An intelligent e-commerce order dispatching and customer life cycle management system

Abstract

The application relates to the technical field of business data processing, in particular to an e-commerce intelligent order dispatching and customer life cycle management system, which comprises a customer data management module, which is used for collecting transaction data, behavior data and after-sales data of customers in an e-commerce platform and constructing a customer portrait; a life cycle division module, which is used for statistically obtaining the last consumption interval, consumption frequency and consumption amount of each customer based on the transaction data in a preset statistical window, obtaining a stage division index of each customer by weighted summation, clustering each customer into a plurality of customer groups based on the stage division index, and generating a life cycle stage identifier of each customer; an order generation module, which is used for generating a service order carrying the life cycle stage identifier in response to an order event, and generating a wake-up order for a dormant period customer; and an order dispatching decision module, which is used for determining the processing priority of the service order and the matching weight of each skill label according to the life cycle stage identifier.

Description

Technical Field

[0001] This invention relates to the field of business data processing, specifically to an e-commerce intelligent order dispatch and customer lifecycle management system. Background Technology

[0002] As e-commerce continues to expand, the number of work orders handled by customer service systems on e-commerce platforms, such as order inquiries and after-sales services, continues to grow. Customer service dispatch systems have become a key link connecting customer demands with customer service resources. At the same time, customer transaction behavior on the platform has obvious stage characteristics, and customers at different stages of their life cycle have significantly different service needs. Customer life cycle management is gradually being introduced into the operation and service scenarios of e-commerce platforms.

[0003] Existing customer service dispatch systems typically assign service orders to suitable customer service personnel based on factors such as the type and urgency of the order, as well as the skills and availability of customer service staff. Some systems also segment customers based on their historical transaction data, dividing them into lifecycle stages such as introduction, growth, maturity, and dormancy, for differentiated marketing and outreach.

[0004] However, in existing solutions, the customer lifecycle segmentation results are independent of the customer service dispatch process. The dispatch system cannot configure service resources differently based on the customer's stage. The customer stage segmentation boundaries are mostly preset fixed values ​​that cannot be corrected with changes in service quality, and the segmentation benchmark is prone to drift. Furthermore, in concurrent dispatch scenarios, when multiple work orders are directed to the same customer service personnel at the same time, there is a lack of a definite allocation decision and occupancy release mechanism, making it easy for customer service personnel to be repeatedly occupied or suspended for a long time.

[0005] Therefore, the present invention provides an e-commerce intelligent order dispatch and customer lifecycle management system. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by the present invention to solve its technical problem is: an e-commerce intelligent order dispatch and customer lifecycle management system, including a customer data management module, used to collect transaction data, behavior data and after-sales data of customers in the e-commerce platform, and build customer profiles;

[0008] The lifecycle segmentation module is used to calculate the last consumption interval, consumption frequency and consumption amount of each customer based on the transaction data in the preset statistical window, and obtain the stage segmentation index of each customer by weighted summation. Based on the stage segmentation index, each customer is clustered into multiple customer groups, and each customer group is divided into the introduction stage, growth stage, maturity stage or dormant stage, and a lifecycle stage identifier of each customer is generated.

[0009] The work order generation module is used to generate service work orders carrying the lifecycle stage identifier in response to order events, and to generate wake-up work orders for dormant customers;

[0010] The dispatch decision module is used to determine the processing priority of the service order and the matching weight of each skill tag according to the life cycle stage identifier, and to determine the matching candidate customer service personnel. The status of the candidate customer service personnel includes idle state, pre-occupancy state and in-process state.

[0011] The dispatching module determines the target customer service personnel based on the processing priority and the predicted load of each candidate customer service personnel, and sets them to a pre-occupancy state. When multiple service orders have the same target customer service personnel, the service order with the highest processing priority is assigned to that target customer service personnel and set to an in-process state. The remaining service orders re-determine the target customer service personnel from the remaining candidate customer service personnel who are in an idle state.

[0012] The feedback update module is used to record the processing results and customer evaluations of the service work orders, calculate the service deviation, and adjust the statistical window according to the positive correlation.

[0013] Preferably, the customer data management module constructs customer profiles in the following way:

[0014] S201. The collected transaction data, behavior data and after-sales data are grouped according to customer identifiers. Data belonging to the same customer identifier is deduplicated and missing fields are filled in to obtain the customer dataset for each customer.

[0015] S202. Extract transaction characteristics, behavioral characteristics and after-sales characteristics of each customer from the customer dataset. The transaction characteristics include the distribution of consumer product categories and the distribution of average order value. The behavioral characteristics include the distribution of browsing categories and the conversion rate of adding to cart. The after-sales characteristics include the distribution of after-sales types and the after-sales evaluation score.

[0016] S203. According to the preset tagging system, the transaction characteristics, behavioral characteristics and after-sales characteristics of each customer are mapped to the corresponding customer tags, and the customer tags of each customer constitute a customer profile.

[0017] Preferably, the lifecycle segmentation module divides each customer group into the introduction, growth, maturity, or dormancy phases in the following way:

[0018] Customers who generate transaction data for the first time within the statistical window are classified as the introductory period;

[0019] For each customer other than those in the introductory period, the last consumption interval, consumption frequency, and consumption amount are normalized respectively. The normalized last consumption interval, consumption frequency, and consumption amount are then weighted and summed according to a preset weighting coefficient to obtain the stage division index for each customer. The weighting coefficients corresponding to consumption frequency and consumption amount are positive, while the weighting term corresponding to the last consumption interval participates in the weighted summation with a negative sign.

[0020] Based on the distribution of the stage segmentation index of each customer, the customers are clustered into three customer groups; according to the mean of the stage segmentation index corresponding to each customer group from low to high, the customers in the three customer groups are successively divided into the dormant period, growth period and mature period.

[0021] Generate corresponding lifecycle stage identifiers for customers segmented into different lifecycles;

[0022] The lifecycle segmentation module sets up a buffer zone between adjacent customer groups. The buffer zone is the index interval between the mean values ​​of the stage segmentation indices corresponding to two adjacent customer groups. After the statistical window is adjusted, the lifecycle segmentation module re-statistics based on the adjusted statistical window to obtain the updated stage segmentation index for each customer, and re-clusters the customers based on the updated stage segmentation index to obtain the updated customer groups. For customers whose updated stage segmentation index falls within the buffer zone, their original lifecycle stage identifier is maintained. For customers whose updated stage segmentation index is higher than the upper limit of the buffer zone or lower than the lower limit of the buffer zone, their lifecycle stage identifier is migrated to the lifecycle stage corresponding to the customer group to which they belong after re-clustering.

