data. This ensures that both types of data fully cover all interaction scenarios related to customers and the enterprise, without missing any key interaction information.

[0070] The collected historical behavioral data is divided into twelve time periods by natural month. For each time period, the total number of customer interactions, the time interval between each interaction, and the number of interaction channels involved are counted. The total number of interactions in the twelve time periods is added together and divided by twelve to obtain the average interaction frequency. The time interval between all two adjacent interactions is added together and divided by the total number of intervals to obtain the average interaction interval. The total number of interaction channels that occurred in the twelve time periods is counted and divided by twelve to obtain the average channel participation. Based on these three averages, a customer historical statistical baseline is constructed, which includes the interaction frequency baseline, the interaction interval baseline, and the channel participation baseline. This baseline intuitively reflects the behavioral patterns of customers under normal conditions.

[0071] The collected interaction behavior data is divided into time periods by natural month. Following the same standards as historical behavior data statistics, statistics are compiled for each time period, including the total number of customer interactions, the time interval between each interaction, and the number of interaction channels involved. The actual value of each statistical item is compared with the average value of the corresponding statistical item in the historical statistical baseline. The difference between the actual value and the average value of each statistical item is calculated, and then the difference is divided by the average value to obtain the proportion. For example, if the actual number of interactions is 50 times and the interaction frequency benchmark in the historical statistical baseline is 40 times, the difference is 10 times, and the proportion is 25%. The difference and proportion corresponding to each statistical item are used together as the behavior deviation index for that statistical item. The behavior deviation indices of all statistical items together constitute the behavior deviation index of the interaction behavior data.

[0072] A preset behavioral deviation threshold is set, where the difference threshold is set to ±10% of the average value of the corresponding statistical item in the historical statistical baseline, and the proportion threshold is set to ±10%. The behavioral deviation index of each statistical item in the interaction behavior data is checked one by one. If the difference of a certain statistical item is within the range of ±10% to ±10% of the average value, and the proportion is also within the range of ±10% to ±10%, then the interaction behavior corresponding to that statistical item is determined to be normal behavior data. All parts determined to be normal behavior data are removed from the interaction behavior data, and the remaining interaction behavior data is the customer churn indicator data.

[0073] The beneficial effects are that by comprehensively extracting customers' historical behavior data and interaction behavior data, a complete and accurate data foundation is provided for subsequent analysis. By statistically analyzing key indicators of historical behavior data at fixed time dimensions and calculating average values, a historical statistical baseline that objectively reflects customers' normal behavior is constructed. Based on this baseline, interaction behavior data is compared and calculated one by one to obtain clear behavioral deviation indicators. Then, based on preset specific thresholds, regular behavior data is accurately judged and eliminated to ensure that the final churn symptom data can accurately capture customers' abnormal behavior information. The entire process has clear operation steps, clear judgment standards, and strong reproducibility, providing reliable and accurate data support for subsequent churn risk-related analysis.

[0074] S2. Perform state evolution analysis on the churn symptom data to obtain the activity decay characteristics of the churn symptom data;

[0075] In this embodiment of the invention, the step of performing state evolution analysis on the churn symptom data to obtain the activity decline characteristics of the churn symptom data includes:

[0076] Heterogeneous feature extraction is performed on the churn symptom data to obtain the frequency characteristics, interval characteristics, and channel diversity characteristics of the churn symptom data;

[0077] The frequency feature, the interval feature, and the channel diversity feature are sorted in a multidimensional order according to time sequence to obtain the customer's feature time sequence.

[0078] The change trend analysis of the characteristic time series is performed to obtain the frequency decay rate, interval decay rate and channel decay rate of the characteristic time series.

[0079] The frequency decay rate, the interval decay rate, and the channel decay rate are integrated into the activity decay feature of the churn symptom data.

[0080] A comprehensive review of churn symptom data is conducted, counting the total number of customer-enterprise interactions. This total number represents the frequency characteristic of the churn symptom data. Next, the occurrence time of each interaction record is examined, and the time length between two adjacent interactions is calculated. All calculated time lengths are compiled and summarized to form the interval characteristic of the churn symptom data. Then, the interaction channels used by the customer in each interaction record are identified, duplicate channel names are removed, and the number of different channel types remaining is counted. This number represents the channel diversity characteristic of the churn symptom data.

[0081] Based on the specific time of the interaction, arranged in order from earliest to latest, first extract the frequency feature value corresponding to each time node, then extract the interval feature value corresponding to that time node, and finally extract the channel diversity feature value of that time node. Combine these three feature values ​​of each time node in sequence to form a sequence arranged in chronological order. This sequence is the customer's feature time sequence.

