Customer loss early warning feedback method, system, equipment and product based on customer behavior analysis
By using historical data on customer delivery volume and employing Fourier transform and STL time series decomposition algorithms to identify seasonal cycles and trend terms, the problem of high false alarm and false negative rates in existing customer churn warning systems has been solved, enabling accurate identification and effective retention of customer churn risks.
Patent Information
- Application Number
- CN202511781841.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-29
- Publication Date
- 2026-03-20
AI Technical Summary
Existing customer churn warning systems rely on manual judgment or rigid rules, which are difficult to adapt to the behavioral differences of different customer groups, resulting in high false alarm and false alarm rates and an inability to accurately identify customer churn risks.
By acquiring historical data on customer delivery volume, Fourier transform is used to identify major seasonal cycles. The STL time series decomposition algorithm is combined to extract trend and seasonal items, mark potential churned customers, and label negative customers through customer communication to implement operational intervention strategies.
It enables accurate identification and timely alerts of customer churn risks, improves the success rate of customer retention and the scientific nature of operational intervention strategies, and reduces false alarm and missed alarm rates.
Smart Images
Figure CN121707607A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data analysis, specifically to a customer churn early warning and feedback method, system, device, and product based on customer behavior analysis. Background Technology
[0002] For courier companies, a small number of high-value customers with large delivery volumes contribute the vast majority of their revenue; losing these customers would cause a significant drop in company income. At the same time, the cost of acquiring new customers is significantly higher than retaining existing customers. Therefore, establishing a customer churn early warning mechanism is particularly necessary for courier companies. In the context of intensifying competition in the courier industry, effective customer churn early warning to prevent customer loss is a key business management activity that integrates financial health, operational optimization, and risk control, directly impacting the profitability and market competitiveness of courier companies.
[0003] Existing customer churn warning systems primarily rely on manual assessment of customer transaction volume through business reports, employing a subjective approach to determine customer churn risk. This is particularly challenging for complex data with unclear patterns. Some CRM (Customer Relationship Management) systems offer transaction-based warning rules and can proactively alert users when transaction volume decreases; however, these rules are rigid and fail to adapt to the behavioral differences among various customer groups. Therefore, existing customer churn warning methods suffer from high false alarm and false negative rates. Summary of the Invention
[0004] The purpose of this application is to overcome the shortcomings of the prior art and provide a customer churn early warning and feedback method, system, device and product based on customer behavior analysis. It can automatically analyze customer churn risk based on historical customer delivery data, provide accurate and timely customer churn risk reminders, and thus implement proactive customer retention strategies.
[0005] Firstly, this application provides a customer churn early warning and feedback method based on customer behavior analysis, the technical solution of which includes the following steps: S1, obtain historical data on customer delivery volume to obtain time-series data on delivery volume; S2, seasonality identification of delivery volume time series data to obtain the main seasonal cycles; S3, based on the main seasonal cycle, uses the STL time series decomposition algorithm to perform characteristic decomposition on the delivery volume time series data to obtain the trend term, seasonal term and residual term of the delivery volume time series data; S4, based on the acquired trend items, marks customers as potential churn customers; S5 pushes early warning information about potential churned customers to account managers, conducts customer communication, and labels potential churned customers with negative tags based on the results of customer communication. S6. Implement operational intervention strategies for customer retention for customers who are labeled negatively.
[0006] By adopting the above technical solution, this application can classify customer behavior types based on historical data of customer delivery volume, and use the STL time series decomposition algorithm based on the classification to use trend items as the basis for judging customer churn risk, thereby realizing the churn risk judgment adapted to customer behavior type and implementing proactive operational intervention strategies, which can significantly improve the accuracy and timeliness of customer churn risk identification and the success rate of customer recovery.
[0007] As a priority, in S2, the main seasonal cycles for obtaining delivery volume time-series data specifically include the following steps: S201, Perform Fourier transform on the delivery quantity time series data to obtain the delivery quantity frequency domain data; S202, calculate the modulus of each frequency component f of the delivery quantity frequency domain data to obtain the amplitude spectrum containing the amplitude information of each frequency component f; S203 converts the frequency component into a period T, where T = 1 / f; S204 maps the period T to the amplitude information, and the period corresponding to the maximum peak amplitude is the main seasonal period.