[0023] Preferably, the work order generation module generates service work orders and wake-up work orders in the following way:

[0024] Obtain the customer identifier corresponding to the customer who initiated the order event, read the lifecycle stage identifier corresponding to the customer identifier, and generate a service ticket carrying the customer identifier, the lifecycle stage identifier, and the event type of the order event; the order event includes order placement events, consultation events, and after-sales events;

[0025] When a customer is classified as dormant, or when the customer's lifecycle stage identifier is migrated to the identifier corresponding to the dormant stage, wake-up content is generated based on the customer tag in the customer profile of that customer, and a wake-up work order carrying the customer identifier, the lifecycle stage identifier and the wake-up content is generated.

[0026] Preferably, the dispatch decision module determines the processing priority of the service work order in the following way:

[0027] Service orders whose lifecycle stage is identified as the mature or growth stage are assigned to the first priority, service orders whose lifecycle stage is identified as the introduction stage are assigned to the second priority, and service orders whose lifecycle stage is identified as the dormant stage and wake-up stage are assigned to the third priority. The first priority is higher than the second priority, and the second priority is higher than the third priority.

[0028] For multiple service orders with the same priority, they are processed and sorted according to the order in which they were generated.

[0029] Preferably, the dispatch decision module determines the matching candidate customer service personnel in the following way:

[0030] The skill tags include consultation and answering skill tags, after-sales service skill tags, and marketing communication skill tags;

[0031] The matching weights of each skill tag are configured according to the lifecycle stage identifiers. Specifically, for service tickets corresponding to the introduction stage identifier, the matching weight of consultation and answering skill tags is greater than that of after-sales processing skill tags; for service tickets corresponding to the growth stage identifier and the maturity stage identifier, the matching weight of after-sales processing skill tags is greater than that of consultation and answering skill tags; for service tickets corresponding to the dormant stage identifier, the matching weight of marketing communication skill tags is greater than that of consultation and answering skill tags and after-sales processing skill tags.

[0032] The skill matching degree between each skill tag and each customer service personnel is weighted and summed according to the matching weight of each skill tag to obtain the comprehensive matching score of each customer service personnel. Customer service personnel are selected as candidate customer service personnel in descending order of the comprehensive matching score.

[0033] Preferably, the dispatching module determines the target customer service personnel by combining the predicted workload of each candidate customer service personnel, specifically as follows:

[0034] Obtain the pending service orders of the candidate customer service personnel. According to the event type of each pending service order, retrieve the historical average processing time of the corresponding event type. Determine the remaining processing time of each pending service order as the difference between the historical average processing time of each pending service order and the processing time of the pending service order. Determine the predicted load of the candidate customer service personnel as the sum of the remaining processing times of each pending service order.

[0035] The candidate customer service representative with the lowest predicted load is identified as the target customer service representative and placed in a pre-occupancy state.

[0036] Preferably, the dispatch execution module further includes:

[0037] When the dispatching module sets the target customer service personnel to a pre-occupancy state, it generates a corresponding occupancy period. If the service order is assigned to the target customer service personnel within the occupancy period, the target customer service personnel's status is changed from the pre-occupancy state to the in-process state. If the occupancy period expires and the service order has not been assigned to the target customer service personnel, the target customer service personnel's status is released from the pre-occupancy state to the idle state.

[0038] When re-determining the target customer service representative among the remaining candidate customer service representatives who are in an idle state for the remaining service work orders, if there are no remaining candidate customer service representatives in an idle state, the remaining service work orders are placed into the waiting queue. When a candidate customer service representative is found to be in an idle state, the process of determining the target customer service representative is re-executed for the service work orders in the waiting queue according to the processing priority.

[0039] Preferably, the feedback update module calculates the service deviation and adjusts the statistical window by a magnitude positively correlated with it, specifically as follows:

[0040] Calculate the first relative deviation between the actual processing time of the service order and the historical average processing time of historical service orders of the same event type, and the second relative deviation between the customer evaluation score of the service order and the average customer evaluation score of historical service orders. Then, perform a weighted summation of the first relative deviation and the second relative deviation to obtain the service deviation amount of the service order.

[0041] The correction amount of the statistical window is determined based on the average service deviation of each service work order. The correction amount is then added to the original length of the statistical window. The correction amount is positively correlated with the average service deviation.

[0042] Preferably, the feedback update module further transfers service resource quotas based on the processing result of the wake-up work order, specifically as follows:

[0043] The number of wake-up work orders that did not trigger a return flow is counted among the wake-up work orders corresponding to the same dormant customer to obtain the wake-up failure rate of that customer. The "not triggering a return flow" means that the customer did not generate transaction data in the statistics window after the wake-up work order was processed.

[0044] The service resource quota corresponding to the customer is reduced according to the number of wake-up failures, and the reduced service resource quota is transferred to the service work orders of the growth customers. The service resource quota includes the quota for the number of work orders generated for the customer in the unit statistics window and the quota for the work order dispatch priority.

[0045] This invention provides an intelligent order dispatching and customer lifecycle management system for e-commerce. It has the following beneficial effects:

[0046] 1. This invention obtains a stage segmentation index by weighted summation of the last consumption interval, consumption frequency, and consumption amount. The segmentation boundary is generated by clustering according to the index distribution. The segmentation boundary is adaptively generated according to the overall distribution of customers. First-time transaction customers are excluded from the index calculation to avoid their statistics from skewing the overall segmentation benchmark. The stage segmentation is closer to the actual status of customers.

[0047] 2. This invention sets up a buffer zone between adjacent customer groups with the average value of the two groups as the range. Customer migration must cross the average value of the adjacent customer group to take effect. Minor fluctuations caused by the adjustment of the statistical window will not trigger the stage identifier to migrate back and forth, avoiding oscillation of matching weights, and customers obtain a stable order dispatch strategy.

[0048] 3. This invention corrects the statistical window length according to the mean of the service deviation. When the service fluctuates greatly, the window is extended to make the statistics tend to be stable. When the service is stable, the window is maintained to keep the response speed, thus avoiding the drift of the stage division benchmark with the fluctuation of service quality. Attached Figure Description

[0049] Figure 1 This is a system flowchart of the present invention;

[0050] Figure 2 This is a schematic diagram of the dispatch execution process of the present invention;

[0051] Figure 3 This is a schematic diagram of the customer service personnel status transition process of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] like Figures 1 to 3 As shown, this invention proposes an e-commerce intelligent order dispatching and customer lifecycle management system, comprising:

[0054] The customer data management module is used to collect transaction data, behavioral data, and after-sales data of customers on the e-commerce platform to build customer profiles.