[0082] The frequency characteristic values ​​of the first three time nodes in the characteristic time series are selected, and the average of these three values ​​is calculated as the initial benchmark. Then, starting from the fourth time node, the frequency characteristic value of each time node is compared with the initial benchmark, and the difference between the two is calculated. This difference is then divided by the time span from the third time node to the current time node to obtain the frequency change at the current time node. The frequency changes of all time nodes are sorted out in turn to form the frequency decay rate. Using the same method, the average of the interval characteristic values ​​of the first three time nodes is calculated as the initial reference. The difference between the interval characteristic value of each subsequent time node and the initial reference is calculated and divided by the corresponding time span to obtain the interval change at each time node. This is then sorted out to form the interval decay rate. Similarly, the average of the channel diversity characteristic values ​​of the first three time nodes is calculated as the initial standard. The difference between the channel diversity characteristic value of each subsequent time node and the initial standard is calculated and divided by the corresponding time span to obtain the channel change at each time node. This is then sorted out to form the channel decay rate.

[0083] The changes at each time point in the frequency decay rate, the interval decay rate, and the channel decay rate are correlated one by one according to the chronological order of the time points, forming a complete set containing the changes in all three dimensions at all time points. This set is the activity decay feature of the churn symptom data.

[0084] The beneficial effects are that the frequency, interval, and channel diversity characteristics of churn symptom data are accurately extracted through a clear statistical method, ensuring that the generation process of each feature is reproducible and the results are clear. The three features are sorted in chronological order to form a feature time series, which clearly presents the distribution of customer behavior characteristics over time. By calculating the changes based on the average value of the initial stage features, the decay rate of the three dimensions is accurately obtained, which comprehensively reflects the decline trend of customer activity. Finally, the three decay rates are integrated into an activity decay feature, providing comprehensive and accurate data support for subsequent time series trajectory fitting. The whole process is clear and logically coherent, effectively avoiding the problems of insufficient feature extraction or vague description of the decay trend.

[0085] S3. Based on the activity decline characteristics, perform time-series trajectory fitting on the churn symptom data to obtain the customer's activity decline trajectory map;

[0086] In this embodiment of the invention, the step of fitting a time-series trajectory of the churn symptom data based on the activity decline characteristics to obtain the customer's activity decline trajectory map includes:

[0087] A multidimensional coordinate system for the customer is constructed with time as the horizontal axis and the multidimensional feature vector of the activity decline feature as the vertical axis.

[0088] The activity decay feature is mapped to the multidimensional coordinate system to obtain the discrete time-series state points of the activity decay feature;

[0089] Nonlinear fitting is performed on the discrete time-series state points to obtain the continuous decay trajectory curve of the discrete time-series state points;

[0090] The smoothness of the continuous decline trajectory curve is optimized, and key information is annotated on the optimized continuous decline trajectory curve to obtain the customer's activity decline trajectory map.

[0091] The horizontal axis is set with calendar days as the unit. The starting point of the horizontal axis is the earliest interaction time corresponding to the churn symptom data, and the ending point is the latest interaction time corresponding to the churn symptom data. Each calendar day is marked in chronological order. The vertical axis has three independent dimensions, which correspond to the frequency decay rate, interval decay rate and channel decay rate in the activity decay characteristics, respectively. The value range of each vertical axis dimension is determined according to the actual maximum and minimum values ​​of the corresponding decay rate. The maximum value is set to 1.2 times the actual maximum value of the corresponding decay rate, and the minimum value is set to 0. In this way, a customer multi-dimensional coordinate system including the time dimension and the three decay rate dimensions is constructed.

[0092] For each natural day, the frequency decay rate, interval decay rate, and channel decay rate are extracted. The frequency decay rate is mapped to the corresponding position in the first dimension of the vertical axis of the multi-dimensional coordinate system, the interval decay rate is mapped to the corresponding position in the second dimension, and the channel decay rate is mapped to the corresponding position in the third dimension. The three coordinates together determine the unique position of the natural day in the multi-dimensional coordinate system. The position points corresponding to all natural days together constitute the discrete time-series state points of the activity decay characteristics.

[0093] The distribution of all discrete time-series state points in the multidimensional coordinate system is analyzed one by one. The numerical change trend between state points corresponding to two adjacent natural days is analyzed. Starting from the state point of the first natural day, a curve that fits the change law of each state point is drawn along the direction of time. The fluctuation range of the curve near each state point does not exceed 5% of the value of each dimension of the vertical axis corresponding to that state point. This ensures that the curve can fully reflect the overall trend of frequency decay rate, interval decay rate and channel decay rate changing with time. This curve is the continuous decay trajectory curve of discrete time-series state points.