[0008] By adopting the above technical solution, the signal strength (amplitude) of the customer's delivery volume time series data at different frequency components can be calculated based on the delivery volume time series data, that is, the significance of the frequency component in the original time series data can be obtained, and the main seasonal cycle of the customer's delivery volume can be obtained through the maximum peak amplitude.
[0009] As a priority, select several periods T with the highest peak amplitude, and manually determine at least one major seasonal period based on the customer's business type.
[0010] By adopting the above technical solution, when the main seasonal cycle patterns of customers are relatively complex, based on the periodic analysis of customer business types by manual analysis, several main seasonal cycles are selected for subsequent time series decomposition calculations.
[0011] As a preferred option, in S3, the STL time series decomposition algorithm is used to perform characteristic decomposition on the delivery volume time series data, which specifically includes the following steps: S301, execute the k-th cycle, subtract the final trend term of the previous cycle from the original delivery time series data, the initial final trend term is 0; S302, the detrended express delivery volume time series data is split into multiple seasonal subsequences according to the main seasonal cycle, each seasonal subsequence is independently smoothed by LOESS to obtain smoothed seasonal values, and the smoothed seasonal values of each subsequence are concatenated to obtain the temporary seasonal term. S303 performs low-pass filtering on the temporary seasonal term to extract the low-frequency component; S304, subtract the low-frequency component from the temporary seasonal term to obtain the final seasonal term for k rounds; S305, Subtract the final seasonal term of round k from the original delivery time series data to obtain the deseasonal term of round k; S306, perform LOESS smoothing on the final seasonal term of round k to obtain the final trend term of round k; then execute the (k+1)th round of the loop until the preset value is reached.
[0012] By using the above technical solution, and after obtaining the main seasonal cycles, the final trend item and the final seasonal item are obtained through the STL time series decomposition algorithm, which can be used as the basis for judging customer churn risk and calculating operational intervention strategies.
[0013] Preferably, in S4, a time-series trend curve for the trend item is plotted, and customers whose time-series trend curve is pointing downwards or whose time-series trend curve fluctuates beyond a threshold are marked as potential churned customers.
[0014] The above technical solution allows for a clear description of the objective and true delivery volume trend of customers after removing periodic and residual fluctuations through time-series trend curves, thereby enabling the identification of potential churned customers.
[0015] As a preferred option, in S5, if any of the following occurs during communication with a customer: invitation request is rejected, customer feedback or complaints are received, shipping cost is discussed, or a competitor is mentioned, then the customer will be labeled negatively.
[0016] Through the above technical solution, based on the specific content of communication with customers, some stable customers marked as potential churn customers are filtered out, while customers who are truly at risk of churn are labeled negatively.
[0017] As a preferred embodiment, in S6, the operational intervention strategy specifically involves calculating freight discounts for peak seasons based on the seasonal data of delivery volume and providing customers with price discount schemes.
[0018] By using the above technical solutions to calculate freight discounts based on seasonal items of delivery volume time-series data, a balance between business volume and discounts can be achieved, improving the scientific nature of operational intervention strategies.
[0019] Secondly, the customer churn early warning and feedback system based on customer behavior analysis proposed in this application adopts the following technical solution: It includes a delivery volume data collection module, a seasonality identification module, an STL time-series decomposition module, and a customer operation and maintenance module; The delivery volume data collection module collects historical delivery volume data from customers and draws a time-series data graph of delivery volume. The seasonality identification module obtains the main seasonal cycles of customer delivery volume time series data based on historical delivery volume data. The STL time series decomposition module performs time series decomposition on the delivery volume time series data based on the main seasonal cycle to obtain the trend item, seasonal item and residual item of the delivery volume time series data. The customer operations module marks customers as potential churn customers based on the trend items of delivery volume time series data, pushes early warning information to account managers, conducts customer communication, marks customers with negative labels based on the results of customer communication, and implements operational intervention strategies for customer retention for customers with negative labels.
[0020] Thirdly, a computer device according to this application adopts a technical solution including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned customer churn early warning and feedback method based on customer behavior analysis.
[0021] Fourthly, a computer program product of this application employs a technical solution including a computer program or instructions, which enables the computer program or instructions to implement the steps in the above-mentioned customer churn early warning feedback method based on customer behavior analysis.