[0055] The lifecycle segmentation module is used to calculate the last consumption interval, consumption frequency and consumption amount of each customer based on the transaction data in the preset statistical window, and to obtain the stage segmentation index of each customer by weighted summation. Based on the stage segmentation index, each customer is clustered into multiple customer groups, and each customer group is divided into the introduction stage, growth stage, maturity stage or dormant stage, generating the lifecycle stage identifier of each customer.

[0056] The statistics window is a pre-configured statistics window of the system. Its initial length can be configured to, for example, ninety days, and is dynamically adjusted according to the correction amount determined by the feedback update module.

[0057] The work order generation module is used to generate service work orders carrying lifecycle stage identifiers in response to order events, and to generate wake-up work orders for dormant customers;

[0058] The dispatch decision module is used to determine the processing priority of service orders and the matching weight of each skill tag according to the life cycle stage identifier, and to determine the matching candidate customer service personnel. The status of the candidate customer service personnel includes idle, pre-occupancy and in progress.

[0059] The dispatching module determines the target customer service personnel based on the processing priority and the predicted load of each candidate customer service personnel, and sets them to a pre-occupancy state. When multiple service orders have the same target customer service personnel, the service order with the highest processing priority is assigned to that target customer service personnel and set to an in-process state. The remaining service orders re-determine the target customer service personnel from the remaining candidate customer service personnel who are in an idle state.

[0060] The feedback update module is used to record the processing results of service work orders and customer evaluations, calculate the service deviation, and adjust the statistics window according to the positive correlation.

[0061] As an optional embodiment, after collecting transaction data, behavioral data, and after-sales data, the customer data management module first aggregates them according to customer identifiers.

[0062] It should be noted that the same customer may have multiple scattered records on an e-commerce platform. For example, transaction data may come from the order system, behavioral data may come from page tracking, and after-sales data may come from the customer service system. If these are directly mixed and statistically analyzed, the behavior of the same customer may be split into different records.

[0063] S201. Therefore, in this embodiment, the customer identifier is used as the primary key, and data from different systems are mounted under the same customer identifier. For data reported repeatedly under the same customer identifier (such as repeated browsing records caused by page retries), only the first record is retained. For data with missing fields, the average value of the corresponding field in the same type of historical data of the same customer is used to fill in the missing fields, so as to obtain a complete and non-repeating customer dataset for each customer, and avoid interference caused by missing data or duplicate counting to subsequent stage division.

[0064] S202. After obtaining the customer dataset, extract three types of features. Transaction features include the distribution of product categories and the distribution of average order value. The product category distribution reflects customer category preferences, such as the percentage of purchases made by customers in primary categories; the average order value distribution reflects customer spending levels. Behavioral features include the distribution of browsing categories and the add-to-cart conversion rate. Calculate according to the following formula:

[0065] ;in, To improve add-to-cart conversion rate, To count the number of times customers add items to their shopping cart within the window, The number of times the items were actually placed after being added to the cart is shown; both are measured in units of "times".

[0066] After-sales features include the distribution of after-sales types and after-sales evaluation scores, where the after-sales evaluation score is the rating value submitted by the customer after the after-sales process is completed. The reason for dividing the features into three categories—transaction, behavior, and after-sales—is that the lifecycle segmentation module mainly consumes transaction-related statistics, while the feedback update module mainly consumes after-sales evaluation data. Categorized extraction allows each downstream module to use the data as needed, without having to repeatedly parse the full dataset.

[0067] S203. The preset tagging system includes category preference tags, consumption level tags, activity level tags, and after-sales sensitivity tags; the tagging system can be expanded according to business needs.

[0068] The customer data management module maps the features extracted by S202 to corresponding tags one by one. For example, the category with the largest proportion in the distribution of consumer goods is mapped to the customer's category preference tag, and the after-sales evaluation score is mapped to the after-sales sensitivity tag. All customer tags for each customer constitute the customer profile. Thus, the customer profile is no longer a static accumulation of registration information, but a set of readable tags extracted from multi-source behaviors. The dispatch decision module can directly call the corresponding tags when configuring skill matching weights, and at the same time, it provides an identifiable customer object for the "reawakening work order generation for dormant customers" section in the manual.

[0069] As an optional implementation, the lifecycle segmentation module first performs the initial transaction determination for customers within the statistical window: iterates through the transaction data of each customer, and determines the customer whose earliest transaction time falls within the current statistical window as a new customer and directly assigns them to the introductory period.

[0070] It should be noted that the statistics such as the last purchase interval and purchase frequency of new customers have not yet formed a meaningful distribution. If they are mixed with the statistics of old customers for calculation, it will skew the overall division benchmark. Therefore, this embodiment will introduce the early customers to be cut off first and not participate in the subsequent index calculation and clustering.

[0071] For all customers except those in the introductory period, the interval between last purchases, purchase frequency, and purchase amount were normalized to eliminate differences in units and magnitudes. For example, each statistic was divided by its maximum value among all customers, ensuring that the normalized values ​​were between 0 and 1. Then, the stage segmentation index for each customer was calculated using the following formula. :

[0072] ;in, Indices for dividing stages, , , These are the normalized consumption frequency, consumption amount, and interval between the last consumption, all of which are dimensionless quantities. , , These are preset positive weighting coefficients.

[0073] The negative coefficient for the last purchase interval is because a longer interval indicates a greater tendency for customer churn, and its contribution to the segmentation is opposite to that of purchase frequency and purchase amount. The specific values ​​of the three coefficients can be configured according to business priorities; for example, they can be increased when focusing on repeat purchase operations. .

[0074] After obtaining the stage segmentation index for each customer, the customers are clustered according to the distribution of the index. In one optional implementation, the K-means clustering algorithm is used, with the number of clusters set to 3, corresponding to the three lifecycle stages other than the introduction period; the clustering uses the stage segmentation index of each customer as a sample, and customers with similar indices are grouped into the same customer group.

[0075] After clustering, the mean of the stage segmentation index corresponding to the three customer groups is calculated. The three means are sorted, and the customers of the customer group with the smallest mean are classified as dormant, the customers of the customer group with the middle mean are classified as growth, and the customers of the customer group with the largest mean are classified as mature.