[0094] The slopes of all adjacent segments in the continuous decline trajectory curve are statistically analyzed, and the average difference of all slopes is calculated. A slope threshold is set at 1.5 times the average difference. If the slope difference of an adjacent segment exceeds this threshold, the slope of the next segment is adjusted to keep the slope difference between it and the previous segment within the threshold range. This process is repeated until the slope differences of all adjacent segments meet the requirements, thus optimizing the smoothness of the continuous decline trajectory curve. Subsequently, the specific values ​​of the natural day and the decline rate in each dimension corresponding to each discrete time-series state point are marked on the optimized curve. Nodes where the slope direction changes are also marked, ultimately forming a customer activity decline trajectory map.

[0095] The beneficial effects are as follows: by constructing a multi-dimensional coordinate system with clearly defined time units and numerical ranges, the standardization and uniformity of the coordinate system are ensured, providing a clear benchmark for subsequent data mapping. The three dimensions of activity decline characteristics are accurately mapped to the coordinate system to form discrete time-series state points, ensuring the accuracy of data presentation. By performing nonlinear fitting in a way that conforms to the changing patterns of discrete points and controls the fluctuation range, a continuous trajectory curve that can truly reflect the decline trend is obtained. Then, by setting a slope threshold to optimize the smoothness of the curve and marking key information, the final activity decline trajectory map is clear, accurate and complete, providing an intuitive and reliable basis for subsequent matching with customer life cycle stages. The entire process has clear steps, specific standards, and strong reproducibility.

[0096] S4. Match the activity decline trajectory map with the customer's life cycle stages and integrate the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk.

[0097] In this embodiment of the invention, matching the activity decline trajectory with the customer's lifecycle stages includes:

[0098] Based on the preset enterprise customer lifecycle relationship, the customer lifecycle is divided into the introduction stage, growth stage, maturity stage and decline stage;

[0099] The trend of the activity decline trajectory is analyzed to identify the key inflection points in the activity decline trajectory.

[0100] Based on the timing and magnitude of the key inflection points, the activity decline trajectory is compared with the introduction, growth, maturity and decline phases to confirm the customer's current lifecycle stage.

[0101] The fusion of the customer's static attribute data yields a multi-dimensional profile of the customer's churn risk, including:

[0102] Obtain the industry attributes, years of cooperation, customer level, and historical value contribution data of the aforementioned customers;

[0103] The static attribute feature vector of the customer is obtained by synthesizing the industry attributes, the years of cooperation, the customer level and the historical value contribution data.

[0104] Multi-source feature fusion is performed on the static attribute feature vector and the stage feature vector of the current life cycle stage to obtain the customer's fused feature vector;

[0105] Based on the fused feature vector, the activity decline trajectory map is reconstructed into a multi-dimensional view to obtain a multi-dimensional profile of the customer's churn risk.

[0106] Based on the pre-defined customer cooperation cycle classification standards set by the enterprise, the time range of each stage of the customer life cycle is clearly defined. The cooperation period of 0 to 6 months is classified as the introduction stage, the cooperation period of 6 to 18 months is classified as the growth stage, the cooperation period of 18 to 36 months is classified as the maturity stage, and the cooperation period of more than 36 months is classified as the decline stage. The time node of each stage is calculated from the date when the customer and the enterprise sign the first cooperation agreement.

[0107] A comprehensive analysis of the frequency decline rate, interval decline rate, and channel decline rate values ​​corresponding to each time point in the activity decline trajectory chart is conducted. The changes in the three decline rates between each time point and the previous time point are calculated. When the absolute value of two or more of the changes in the three decline rates corresponding to a certain time point exceeds 20% of the decline rate value corresponding to the previous time point, and this trend continues to the next time point, that time point is determined as a key inflection point in the activity decline trajectory chart.

[0108] Typical key inflection point characteristics are extracted for each of the introduction, growth, maturity, and decline stages. Typical characteristics for the introduction stage include no more than one key inflection point and an absolute change value between 20% and 30%. Typical characteristics for the growth stage include one to two key inflection points and an absolute change value between 30% and 40%. Typical characteristics for the maturity stage include no more than one key inflection point and an absolute change value below 20%. Typical characteristics for the decline stage include at least two key inflection points, with at least one having an absolute change value exceeding 40%. The customer cooperation duration and change magnitude corresponding to the occurrence of the identified key inflection points are compared one by one with the typical characteristics of each lifecycle stage. If the match reaches 80% or higher, the lifecycle stage is confirmed as the customer's current lifecycle stage.

[0109] Through the enterprise customer information management system, we retrieve customer registration information, cooperation agreement documents, and transaction records. We extract the specific industry name of the customer as the industry attribute, calculate the cooperation period based on the difference between the first cooperation agreement signing date and the current date, and determine the customer level according to the enterprise customer rating standards, such as consumption amount, number of cooperation projects, and other comprehensive indicators. The customer level is divided into four levels: A, B, C, and D. We also collect the total transaction amount or project revenue generated by the customer in the past 12 months as historical value contribution data.