[0022] In summary, the beneficial effects of this application are as follows: Based on historical customer delivery volume data, it can differentiate customer delivery behavior patterns according to major seasonal cycles, and calculate trend items through time-series decomposition according to different major seasonal cycles. This allows it to adapt to customers with different behavior patterns, intuitively and accurately perceive the true business trends of customers, and improve the accuracy of customer churn prediction. Furthermore, the application's determination of operational intervention strategies based on the seasonal items obtained from time-series decomposition calculations can achieve a balance between customer recovery and cost control, improve the scientific nature of operational intervention strategies, and ensure the healthy and sustainable development of the business. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating a customer churn early warning and feedback method based on customer behavior analysis, as described in an embodiment of this application. Figure 2 This is a flowchart illustrating step S2 of a customer churn early warning and feedback method based on customer behavior analysis in an embodiment of this application. Figure 3 This is a flowchart illustrating step S3 of a customer churn early warning and feedback method based on customer behavior analysis in an embodiment of this application. Figure 4 This is a schematic diagram of the architecture of a customer churn early warning and feedback system based on customer behavior analysis, as described in an embodiment of this application. Figure 5 This is a schematic diagram of the architecture of an exemplary computer device in an embodiment of this application. Detailed Implementation
[0024] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of this application.
[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. It should be noted that in the optional embodiments of this application, the object information and other related data involved require the permission or consent of the object when the embodiments of this application are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. That is to say, if the embodiments of this application involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations, and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.
[0026] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.
[0027] Customers in the express delivery industry often exhibit significant differences in their delivery patterns based on their business types and work habits. For example, they can be categorized as follows: Regular daily delivery customers collect and send packages daily, with daily delivery volume fluctuating according to their business cycles; low-frequency cyclical customers deliver packages in batches weekly or monthly, with lower frequency but potentially larger single shipments, such as remote factory customers shipping concentrated shipments weekly; and flexible peak-hour customers, whose delivery frequency and volume fluctuate greatly due to seasonality or marketing activities, with relatively consistent daily delivery volumes but a surge in frequency and volume on specific dates (such as monthly member days for platform merchants).
[0028] Because delivery volume curves for different types of customers exhibit entirely different characteristics, it's difficult to truly capture the characteristics of customer churn risk using either manual judgment or a CRM system's fixed threshold-based alerts. This results in high false positive and false negative rates. For example, for low-frequency or highly volatile but loyal customers, triggering thresholds during off-peak periods may lead to misjudgments of churn risk, wasting customer maintenance resources and potentially negatively impacting customer experience. For high-volume customers, decreased willingness to cooperate manifests as a gradual decline in delivery volume as they seek alternative courier companies. However, because the cyclical pattern may not be detected manually or the system's preset thresholds are not yet reached, customer churn risk remains difficult to identify. For high-frequency customers, a sudden decrease in delivery volume or even cessation of cooperation may occur, but if the average decrease in delivery volume over the detection period does not reach the threshold, customer churn risk alerts become delayed, missing the optimal opportunity for customer maintenance and retention.
[0029] This application provides a customer churn early warning and feedback method based on customer behavior analysis, used to capture the true business trends of customers with different behavioral types, thereby accurately identifying customer churn risk. Please refer to... Figure 1 Specifically, it includes the following steps.
[0030] S1: Obtain historical data on customer delivery volume to obtain time-series data on delivery volume.
[0031] For the existing express delivery system, all express deliveries must have their waybills scanned and entered to ensure that the waybill data corresponds to the express delivery itself. Therefore, the delivery volume data can be considered to be real, reliable, and clean data with no obvious anomalies. The fluctuations in the data originate from the real fluctuations in customer delivery volume demand.
[0032] S2 performs seasonality identification on delivery volume time-series data to obtain the main seasonal cycles.
[0033] Although seasonal cycles can be inferred manually based on domain knowledge and the client's industry and business model, relying solely on manual inference has significant limitations. Overly subjective judgments can easily overlook subtle seasonal cycle patterns, and the need to check seasonal cycles for different time windows one by one leads to inefficiency, making it difficult to apply to large-scale data analysis.
[0034] Please see Figure 2 In S2 of this application embodiment, obtaining the main seasonal cycle frequency of delivery volume time series data specifically includes the following steps.