[0076] It should be noted that the division boundary in this embodiment is not a preset fixed value or ratio, but is adaptively generated by the distribution of customer index: when the feedback update module adjusts the length of the statistical window, the statistics, stage division index and their distribution change accordingly, the cluster boundary moves accordingly, and the customer life cycle stage identifier migrates accordingly, thus forming a self-correcting closed loop with the feedback update mechanism described later in the specification.

[0077] The lifecycle stage identifiers are numerical identifiers, with the introduction stage, growth stage, maturity stage, and dormant stage corresponding to identifier values ​​1, 2, 3, and 4, respectively. The lifecycle segmentation module writes the identifier values ​​into the corresponding customer's customer data. When generating a service work order, the work order generation module reads the identifier value of the initiating customer and includes it in the service work order. The dispatch decision module determines the processing priority of the service work order based on the identifier value and configures the matching weight of each skill tag. In other embodiments, the lifecycle stage identifiers can also be character identifiers (such as S1 to S4) or text identifiers (such as "introduction stage"). This embodiment does not limit this.

[0078] It should also be noted that customers in the introductory phase are in the stage of establishing consumption habits and require guiding services; customers in the growth and maturity phases contribute the main transactions and their service response needs to be prioritized; customers in the dormant phase are at risk of churn and need to be encouraged to return through reactivation services; in order to facilitate the order dispatching decision module to allocate service resources differently according to the customer's stage, the life cycle segmentation module classifies each customer into the introductory, growth, maturity or dormant phase based on the transaction data in the statistical window;

[0079] Furthermore, after each adjustment of the statistical window, the stage segmentation index of each customer is updated accordingly. Customers near the boundary of adjacent customer groups may have their index fluctuate repeatedly between the two adjacent customer groups due to slight fluctuations in the statistical quantity. If the life cycle stage identifier is migrated immediately each time the segmentation is redefined, the customer's stage identifier will change frequently, and the matching weight configured in the dispatch decision module will switch repeatedly. The same customer's two consecutive service work orders will be assigned to customer service personnel with significantly different skill directions, resulting in unstable dispatch results.

[0080] Therefore, in this embodiment, the lifecycle segmentation module sets up a buffer zone between adjacent customer groups. The buffer zone is the index interval between the average values ​​of the stage segmentation indices corresponding to two adjacent customer groups. Taking a dormant customer group and a growth customer group as an example, let the average value of the stage segmentation index corresponding to the dormant customer group be... The mean of the stage segmentation index corresponding to the growth stage customer group is , Less than The buffer zone between the two customer groups is then... to The exponential range between; where, , All are dimensionless quantities.

[0081] After the statistical window is adjusted, the lifecycle segmentation module re-statistics based on the adjusted statistical window to obtain the updated stage segmentation index for each customer, and then re-clusters the customers based on the updated stage segmentation index to obtain the updated customer groups. For customers whose updated stage segmentation index falls within the buffer zone, their original lifecycle stage identifier is maintained. For customers whose updated stage segmentation index is higher than the upper limit of the buffer zone or lower than the lower limit of the buffer zone, their lifecycle stage identifier is migrated to the lifecycle stage corresponding to the customer group to which they belong after re-clustering. The buffer zone is updated accordingly with the mean of each customer group after re-clustering.

[0082] For example, a customer who was originally in a dormant customer group, after the statistical window was adjusted, had an updated stage segmentation index that exceeded [a certain threshold]. But still smaller If the value falls within the buffer zone, the customer maintains the identifier corresponding to the dormant period; when its updated stage division index exceeds... At that time, its lifecycle stage identifier is migrated to the identifier corresponding to the growth stage;

[0083] Similarly, growth-stage customers need their updated stage segmentation index to be lower than [a certain value]. The customer migration is initiated during the dormant period. Therefore, the migration of a customer stage only takes effect after exceeding the mean of the adjacent customer groups. The buffer is adaptively generated according to the clustering results. Minor fluctuations in statistics caused by the adjustment of the statistical window will not trigger the back-and-forth migration of the stage identifier, avoiding oscillations in the matching weights. This ensures that the same customer receives a stable order dispatch strategy within adjacent statistical periods. Customers in the introductory period participate in the calculation and clustering of the stage division index in the first statistical window afterward and migrate to the corresponding customer group according to the clustering results. There is no buffer between the introductory period and adjacent customer groups, and the migration takes effect immediately.

[0084] As an optional embodiment, order events include order placement events, consultation events, and after-sales events. The order placement event is pushed by the order system after the customer submits the order, the consultation event is pushed by the conversation system after the customer initiates a consultation conversation, and the after-sales event is pushed by the after-sales system after the customer submits a return, exchange, or complaint application.

[0085] The work order generation module responds to the order event by first parsing the account information carried in the order event and mapping the account information to the corresponding customer identifier; then, using the customer identifier as an index, it reads the customer's lifecycle stage identifier stored in the customer data management module; and finally, it generates a service work order.

[0086] The service order fields include at least a customer identifier field, a lifecycle stage identifier field, and an event type field. The event type field records whether the order event is an order placement event, a consultation event, or an after-sales event.

[0087] Furthermore, the reason for writing the lifecycle stage identifier into the work order field instead of having it temporarily queried by the dispatch decision module is that the customer's stage identifier may be migrated due to feedback updates after the work order is generated. Using the identifier at the time of work order generation as the standard can ensure that the processing priority and skill matching weight corresponding to the same work order remain consistent within its processing cycle, avoiding policy jumps in the middle of processing.

[0088] The generation of wake-up work orders does not depend on order events, but is triggered by the results of lifecycle division.

[0089] Furthermore, when the lifecycle segmentation module classifies a customer into the dormant period, or re-counts the data based on the adjusted statistical window and migrates the customer's lifecycle stage identifier to the identifier value corresponding to the dormant period (such as identifier value 4), it sends a wake-up trigger information to the work order generation module. The wake-up trigger information carries the customer identifier of the corresponding customer.

[0090] The work order generation module responds to the wake-up trigger information by reading the customer profile of the customer and generating wake-up content based on the customer tags in the customer profile. For example, it matches the corresponding category's discount information or service follow-up content based on the category preference tags. Then, it generates a wake-up work order, which carries a wake-up content field in addition to the customer identification field and the lifecycle stage identification field.

[0091] The field structure of wake-up work orders is consistent with that of service work orders. Therefore, after wake-up work orders enter the dispatch decision module, they are subject to the same priority determination and candidate customer service personnel determination logic as ordinary service work orders.