[0110] The extracted industry attributes are digitized and coded. The manufacturing industry is coded as 1, the service industry as 2, the financial industry as 3, and other industries as 4. The cooperation period is converted into specific months, such as 1 year and 3 months as 15. Customer level A is coded as 4, B as 3, C as 2, and D as 1. The historical value contribution data retains the original monetary value. The coded industry attributes, cooperation period, customer level, and historical value contribution data are arranged in order to form a one-dimensional array of four values. This array is the static attribute feature vector of the customer.

[0111] Obtain the stage feature vector corresponding to the current life cycle stage. The preset stage feature vector is [1,0,0,0] for the introduction stage, [0,1,0,0] for the growth stage, [0,0,1,0] for the maturity stage, and [0,0,0,1] for the decline stage. Multiply each value in the static attribute feature vector with the value at the corresponding position in the stage feature vector and add them together to obtain a new value. Calculate the values ​​at the four positions to form a new one-dimensional array. This array is the customer's fused feature vector.

[0112] The system integrates four new display dimensions: industry attribute code, number of months of cooperation, customer level code, and historical value contribution amount from the feature vector. These dimensions are combined with the original frequency decline rate, interval decline rate, and channel decline rate dimensions. Independent numerical labels and change curves are added to each dimension in the activity decline trajectory map. At the same time, the correlation between each dimension is clarified, such as the synchronous change of historical value contribution amount and frequency decline rate. This forms a visual chart that contains multi-dimensional customer information and decline trend. This chart is a multi-dimensional profile of customer churn risk.

[0113] The beneficial effects include: defining customer lifecycle stages through clear timeframe standards, ensuring the uniformity and operability of stage division; identifying key inflection points through specific change thresholds and duration conditions, ensuring the accuracy of inflection point identification; confirming the current lifecycle stage based on a high degree of consistency between typical characteristics of each stage and actual inflection point characteristics, improving the accuracy of stage matching; obtaining static attribute feature vectors through system extraction and digital encoding, ensuring the integrity and standardization of data; achieving multi-source feature fusion through numerical operations at corresponding positions, enabling the fused feature vectors to comprehensively reflect the integrated information of customer static attributes and lifecycle stages; expanding the trajectory map based on the fused feature vectors to display dimensions and clarify correlations; and finally forming a comprehensive and highly visualized multi-dimensional profile of churn risk, providing a comprehensive and accurate basis for subsequent analysis of churn risk drivers.

[0114] S5. Based on the multi-dimensional profile of churn risk, perform decision factor analysis on the driving factors of customer churn risk, and conduct a comprehensive risk assessment on the root cause characteristics after analysis to obtain the churn risk level of the customer.

[0115] In this embodiment of the invention, the step of performing decision factor analysis on the driving factors of customer churn risk based on the multi-dimensional profile of churn risk, and conducting a comprehensive risk assessment on the root cause characteristics after analysis to obtain the customer's churn risk level includes:

[0116] Based on a pre-defined decision factor weight rule base, an importance analysis is performed on the multi-dimensional features of the multi-dimensional profile of churn risk to obtain the churn risk contribution of the multi-dimensional features.

[0117] Based on the contribution of the churn risk, a dominant factor analysis is performed on the churn risk drivers of the customer to obtain the root cause characteristics of the customer. The churn risk drivers are extracted from the churn risk multidimensional profile and include static attribute characteristics of the customer and dynamic characteristics of customer activity decline.

[0118] The risk quantification assessment of the root cause characteristics is performed to obtain the loss risk degree of the root cause characteristics;

[0119] The customer's churn risk level is obtained by associating and matching the churn risk level with a preset risk level rule base.

[0120] The formula for calculating the risk of churn is as follows:

[0121] ;

[0122] In the formula, Indicates the first The risk of loss of each root cause characteristic Indicates the first The contribution of each root cause characteristic to the risk of loss. This represents the total number of the root cause features. Represents the natural logarithm function. This represents the square root operation. This indicates a summation operation.

[0123] Churn risk drivers refer to a set of multi-dimensional features extracted from the multi-dimensional customer churn risk profile that can directly or indirectly trigger a decline in customer activity and generate churn risk. These are the core triggers leading to signs of customer churn and the core analytical object for assessing the level of customer churn risk. The specific dimensional features of churn risk drivers all come from the multi-dimensional customer churn risk profile, including two main categories: static customer attribute features and dynamic features of customer activity decline. Specifically, static attribute features include: industry attributes, years of cooperation, customer level, and historical value contribution; dynamic decline features include: frequency decline rate, interval decline rate, and channel decline rate.