[0035] S201, perform FFT (Fourier Transform) calculation on the delivery volume time-series data to obtain the delivery volume frequency domain data. Through Fourier Transform, the data on delivery volume changing over time can be converted into a description of the various frequencies of the delivery volume and their relative importance. The following equation applies:
[0036] In the formula, F(f) is a function of the frequency domain data of the delivery quantity, and f(t) is the original function of the time series data of the delivery quantity. It is a complex exponential function used as a frequency analyzer, where t0 is the start time in the time domain and t1 is the end time in the time domain.
[0037] For example, if the sampled data points are daily data, it means that the sampling interval dt is 1 day. Correspondingly, for customers with low-frequency delivery, the sampled data points can be defined as the interval between one delivery.
[0038] The result of performing an FFT calculation is an array of complex numbers, with each complex number corresponding to a frequency component. Then, based on the total number of data points N and the sampling interval dt, the actual frequency value f corresponding to each frequency component can be obtained using the np.fft.fftfreq function. Taking the sampled data points as daily data as an example, the unit is "period / day".
[0039] S202, calculate the magnitude f of each frequency component of the delivered quantity frequency domain data to obtain the amplitude spectrum containing the amplitude information of each frequency component. The amplitude of the frequency component reflects the intensity of the frequency signal in the original data, that is, the magnitude of the influence of the frequency on the original data.
[0040] S203 converts the frequency component into a period T, where T = 1 / f.
[0041] We will still use the daily delivery volume data as the sampling data point for explanation. If a frequency component f = 1 / 7 period / day, its period T = 7 days, that is, the period is a week.
[0042] S204 maps the period T to the amplitude information, and the period corresponding to the maximum peak amplitude is the main seasonal period.
[0043] It should also be noted that customer delivery behavior often involves more than one cyclical pattern. This means that besides the cyclical component with the strongest amplitude peak, other secondary peaks may also be meaningful cyclical patterns, such as semi-annual cycles or seasonal cycles. Therefore, in the embodiments of this application, several cycles T with the highest amplitude peaks are selected. By manually combining cycle T with business knowledge and domain analysis, at least one major seasonal cycle is finally determined. Typically, multiple meaningful major seasonal cycles can be determined.
[0044] S3, based on the main seasonal cycle, uses the STL time series decomposition algorithm to perform characteristic decomposition on the delivery volume time series data, and obtains the trend term, seasonal term and residual term of the delivery volume time series data.
[0045] Given a fixed seasonal period, input the STL algorithm, determine the total number of cycles, and then perform characteristic decomposition. Other preset parameters of the STL algorithm can be set conventionally and optimized based on the results of the characteristic decomposition. Please refer to [link to relevant documentation]. Figure 3 Specifically, it includes the following steps.
[0046] S301, execute the k-th cycle, subtract the final trend term from the previous cycle from the original delivery time series data, the initial final trend term is 0.
[0047] S302, the detrended express delivery volume time series data is split into multiple seasonal subsequences according to the main seasonal cycle, each seasonal subsequence is independently smoothed by LOESS to obtain smoothed seasonal values, and the smoothed seasonal values of each subsequence are concatenated to obtain the temporary seasonal term.
[0048] S303 performs low-pass filtering on the temporary seasonal term to extract the low-frequency components.
[0049] S304, subtract the low-frequency component from the temporary seasonal term to obtain the final seasonal term for k rounds.
[0050] S305, subtract the final seasonal term of round k from the original delivery time series data to obtain the de-seasonal term of round k.
[0051] S306, perform LOESS smoothing on the final seasonal term of round k to obtain the final trend term of round k; then execute the (k+1)th round of the loop until the preset value is reached.
[0052] Meanwhile, in one specific embodiment, the convergence of the time decomposition of the STL algorithm can be determined by observing whether the residual terms are approximately random. If they do not converge, the parameters of the STL algorithm are tuned.
[0053] After completing all rounds of the STL algorithm and confirming the convergence of the time series decomposition, the final trend term and final seasonal term are obtained, which are the trend term and seasonal term of the delivery volume time series data. The trend term, after removing the influence of the seasonal term and residual term, represents the true feedback on customer delivery volume and has good objectivity and reliability. Different trend terms are obtained for each of the major seasonal periods, enabling analysis of user delivery volume across different periodic dimensions.
[0054] S4, based on acquired trend items, marks customers as potential churn customers.