[0092] Specifically, in the above-mentioned work order generation methods, ordinary service work orders are driven by order events, and wake-up work orders are driven by stage division results. The two types of trigger entry points cover two service scenarios: customer-initiated and system-initiated. The lifecycle stage identifier is written into the work order field at the time of work order generation, so that the dispatch strategy corresponding to the work order remains stable within its processing cycle, and avoids the strategy jump caused by stage migration during processing.

[0093] The wake-up content is generated based on customer tags in the customer profile, so that the wake-up service matches the customer's preferences. The wake-up work order and the ordinary service work order use the same field structure. The two types of work orders reuse the same set of processing logic in the dispatch decision and dispatch execution stages, which reduces the processing branches and maintenance costs of the system.

[0094] As an optional implementation, the priority of processing is determined based on the contribution of various types of customers to revenue and the risk of churn: customers in the growth stage and mature stage are the main users of platform transactions, and their order placement, consultation and after-sales requests directly affect the completion of transactions. Therefore, service tickets with the corresponding life cycle stage identifier values ​​2 and 3 are determined as the first priority; customers in the introduction stage are in the stage of establishing consumption habits, and service experience affects their retention and conversion. Therefore, service tickets with the corresponding identifier value 1 are determined as the second priority; wake-up tickets are services that the system actively triggers for dormant customers. They are not requests initiated by customers in real time, and the timeliness requirements are relatively relaxed. Therefore, service tickets with the corresponding identifier value 4 and wake-up tickets are determined as the third priority. The first priority is higher than the second priority, and the second priority is higher than the third priority.

[0095] Multiple service orders with the same priority are processed and sorted according to the order in which they were generated, i.e., the service orders generated earlier are assigned first.

[0096] It should be noted that when multiple service orders correspond to the same target customer service personnel, the dispatching module determines the allocation order based on processing priority; for service orders with the same processing priority, the allocation order is determined based on the generation time, thus ensuring that the allocation order among any number of concurrent service orders is uniquely determined.

[0097] As an optional embodiment, the skill tags for customer service personnel include consultation and answering skill tags, after-sales processing skill tags, and marketing communication skill tags. Among them, consultation and answering skill tags include specific tags such as product consultation and logistics tracking; after-sales processing skill tags include specific tags such as return and exchange processing and complaint handling; and marketing communication skill tags include specific tags such as event promotion and customer follow-up.

[0098] The skill tags for each customer service representative are pre-configured by the administrator or generated based on the representative's historical work order processing records. For example, the skill tags corresponding to the work order types with the highest processing volume for a particular customer service representative are configured as that representative's skill tags. The skill matching degree between each skill tag and each customer service representative is used to characterize the customer service representative's proficiency level with the corresponding skill tag. Its value can be determined based on the customer service representative's historical processing volume and historical evaluation scores for the corresponding type of work order.

[0099] The dispatch decision module configures the matching weights of each skill tag according to the lifecycle stage identifier carried in the service order, and the configuration direction corresponds to the service needs of the customer at the current stage:

[0100] For customers in the introductory phase, their main needs are for product understanding and order placement guidance. Therefore, for service orders with a value of 1, the matching weight of the consultation and answering skill tags should be set higher than that of the after-sales handling skill tags. For customers in the growth and maturity phases, their needs are concentrated on order fulfillment and after-sales service. Therefore, for service orders with values ​​of 2 and 3, the matching weight of the after-sales handling skill tags should be set higher than that of the consultation and answering skill tags. For customers in the dormant phase, the reactivation service orders aim to guide customer return. Therefore, for reactivation service orders with a value of 4, the matching weight of the marketing communication skill tags should be set higher than that of the other two skill tags.

[0101] The specific values ​​of each matching weight can be pre-configured. For example, the matching weight of the primary skill category can be configured as 0.6, the secondary skill category as 0.3, and the other categories as 0.1. This embodiment does not limit the specific values.

[0102] After configuring the matching weights for each skill tag, the dispatch decision module calculates the comprehensive matching score for each customer service representative using the following formula. : ;

[0103] in, This is a dimensionless overall matching score for customer service personnel, and its value is only used for relative ranking among different customer service personnel. The skill match degree between the k-th skill tag and the customer service personnel is dimensionless. Let k be the matching weight for the k-th skill tag. It is dimensionless, and k is the category number of the skill tag.

[0104] Furthermore, the skill matching degree between the k-th skill tag and the customer service personnel. The specific calculation method is as follows: ,in These are preset weighting coefficients, dimensionless, with values ​​ranging from 0 to 1; M represents the number of service orders corresponding to the kth skill tag processed by the customer service representative within the statistics window, in units of times; M represents the total number of all service orders processed by the customer service representative within the statistics window, in units of times. The historical average rating score for the service tickets corresponding to the kth skill tag handled by this customer service representative, in points; This represents the upper limit of the evaluation score, expressed in points. This represents the percentage of experience the customer service representative has in handling skill category k. This represents the service quality level of the customer service representative in skill type k. Both are dimensionless quantities, and the result is obtained by weighting. It is also a dimensionless quantity. For customer service personnel who have no historical processing records in the statistics window, the skill matching degree corresponding to their various skill tags is configured according to the preset initial value and updated according to the above formula as their historical processing records accumulate.

[0105] Then, the customer service personnel are sorted from highest to lowest according to their comprehensive matching scores, and the top N customer service personnel are selected as candidate customer service personnel, for example, the top three customer service personnel.

[0106] It should be noted that when the customer's lifecycle stage identifier is migrated with the feedback update, the dispatch decision module re-executes the above weight configuration and candidate determination process according to the migrated stage identifier, so that the selection of candidate customer service personnel is adjusted as the customer's stage evolves.

[0107] It should be noted that the idle state indicates that the customer service representative is not currently handling any service orders to be processed, and can be identified as the target customer service representative for a new service order; the pre-occupied state indicates that the customer service representative has been identified as the target customer service representative for a certain service order but has not yet started processing it, and the service order is in the processing buffer stage of pending confirmation or pending access; the in-process state indicates that the customer service representative is processing an assigned service order.

[0108] When determining the target customer service personnel, the dispatching module needs to consider not only the processing priority but also the current workload of each candidate customer service personnel. If the target is determined solely based on the skill matching degree, customer service personnel with high skill matching degree will be continuously selected, and the actual number of work orders they are handling will continue to accumulate, ultimately leading to a longer waiting time for work orders.

[0109] Therefore, this embodiment introduces a predicted workload to characterize the estimated time that candidate customer service personnel will need to process pending service orders, so that the determination of target customer service personnel takes into account both skill matching direction and current workload.