[0124] Decision factors refer to the core features selected from the churn risk drivers, assigned preset weights, and used to quantify the contribution of churn risk. They are key decision-making bases for analyzing the root causes of customer churn and assessing the level of churn risk. All decision factors are subsets of the churn risk drivers. All seven dimensions of the churn risk drivers are consistent with the specific content of the churn risk drivers, namely: industry attributes, years of cooperation, customer level, historical value contribution, frequency decline rate, interval decline rate, and channel decline rate. The preset weights for each factor are: industry attributes 20%, years of cooperation 25%, customer level 20%, historical value contribution 15%, frequency decline rate 10%, interval decline rate 5%, and channel decline rate 5%.

[0125] The pre-defined decision factor weighting rule base clearly specifies the weight percentages of each dimension feature in the multi-dimensional profile of churn risk. Specifically, industry attribute weight is 20%, cooperation duration weight is 25%, customer level weight is 20%, historical value contribution weight is 15%, frequency decay rate weight is 10%, interval decay rate weight is 5%, and channel decay rate weight is 5%. Specific numerical values ​​for each dimension feature are extracted from the multi-dimensional profile of churn risk. For industry attribute, the digitized value is extracted (Manufacturing = 1, Service Industry = 2, Finance Industry = 3, Other Industries = 4). For cooperation duration, the value is extracted in months. For customer level, the digitized value is extracted (A Level = 4, B Level = 3, C Level = 2, D Level = 1). For historical value contribution, the total transaction amount or project revenue is extracted in yuan. For frequency decay rate, interval decay rate, and channel decay rate, the previously calculated change values ​​are extracted. The specific value of each dimension feature is multiplied by its corresponding weight percentage, and the result is rounded to two decimal places. This result represents the churn risk contribution of that dimension feature.

[0126] The preset screening threshold for churn risk contribution is 15%. Starting with the churn risk contribution corresponding to industry attributes, the churn risk contribution of each dimension feature is compared with 15% one by one, with the comparison accurate to two decimal places. If the churn risk contribution value of a certain dimension feature is greater than or equal to 15.00%, then that dimension feature is directly identified as the dominant factor of customer churn risk. After completing all comparisons according to the order of dimension features, all dimension features identified as dominant factors are collected. These dimension features together constitute the root cause features of the customer.

[0127] For each root cause feature, first obtain its corresponding churn risk contribution value. Add this value to 1 to obtain a sum. Select the natural constant 2.71828 as the benchmark. By consulting the natural logarithm table or by stepping through the powers of 2.71828, find the exponent that the power of the sum equals the sum. This exponent is the natural logarithm of the sum. Simultaneously, by verifying whether the product of non-negative values ​​equals the churn risk contribution of the root cause feature, find the non-negative values ​​that meet the condition. This value is the square root of the churn risk contribution. Multiply the obtained natural logarithm by the square root and retain three decimal places to obtain the intermediate calculated value for the root cause feature. After all the intermediate calculated values ​​for the root cause features have been calculated, add these intermediate calculated values ​​to obtain the intermediate sum. Divide the intermediate calculated value for each root cause feature by the intermediate sum and retain three decimal places. This result is the churn risk level of the root cause feature.

[0128] The pre-defined risk level rule base clearly defines four risk levels corresponding to churn risk ranges: 0.000 to 0.200 is low risk, 0.200 to 0.500 is medium risk, 0.500 to 0.800 is high risk, and 0.800 to 1.000 is extremely high risk. Starting with the churn risk of the first root cause feature, the churn risk of each root cause feature is compared one by one with the ranges corresponding to the four risk levels. The risk level corresponding to the root cause feature is determined based on the range in which the value falls. After the risk levels of all root cause features are determined, the upper limit values ​​of the ranges corresponding to all risk levels are listed. The risk level corresponding to the largest upper limit value is selected as the customer's churn risk level.

[0129] The beneficial effects are as follows: By pre-setting clear and fixed weight proportions for each dimension of features, and clarifying the extraction standards and calculation requirements for specific numerical values ​​of each dimension of features, the calculation process for the contribution of churn risk is ensured to be uniform, reproducible, and accurate and unambiguous. Using a clear screening threshold of 15.00% as the criterion for determining the dominant factor, and comparing them one by one in a fixed order, subjective assumptions in the root cause feature identification process are avoided, ensuring the objectivity and fairness of the selection of dominant factors. By clarifying the acquisition methods of the natural constant benchmark, logarithm, and square root results, as well as the calculation accuracy requirements for each step, the calculation process of churn risk is broken down into operable, purely textual steps, ensuring the accuracy and reproducibility of risk quantification assessment. A risk level rule base is constructed with clear interval divisions and explicit inclusion rules, comparing them sequentially and selecting the highest risk level, making the determination of churn risk level systematic and the results clear. The entire process is detailed and standardized, with clear operational basis and result requirements for each step, significantly improving the scientificity and accuracy of churn risk level assessment, and providing accurate and reliable risk basis for the generation of subsequent personalized intervention strategies.