[0055] More specifically, in the embodiments of this application, a time-series trend curve is plotted using trend items to visually observe user delivery volume in a graphical form. Ideally, the time-series trend curve should be a smooth curve. In one case, if the overall direction of the time-series trend curve is downward, it indicates that after excluding the interference of seasonal and residual items, the customer's delivery volume is generally declining, and they should be marked as potential churned customers for follow-up. In another case, if the fluctuation range of the time-series trend curve exceeds a threshold, exhibiting non-ideal behavior, it indicates that the customer's delivery volume is uncertain in the short term, signaling short-term business fluctuations, and they should also be marked as potential churned customers for follow-up.
[0056] S5 sends alerts to account managers, facilitates communication with clients, and assigns negative labels to clients based on the results of these communications.
[0057] The purpose of customer communication is twofold: to maintain customer relationships and to assess the risk of customer churn through actual communication. Positive customer feedback, smooth communication, or explanations regarding changes in delivery volume trends are considered positive signals, allowing for positive labeling of the customer and suspending further operational intervention strategies. Conversely, negative signals include rejected invitations, customer complaints, discussions about shipping costs, or mentions of competitors, requiring negative labeling of the customer. Customers with negative labels have a higher risk of churn, necessitating proactive operational intervention strategies for customer retention.
[0058] S6. Implement operational intervention strategies for customer retention for customers who are labeled negatively.
[0059] In the embodiments of this application, the operational intervention strategy can specifically involve calculating freight discounts for peak seasons based on the seasonal component of delivery volume time-series data and offering price incentives to customers. Typically, in the express delivery industry, a target delivery volume is pre-agreed upon in the contract signing process, based on communication with customers, and a corresponding agreed-upon discount is offered based on that target volume. However, in operational intervention strategies, due to considerations of maintaining customer relationships, it is difficult to implement discount schemes by pre-agreing on target delivery volumes. In the embodiments of this application, based on the seasonal component of the acquired delivery volume time-series data, the seasonal cycle of increased customer delivery demand can be perceived. Based on this, freight discounts can be calculated, and price incentive schemes can be proactively proposed. This effectively reduces the risk of customer churn and provides a basis for calculating freight discounts, balancing costs and incentives, making the operational intervention strategy more scientific and reasonable.
[0060] In other embodiments, operational intervention strategies for customer retention that can be implemented also include fully sharing the characteristic decomposition results of delivery volume time-series data, fully communicating with customers on trend items and seasonal items, conducting analysis and discussion, providing data references for customers' business status at the data layer, and empowering customer service to improve customer satisfaction.
[0061] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0062] Please see Figure 4 An embodiment of this application provides a customer churn early warning and feedback system based on customer behavior analysis, which includes a delivery volume data collection module 1, a seasonality identification module 2, an STL time series decomposition module 3, and a customer operation and maintenance module 4.
[0063] The delivery volume data collection module 1 collects historical delivery volume data from customers and draws a time-series data chart of delivery volume.
[0064] Seasonal identification module 2 obtains the main seasonal cycles of customer delivery volume time series data based on historical delivery volume data.
[0065] STL Time Series Decomposition Module 3 performs time series decomposition on delivery volume time series data based on the main seasonal cycle, and obtains the trend term, seasonal term and residual term of delivery volume time series data.
[0066] Customer Operations Module 4 marks customers as potential churn customers based on the trend items of delivery volume time series data, pushes early warning information to account managers, conducts customer communication, marks customers with negative labels based on the results of customer communication, and implements operational intervention strategies for customer retention for customers with negative labels.
[0067] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the customer churn early warning and feedback system based on customer behavior analysis described above can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here.
[0068] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0069] Embodiments of this application also provide a computer device, which may be a server, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores delivery volume time-series data obtained from historical customer delivery volume data, the main seasonal cycles obtained through seasonal identification of the delivery volume time-series data, and trend, seasonal, and residual terms obtained through time-series decomposition of the delivery volume time-series data. The network interface is used for communication with external terminals via a network connection. When the processor executes the computer program, it implements the aforementioned customer churn early warning feedback based on customer behavior analysis.