[0110] Predicted load Calculate according to the following formula: ;

[0111] in, The predicted workload for candidate customer service personnel, in seconds; The historical average processing time for the event type corresponding to the j-th in-process service order of the candidate customer service representative, in seconds; This represents the processing time (in seconds) for the j-th pending service order of the candidate customer service representative; j is the sequence number of the pending service order. Let be the remaining processing time for the j-th in-process service order. The sum of the remaining processing times is the estimated time required for the candidate customer service representative to complete all in-process service orders.

[0112] When the processing time of an ongoing service order exceeds the historical average processing time for the corresponding event type, the remaining processing time of the ongoing service order is counted as zero.

[0113] The candidate customer service representative with the lowest predicted load is identified as the target customer service representative and set to a pre-occupancy state.

[0114] The historical average processing time is obtained by statistically analyzing the historical processing results recorded by the feedback update module: according to the event type of the service work order, the actual processing time of the same type of closed service work orders is averaged to obtain the historical average processing time corresponding to order placement events, consultation events and after-sales events, and is updated on a rolling basis as the number of closed service work orders increases.

[0115] Specifically, this can be further explained using the following example: If a candidate customer service representative currently has one inquiry service order and one after-sales service order in the pending order status, and the historical average processing time for the inquiry service order is 300 seconds and the processed time is 120 seconds, and the historical average processing time for the after-sales service order is 600 seconds and the processed time is 100 seconds, then the predicted workload of this candidate customer service representative is... .

[0116] When determining the target customer service personnel based on processing priority and matching weight of each skill tag, the dispatching module combines the predicted load of each candidate customer service personnel: the candidate customer service personnel with the lowest predicted load is determined as the target customer service personnel; when multiple candidate customer service personnel have the same predicted load, the target customer service personnel are determined in descending order of comprehensive matching score. After determining the target customer service personnel, their status is changed from idle to pre-occupied.

[0117] As an optional embodiment, after the dispatching module sets the target customer service personnel to a pre-occupied state, the target customer service personnel will no longer be in an idle state in the status query. If the service work order fails to be actually assigned due to network abnormalities, session interruption or other reasons, the corresponding target customer service personnel will remain in the pre-occupied state for a long time and will not be able to accept new service work orders, resulting in the invalid suspension of the customer service personnel. Therefore, this embodiment generates a corresponding occupancy time limit when setting to the pre-occupied state to constrain the duration of the pre-occupied state, as follows.

[0118] The time limit is a preset duration, such as 30 seconds, or it can be configured according to the processing priority of the service order. For example, the time limit for a service order with the first priority is longer than the time limit for a wake-up order with the third priority. If the service order is assigned to the target customer service personnel within the time limit, that is, when the service order processing session is established or the processing record is generated, the dispatching module will change the status of the target customer service personnel from the pre-occupied state to the in-process state, and the time limit will expire.

[0119] The time limit for each customer service representative is configured according to their respective pre-occupancy fulfillment ratio: The feedback update module counts the number of times each customer service representative's pre-occupancy status has been migrated to the in-process status and the total number of times they have been set to the pre-occupancy status within the set statistical period, and the ratio of the two is determined as the pre-occupancy fulfillment ratio of that customer service representative; The order dispatching module configures the time limit for each customer service representative according to the pre-occupancy fulfillment ratio. The smaller the pre-occupancy fulfillment ratio, the shorter the corresponding time limit.

[0120] It should be noted that a low pre-occupancy fulfillment rate indicates that the customer service representative's pre-occupancy is frequently not fulfilled due to session establishment failures, customer interruptions, or other reasons. During the pre-occupancy period, the representative cannot accept new service orders. Shortening the pre-occupancy period can release the representative to an idle state more quickly, reducing the ineffective locking of customer service resources. Customer service representatives with a high pre-occupancy fulfillment rate have high allocation reliability and are configured with a relatively long pre-occupancy period, allowing them sufficient time for allocation confirmation.

[0121] If the occupancy period expires and the service order has not been assigned to the target customer service personnel, the dispatching module will release the target customer service personnel from the pre-occupancy state to the idle state, allowing the customer service personnel to re-enter the state where they can be identified as target customer service personnel.

[0122] Furthermore, in scenarios with high concurrency, there may be situations where, when other service work orders are re-selecting target customer service personnel from the remaining candidate customer service personnel who are in an idle state, there are no remaining candidate customer service personnel in an idle state.

[0123] In one optional implementation, the dispatching module places the service orders for which no target customer service representative has been identified into a waiting queue and records the processing priority and generation time of each service order. When any candidate customer service representative transitions to an idle state, the dispatching module, according to the processing priority from high to low, sequentially executes the target customer service representative identification process for service orders of the same priority in order of generation time from earliest to latest. It should be noted that the re-identification of the waiting queue is triggered by a state transition event, without the need for periodically polling the status of each candidate customer service representative.

[0124] In the above approach, the occupancy time limit provides a definite exit mechanism for the duration of the pre-occupancy state. Occupancy that fails to be allocated within the time limit is automatically released, preventing customer service personnel from being in an ineffective suspended state for a long time. The occupancy time limit is configured according to the processing priority, so that high-priority work orders can obtain relatively sufficient allocation confirmation time. The queue to be allocated is redefined driven by state transition events. Even in extreme scenarios where customer service personnel are all busy, service work orders still have a definite temporary destination. Moreover, the processing order in the queue follows the sorting rules of processing priority and generation time, which is consistent with the sorting rules of conflict resolution. The concurrent processing logic of the system is completely closed.

[0125] As an optional embodiment, after each service order is processed, the feedback update module records the actual processing time and customer evaluation score of the service order, and calculates the service deviation of the service order according to the following formula. ; ;in The actual processing time for this service order, in seconds; Average processing time (seconds) for historical service orders of the same event type. The customer rating score for this service order, in points; The average customer rating score (points) for historical service orders; , Here, is the weighting coefficient, and is a dimensionless quantity. ,For example , All values ​​are 0.5;

[0126] The relative deviation in processing time characterizes the extent to which the processing time of this work order deviates from the average level of similar work orders; The relative deviation of customer evaluation represents the degree to which customer evaluation of this service deviates from the historical average level; the two are weighted and summed to obtain the service deviation amount, which comprehensively represents the degree to which this service deviates from the normal level from two dimensions: processing time and service quality.

[0127] The reason for separating statistics by event type is that the processing time for order placement, consultation and after-sales work orders varies greatly. Using work orders of the same type as the comparison benchmark can avoid the interference of type differences on the deviation.