[0130] S6. Map the root cause characteristics and the churn risk level to a preset strategy intervention library, and combine the static attribute data to generate a personalized intervention strategy for the customer.

[0131] In this embodiment of the invention, mapping the root cause characteristics and the churn risk level to a preset strategy intervention library, and combining the static attribute data to generate a personalized intervention strategy for the customer, includes:

[0132] Logical association parsing is performed on the preset strategy intervention library to obtain the intervention matching rules of the strategy intervention library;

[0133] Based on the intervention matching rules, the root cause characteristics and the churn risk level are correlated and mapped to obtain the basic intervention strategy for the customer.

[0134] Based on the customer level and industry attributes in the static attribute data, the key parameters of the basic intervention strategy are corrected to obtain the personalized intervention strategy for the customer.

[0135] The pre-set strategy intervention library stores basic intervention strategy entries corresponding to various root cause characteristics and different churn risk levels. Each entry clearly indicates the applicable root cause characteristic type, churn risk level range, and corresponding specific basic intervention measures. During parsing, all strategy entries in the library are reviewed one by one to clarify the combination conditions of root cause characteristics and churn risk levels under each entry, as well as the basic intervention measures that uniquely correspond to the combination conditions. For example, when the root cause characteristic is a short cooperation period and the churn risk level is low, the corresponding basic intervention measure is to extend the cooperation preferential period. When the root cause characteristic is a fast frequency decline rate and the churn risk level is medium, the corresponding basic intervention measure is a high-frequency interactive incentive activity. The correspondence between the combination conditions and basic intervention measures corresponding to all entries is systematically organized to form clear and explicit intervention matching rules. The rules must fully cover all possible combinations of root cause characteristics and churn risk levels.

[0136] Extract the previously determined root cause characteristics and churn risk levels of customers. According to the combination conditions in the intervention matching rules, first, accurately compare each root cause characteristic of the customer with the preset root cause characteristic type in the rules. Then, match the customer's churn risk level with the corresponding churn risk level range in the rules. Find the combination conditions that completely match both the customer's root cause characteristics and churn risk level. The basic intervention measures associated with this combination condition in the rules are the basic intervention strategies corresponding to the customer's single root cause characteristic. If the customer has multiple root cause characteristics, integrate the basic intervention strategies corresponding to each root cause characteristic to ensure that all relevant basic intervention measures are included, forming a customer basic intervention strategy containing complete intervention content.

[0137] Customer levels and industry attributes are extracted from static attribute data. Customer levels are preset to four levels: A, B, C, and D. Industry attributes are preset to four categories: manufacturing, service industries, financial industries, and other industries. Customer level correction parameters are also preset: for A-level customers, the service resource allocation ratio in the basic intervention strategy is increased by 30%; for B-level customers, by 20%; for C-level customers, by 10%; and for D-level customers, the basic allocation remains unchanged. Industry attribute correction schemes are preset: manufacturing customers receive industry-specific technical support services; service industry customers receive service process optimization guidance; financial industry customers receive additional services such as fund security guarantees; and other industry customers receive general customized consulting services. The corresponding resource allocation increase ratio is determined based on the customer's actual customer level, and corresponding additional services are matched according to the customer's industry attribute. These correction contents are integrated into the basic intervention strategy, adjusting key parameters such as service resource allocation and service content supplementation. The adjusted intervention strategy is the customer's personalized intervention strategy.

[0138] The beneficial effects are as follows: by systematically sorting out the entries in the strategy intervention library, a unique correspondence between the root cause characteristics, the combination of churn risk levels, and the basic intervention strategies is clearly defined. The resulting intervention matching rules are clear and executable, ensuring the accuracy of the basic intervention strategy mapping. Based on the customer's actual root cause characteristics and churn risk level, precise matching and integration are performed to ensure that the basic intervention strategies fully cover the customer churn drivers. Combined with the resource allocation ratio corresponding to the customer level and the additional service plan corresponding to the industry attributes, the basic intervention strategies are specifically corrected. This makes the final personalized intervention strategy not only fit the root cause and risk level of customer churn but also adapt to the customer's personalized attributes, significantly improving the targeting and effectiveness of marketing interventions. The entire process is clearly defined, standardized, and highly reproducible, providing strong support for accurately curbing customer churn.

[0139] like Figure 2 The diagram shown is a functional block diagram of a marketing intervention system based on customer churn prediction provided in an embodiment of the present invention.

[0140] The marketing intervention system 100 based on customer churn prediction described in this invention can be installed in an electronic device. Depending on the functions implemented, the marketing intervention system 100 based on customer churn prediction may include an abnormal fluctuation detection module 101, a state evolution analysis module 102, a time-series trajectory fitting module 103, a risk profile construction module 104, a risk assessment and rating module 105, and an intervention strategy generation module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0141] In this embodiment, the functions of each module / unit are as follows:

[0142] The abnormal fluctuation detection module 101 is used to detect abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data.