[0070] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A customer churn early warning and feedback method based on customer behavior analysis, characterized in that, Includes the following steps: S1, obtain historical data on customer delivery volume to obtain time-series data on delivery volume; S2, seasonality identification of delivery volume time series data to obtain the main seasonal cycles; S3, based on the main seasonal cycle, uses the STL time series decomposition algorithm to perform characteristic decomposition on the delivery volume time series data to obtain the trend term, seasonal term and residual term of the delivery volume time series data; S4, based on the acquired trend items, marks customers as potential churn customers; S5 pushes early warning information about potential churned customers to account managers, conducts customer communication, and labels potential churned customers with negative tags based on the results of customer communication. S6. Implement operational intervention strategies for customer retention for customers who are labeled negatively.
2. The customer churn early warning and feedback method based on customer behavior analysis according to claim 1, characterized in that, In S2, obtaining the main seasonal cycles of delivery volume time-series data specifically includes the following steps: S201, Perform Fourier transform on the delivery quantity time series data to obtain the delivery quantity frequency domain data; S202, calculate the modulus of each frequency component f of the delivery quantity frequency domain data to obtain the amplitude spectrum containing the amplitude information of each frequency component f; S203 converts the frequency component into a period T, where T = 1 / f; S204 maps the period T to the amplitude information, and the period corresponding to the maximum peak amplitude is the main seasonal period.
3. The customer churn early warning and feedback method based on customer behavior analysis according to claim 2, characterized in that, Select several periods T with the highest amplitude ranking, and manually determine at least one major seasonal period based on the customer's business type.
4. The customer churn early warning and feedback method based on customer behavior analysis according to claim 1, characterized in that, In S3, the STL time series decomposition algorithm is used to perform characteristic decomposition on the delivery volume time series data, which specifically includes the following steps: S301, execute the k-th cycle, subtract the final trend term of the previous cycle from the original delivery time series data, the initial final trend term is 0; S302, the detrended express delivery volume time series data is split into multiple seasonal subsequences according to the main seasonal cycle, each seasonal subsequence is independently smoothed by LOESS to obtain smoothed seasonal values, and the smoothed seasonal values of each subsequence are concatenated to obtain the temporary seasonal term. S303 performs low-pass filtering on the temporary seasonal term to extract the low-frequency component; S304, subtract the low-frequency component from the temporary seasonal term to obtain the final seasonal term for k rounds; S305, Subtract the final seasonal term of round k from the original delivery time series data to obtain the deseasonal term of round k; S306, perform LOESS smoothing on the final seasonal term of round k to obtain the final trend term of round k; then execute the (k+1)th round of the loop until the preset value is reached.
5. The customer churn early warning and feedback method based on customer behavior analysis according to claim 1, characterized in that, In S4, plot the time-series trend curve for the trend item, and mark customers whose time-series trend curve is pointing downwards or whose time-series trend curve fluctuates beyond a threshold as potential churn customers.
6. The customer churn early warning and feedback method based on customer behavior analysis according to claim 1 is characterized in that, In S5, if a communication with a customer includes any of the following: an invitation request being rejected, a customer complaint, a discussion about shipping costs, or a mention of a competitor, then the customer will be labeled negatively.
7. The customer churn early warning and feedback method based on customer behavior analysis according to claim 1 is characterized in that, In S6, the operational intervention strategy specifically involves calculating freight discounts for peak seasons based on the seasonal data of delivery volume and providing customers with price discount schemes.
8. A customer churn early warning and feedback system based on customer behavior analysis, characterized in that, It includes a delivery volume data collection module, a seasonality identification module, an STL time-series decomposition module, and a customer operation and maintenance module; The delivery volume data collection module collects historical delivery volume data from customers and draws a time-series data graph of delivery volume. The seasonality identification module obtains the main seasonal cycles of customer delivery volume time series data based on historical delivery volume data. The STL time series decomposition module performs time series decomposition on the delivery volume time series data based on the main seasonal cycle to obtain the trend item, seasonal item and residual item of the delivery volume time series data. The customer operations module marks customers as potential churn customers based on the trend items of delivery volume time series data, pushes early warning information to account managers, conducts customer communication, marks potential churn customers with negative labels based on the results of customer communication, and implements operational intervention strategies for customer retention for customers with negative labels.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the customer churn early warning and feedback method based on customer behavior analysis as described in any one of claims 1 to 7.
10. A computer program product, characterized in that, The computer program product includes a computer program or instructions that enable the computer program or instructions to perform the steps in the customer churn early warning and feedback method based on customer behavior analysis as described in any one of claims 1 to 7.