[0128] Furthermore, the feedback update module periodically calculates the average service deviation of each service work order, determines the correction amount of the statistical window according to the product of the average and the preset coefficient, and adds the correction amount to the original length of the statistical window. That is, the correction amount is positively correlated with the average service deviation. The unit of the preset coefficient is consistent with the length unit of the statistical window, so that the correction amount is consistent with the original length of the statistical window.

[0129] The larger the mean of the service deviation, the greater the extension of the statistical window.

[0130] It should be noted that a large mean of service deviation indicates that the service processing level fluctuates greatly within the current statistical window. The reference value of the last consumption interval, consumption frequency, and consumption amount based on the statistics of this window for the stage division decreases. In this case, extending the statistical window can include more historical transaction data, so that the statistics on which the stage division is based tend to be stable. A small mean of service deviation indicates that the service processing level is stable. The statistical window is only slightly extended or kept at its original length to maintain the responsiveness of the stage division to changes in customer status.

[0131] After the length of the statistics window is adjusted, the statistics and stage division index of each customer will change accordingly, and the life cycle stage identifier of some customers will be migrated. When a large number of customers migrate into the dormant period, the system will generate a large number of wake-up work orders. If customer service resources are allocated after the adjustment takes effect and the wake-up work orders are actually generated, the dispatch side will not be able to respond in time, and the wake-up work orders will be backed up in the queue to be allocated.

[0132] Therefore, in this embodiment, before the correction amount of the statistical window takes effect, the feedback update module calculates the stage segmentation index of each customer according to the adjusted statistical window length, counts the number of customers whose stage segmentation index exceeds each segmentation boundary, obtains the scale of inbound and outbound customers for each life cycle stage, and determines the incremental estimated value of the wake-up work order according to the difference between the scale of inbound and outbound customers during the dormant period; the dispatch decision module expands the selection number of candidate customer service personnel in advance according to the incremental estimated value, so that customer service personnel who are idle and have the second highest comprehensive matching score are pre-entered into the candidate range; after the rehearsal is completed, the correction amount takes effect, and the life cycle segmentation module re-counts and migrates the life cycle stage identifier of each customer based on the adjusted statistical window.

[0133] Therefore, the impact of the statistical window adjustment on customer distribution is quantified and extrapolated before it takes effect. The dispatching side expands the candidate range in advance according to the extrapolation results. After the adjustment takes effect, the batch of wake-up work orders can directly obtain sufficient candidate customer service personnel, avoiding the impact of work order flood on the waiting queue. Moreover, the pre-calculation is carried out in the form of trial calculation, without the need for actual migration, and the processing overhead is small.

[0134] Reactivation work orders for dormant customers are services proactively initiated by the system, and their processing requires the service resources of customer service personnel. If a dormant customer fails to respond after multiple reactivations, continuing to allocate the same amount of reactivation service resources to that customer will consume service resources for customers with low responsiveness. Meanwhile, growing customers are in a phase of rapid transaction growth, and their service demands increase accordingly, making service resources relatively scarce.

[0135] Therefore, this embodiment redistributes service resource quotas among customers based on the processing results of the wake-up work order, so that wake-up service resources are tilted towards stages with higher response tendencies.

[0136] After the wake-up work order is processed, the feedback update module checks whether the initiating customer has generated transaction data in the statistics window: wake-up work orders that have generated transaction data are recorded as successful wake-up, and wake-up work orders that have not generated transaction data are recorded as not triggering a return flow; the number of wake-up work orders that have not triggered a return flow for the same dormant customer is counted to obtain the wake-up failure count for that customer; for example, if a dormant customer has three wake-up work orders in the statistics window, and the customer has not generated any transaction data in the statistics window after processing, then the wake-up failure count for that customer is three.

[0137] The service resource quota includes the quota for the number of work orders generated for customers within the unit's statistics window and the quota for the priority of work order assignment.

[0138] In one optional implementation, the service resource quota corresponding to the dormant customer is reduced according to the number of wake-up failures: the quantity quota of the customer is reduced, for example, for each additional wake-up failure, the quantity quota of the customer in the next statistical window is reduced by one; and the dispatch priority quota of the service work order corresponding to the customer is reduced, for example, the sorting position of its wake-up work order in the original processing priority is moved to the back.

[0139] The reduced service resource quota is transferred to the service work orders of growth-stage customers: The reduced quantity quota is added to the quantity quota of growth-stage customers, increasing the number of service work orders that can be generated for growth-stage customers within the unit statistics window; and the released dispatch priority quota is allocated to the service work orders of growth-stage customers, so that the service work orders of growth-stage customers are ranked higher within the same processing priority.

[0140] The allocation of service resource quotas transferred to customers in each growth stage can be based on the proportion of the stage index of each customer, so that customers with obvious transaction growth trends can obtain more service resources.

[0141] In summary, this invention connects the customer lifecycle stage division, service order generation and dispatch, and service result feedback and update into a closed loop. The customer's stage is updated with transaction data, the dispatch strategy is configured differently according to the stage, and the service result corrects the division benchmark. The three are mutually integrated and continuously self-correcting. The stage division index generates boundaries based on customer distribution clustering, the division benchmark is adaptive, and a buffer zone is set between adjacent customer groups to avoid the back-and-forth migration of stage identifiers and oscillation of matching weights caused by small fluctuations caused by window adjustments.

[0142] The system introduces a pre-occupancy state and concurrent conflict resolution during work order execution, ensuring a unique and determined order allocation for concurrent work orders. The occupancy time limit provides a definite exit mechanism for unfulfilled occupancy, preventing customer service personnel from suspending work orders ineffectively. The statistics window is adaptively adjusted according to the service deviation, avoiding baseline drift. Before the adjustment takes effect, the system performs a trial calculation and pre-expands the candidate range to prevent a surge of wake-up work orders from impacting the waiting queue, ensuring stable and reliable system operation.