[0143] The state evolution analysis module 102 is used to perform state evolution analysis on the churn symptom data to obtain the activity decline characteristics of the churn symptom data.

[0144] The time-series trajectory fitting module 103 is used to perform time-series trajectory fitting on the churn symptom data based on the activity decline characteristics to obtain the customer's activity decline trajectory map.

[0145] The risk profile construction module 104 is used to match the activity decline trajectory map with the customer's life cycle in stages and integrate the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk.

[0146] The risk assessment and rating module 105 is used to analyze the decision factors driving the customer's churn risk based on the multi-dimensional profile of churn risk, and to conduct a comprehensive risk assessment on the root cause characteristics after analysis to obtain the customer's churn risk level.

[0147] The intervention strategy generation module 106 is used to map the root cause characteristics and the churn risk level to a preset strategy intervention library, and combine the static attribute data to generate a personalized intervention strategy for the customer.

[0148] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0149] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0150] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0151] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0152] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A marketing intervention method based on customer churn prediction, characterized in that, The method includes: S1. Detect abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data; S2. Perform state evolution analysis on the churn symptom data to obtain the activity decay characteristics of the churn symptom data, including: Heterogeneous feature extraction is performed on the churn symptom data to obtain the frequency characteristics, interval characteristics, and channel diversity characteristics of the churn symptom data; The frequency feature, the interval feature, and the channel diversity feature are sorted in a multidimensional order according to time sequence to obtain the customer's feature time sequence. The change trend analysis of the characteristic time series is performed to obtain the frequency decay rate, interval decay rate and channel decay rate of the characteristic time series. The frequency decay rate, the interval decay rate, and the channel decay rate are integrated into the activity decay feature of the churn symptom data; S3. Based on the activity decline characteristics, perform time-series trajectory fitting on the churn symptom data to obtain the customer's activity decline trajectory map, including: A multidimensional coordinate system for the customer is constructed with time as the horizontal axis and the multidimensional feature vector of the activity decline feature as the vertical axis. The activity decay feature is mapped to the multidimensional coordinate system to obtain the discrete time-series state points of the activity decay feature; Nonlinear fitting is performed on the discrete time-series state points to obtain the continuous decay trajectory curve of the discrete time-series state points; The smoothness of the continuous decline trajectory curve is optimized, and key information is annotated on the optimized continuous decline trajectory curve to obtain the customer's activity decline trajectory map. S4. Match the activity decline trajectory map with the customer's life cycle stages and integrate the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk. S5. Based on the multi-dimensional profile of churn risk, perform decision factor analysis on the driving factors of customer churn risk, and conduct a comprehensive risk assessment on the root cause characteristics after analysis to obtain the customer's churn risk level, including: Based on a pre-defined decision factor weight rule base, an importance analysis is performed on the multi-dimensional features of the multi-dimensional profile of churn risk to obtain the churn risk contribution of the multi-dimensional features. Based on the contribution of the churn risk, a dominant factor analysis is performed on the churn risk drivers of the customer to obtain the root cause characteristics of the customer. The churn risk drivers are extracted from the churn risk multidimensional profile and include static attribute characteristics of the customer and dynamic characteristics of customer activity decline. The risk quantification assessment of the root cause characteristics is performed to obtain the loss risk degree of the root cause characteristics; The churn risk level is matched with a preset risk level rule base to obtain the customer's churn risk level; S6. Map the root cause characteristics and the churn risk level to a preset strategy intervention library, and combine them with the static attribute data to generate a personalized intervention strategy for the customer, including: Logical association parsing is performed on the preset strategy intervention library to obtain the intervention matching rules of the strategy intervention library; Based on the intervention matching rules, the root cause characteristics and the churn risk level are correlated and mapped to obtain the basic intervention strategy for the customer. Based on the customer level and industry attributes in the static attribute data, the key parameters of the basic intervention strategy are corrected to obtain the personalized intervention strategy for the customer.

2. The marketing intervention method based on customer churn prediction as described in claim 1, characterized in that, The process of detecting abnormal fluctuations in customer interaction behavior data to obtain customer churn indicator data includes: Obtain customers' historical behavior data and interaction behavior data; The historical behavior data is analyzed for routine behavior to obtain the customer's historical statistical baseline; Based on the historical statistical baseline, a difference comparison analysis is performed on the interaction behavior data to obtain the behavior deviation index of the interaction behavior data; Based on the behavioral deviation index, routine behavioral data is removed from the interaction behavior data to obtain the customer churn symptom data.

3. The marketing intervention method based on customer churn prediction as described in claim 1, characterized in that, The step of matching the activity decline trajectory with the customer's lifecycle stages includes: Based on the preset enterprise customer lifecycle relationship, the customer lifecycle is divided into the introduction stage, growth stage, maturity stage and decline stage; The trend of the activity decline trajectory is analyzed to identify the key inflection points in the activity decline trajectory. Based on the timing and magnitude of the key inflection points, the activity decline trajectory is compared with the introduction, growth, maturity and decline phases to confirm the customer's current lifecycle stage.

4. The marketing intervention method based on customer churn prediction as described in claim 3, characterized in that, The fusion of the customer's static attribute data yields a multi-dimensional profile of the customer's churn risk, including: Obtain the industry attributes, years of cooperation, customer level, and historical value contribution data of the aforementioned customers; The static attribute feature vector of the customer is obtained by synthesizing the industry attributes, the years of cooperation, the customer level and the historical value contribution data. The static attribute feature vector and the stage feature vector of the current life cycle stage are fused using multi-source features to obtain the customer's fused feature vector; Based on the fused feature vector, the activity decline trajectory map is reconstructed into a multi-dimensional view to obtain a multi-dimensional profile of the customer's churn risk.

5. The marketing intervention method based on customer churn prediction as described in claim 1, characterized in that, The formula for calculating the risk of churn is as follows: ； In the formula, Indicates the first The risk of loss of each root cause characteristic Indicates the first The contribution of each root cause characteristic to the risk of loss. This represents the total number of the root cause features. Represents the natural logarithm function. This represents the square root operation. This indicates a summation operation.

6. A marketing intervention system based on customer churn prediction, characterized in that, The system for implementing the marketing intervention method based on customer churn prediction as described in claim 1 includes: The abnormal fluctuation detection module is used to detect abnormal fluctuations in customer interaction behavior data to obtain customer churn symptom data. The state evolution analysis module is used to perform state evolution analysis on the churn symptom data to obtain the activity decay characteristics of the churn symptom data, including: Heterogeneous feature extraction is performed on the churn symptom data to obtain the frequency characteristics, interval characteristics, and channel diversity characteristics of the churn symptom data; The frequency feature, the interval feature, and the channel diversity feature are sorted in a multidimensional order according to time sequence to obtain the customer's feature time sequence. The change trend analysis of the characteristic time series is performed to obtain the frequency decay rate, interval decay rate and channel decay rate of the characteristic time series. The frequency decay rate, the interval decay rate, and the channel decay rate are integrated into the activity decay feature of the churn symptom data; A time-series trajectory fitting module is used to perform time-series trajectory fitting on the churn symptom data based on the activity decline characteristics to obtain the customer's activity decline trajectory map, including: A multidimensional coordinate system for the customer is constructed with time as the horizontal axis and the multidimensional feature vector of the activity decline feature as the vertical axis. The activity decay feature is mapped to the multidimensional coordinate system to obtain the discrete time-series state points of the activity decay feature; Nonlinear fitting is performed on the discrete time-series state points to obtain the continuous decay trajectory curve of the discrete time-series state points; The smoothness of the continuous decline trajectory curve is optimized, and key information is annotated on the optimized continuous decline trajectory curve to obtain the customer's activity decline trajectory map. The risk profile building module is used to match the activity decline trajectory map with the customer's life cycle stages and integrate the customer's static attribute data to obtain a multi-dimensional profile of the customer's churn risk. The risk assessment and rating module is used to analyze the decision factors driving the customer's churn risk based on the multi-dimensional churn risk profile, and to conduct a comprehensive risk assessment of the root cause characteristics after analysis to obtain the customer's churn risk level, including: Based on a pre-defined decision factor weight rule base, an importance analysis is performed on the multi-dimensional features of the multi-dimensional profile of churn risk to obtain the churn risk contribution of the multi-dimensional features. Based on the contribution of the churn risk, a dominant factor analysis is performed on the churn risk drivers of the customer to obtain the root cause characteristics of the customer. The churn risk drivers are extracted from the churn risk multidimensional profile and include static attribute characteristics of the customer and dynamic characteristics of customer activity decline. The risk quantification assessment of the root cause characteristics is performed to obtain the loss risk degree of the root cause characteristics; The churn risk level is matched with a preset risk level rule base to obtain the customer's churn risk level; An intervention strategy generation module is used to map the root cause characteristics and the churn risk level to a preset strategy intervention library, and combine the static attribute data to generate a personalized intervention strategy for the customer, including: Logical association parsing is performed on the preset strategy intervention library to obtain the intervention matching rules of the strategy intervention library; Based on the intervention matching rules, the root cause characteristics and the churn risk level are correlated and mapped to obtain the basic intervention strategy for the customer. Based on the customer level and industry attributes in the static attribute data, the key parameters of the basic intervention strategy are corrected to obtain the personalized intervention strategy for the customer.

Images