[0143] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An e-commerce intelligent order dispatching and customer lifecycle management system, characterized in that, include: The customer data management module is used to collect transaction data, behavioral data, and after-sales data of customers on the e-commerce platform to build customer profiles. The lifecycle segmentation module is used to calculate the last consumption interval, consumption frequency and consumption amount of each customer based on the transaction data in the preset statistical window, and obtain the stage segmentation index of each customer by weighted summation. Based on the stage segmentation index, each customer is clustered into multiple customer groups, and each customer group is divided into the introduction stage, growth stage, maturity stage or dormant stage, and a lifecycle stage identifier of each customer is generated. The work order generation module is used to generate service work orders carrying the lifecycle stage identifier in response to order events, and to generate wake-up work orders for dormant customers; The dispatch decision module is used to determine the processing priority of the service order and the matching weight of each skill tag according to the life cycle stage identifier, and to determine the matching candidate customer service personnel. The status of the candidate customer service personnel includes idle state, pre-occupancy state and in-process state. The dispatching module determines the target customer service personnel based on the processing priority and the predicted load of each candidate customer service personnel, and sets them to a pre-occupancy state. When multiple service orders have the same target customer service personnel, the service order with the highest processing priority is assigned to that target customer service personnel and set to an in-process state. The remaining service orders re-determine the target customer service personnel from the remaining candidate customer service personnel who are in an idle state. The feedback update module is used to record the processing results and customer evaluations of the service work orders, calculate the service deviation, and adjust the statistical window according to the positive correlation.

2. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 1, characterized in that: The customer data management module constructs customer profiles in the following way: S201. The collected transaction data, behavior data and after-sales data are grouped according to customer identifiers. Data belonging to the same customer identifier is deduplicated and missing fields are filled in to obtain the customer dataset for each customer. S202. Extract transaction characteristics, behavioral characteristics and after-sales characteristics of each customer from the customer dataset. The transaction characteristics include the distribution of consumer product categories and the distribution of average order value. The behavioral characteristics include the distribution of browsing categories and the conversion rate of adding to cart. The after-sales characteristics include the distribution of after-sales types and the after-sales evaluation score. S203. According to the preset tagging system, the transaction characteristics, behavioral characteristics and after-sales characteristics of each customer are mapped to the corresponding customer tags, and the customer tags of each customer constitute a customer profile.

3. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 1, characterized in that: The lifecycle segmentation module divides customer groups into introduction, growth, maturity, or dormancy phases in the following way: Customers who generate transaction data for the first time within the statistical window are classified as the introductory period; For each customer other than those in the introductory period, the last consumption interval, consumption frequency, and consumption amount are normalized respectively. The normalized last consumption interval, consumption frequency, and consumption amount are then weighted and summed according to a preset weighting coefficient to obtain the stage division index for each customer. The weighting coefficients corresponding to consumption frequency and consumption amount are positive, while the weighting term corresponding to the last consumption interval participates in the weighted summation with a negative sign. Based on the distribution of the stage segmentation index of each customer, the customers are clustered into three customer groups; according to the mean of the stage segmentation index corresponding to each customer group from low to high, the customers in the three customer groups are successively divided into the dormant period, growth period and mature period. Generate corresponding lifecycle stage identifiers for customers segmented into different lifecycles; The lifecycle segmentation module sets up a buffer zone between adjacent customer groups, which is the index interval between the mean values ​​of the stage segmentation indices corresponding to two adjacent customer groups. After the statistical window is adjusted, the lifecycle segmentation module re-statistics based on the adjusted statistical window to obtain the updated stage segmentation index for each customer, and re-clusters the customers based on the updated stage segmentation index to obtain the updated customer groups; for customers whose updated stage segmentation index falls within the buffer zone, their original lifecycle stage identifier is maintained; for customers whose updated stage segmentation index is higher than the upper limit of the buffer zone or lower than the lower limit of the buffer zone, their lifecycle stage identifier is migrated to the lifecycle stage corresponding to the customer group to which they belong after re-clustering.

4. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 2, characterized in that: The work order generation module generates service work orders and wake-up work orders in the following way: Obtain the customer identifier corresponding to the customer who initiated the order event, read the lifecycle stage identifier corresponding to the customer identifier, and generate a service ticket carrying the customer identifier, the lifecycle stage identifier, and the event type of the order event; the order event includes order placement events, consultation events, and after-sales events; When a customer is classified as dormant, or when the customer's lifecycle stage identifier is migrated to the identifier corresponding to the dormant stage, wake-up content is generated based on the customer tag in the customer profile of that customer, and a wake-up work order carrying the customer identifier, the lifecycle stage identifier and the wake-up content is generated.

5. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 4, characterized in that: The dispatch decision module determines the processing priority of the service order in the following way: Service orders whose lifecycle stage is identified as the mature or growth stage are assigned to the first priority, service orders whose lifecycle stage is identified as the introduction stage are assigned to the second priority, and service orders whose lifecycle stage is identified as the dormant stage and wake-up stage are assigned to the third priority. The first priority is higher than the second priority, and the second priority is higher than the third priority. For multiple service orders with the same priority, they are processed and sorted according to the order in which they were generated.

6. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 1, characterized in that: The dispatch decision module determines the matching candidate customer service personnel in the following way: The skill tags include consultation and answering skill tags, after-sales service skill tags, and marketing communication skill tags; The matching weights of each skill tag are configured according to the lifecycle stage identifiers. Specifically, for service tickets corresponding to the introduction stage identifier, the matching weight of consultation and answering skill tags is greater than that of after-sales processing skill tags; for service tickets corresponding to the growth stage identifier and the maturity stage identifier, the matching weight of after-sales processing skill tags is greater than that of consultation and answering skill tags; for service tickets corresponding to the dormant stage identifier, the matching weight of marketing communication skill tags is greater than that of consultation and answering skill tags and after-sales processing skill tags. The skill matching degree between each skill tag and each customer service personnel is weighted and summed according to the matching weight of each skill tag to obtain the comprehensive matching score of each customer service personnel. Customer service personnel are selected as candidate customer service personnel in descending order of the comprehensive matching score.

7. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 1, characterized in that: The dispatching module determines the target customer service personnel based on the predicted workload of each candidate customer service personnel, specifically as follows: Obtain the pending service orders of the candidate customer service personnel. According to the event type of each pending service order, retrieve the historical average processing time of the corresponding event type. Determine the remaining processing time of each pending service order as the difference between the historical average processing time of each pending service order and the processing time of the pending service order. Determine the predicted load of the candidate customer service personnel as the sum of the remaining processing times of each pending service order. The candidate customer service representative with the lowest predicted load is identified as the target customer service representative and placed in a pre-occupancy state.

8. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 7, characterized in that: The dispatch execution module also includes: When the dispatching module sets the target customer service personnel to a pre-occupancy state, it generates a corresponding occupancy period. If the service order is assigned to the target customer service personnel within the occupancy period, the target customer service personnel's status is changed from the pre-occupancy state to the in-process state. If the occupancy period expires and the service order has not been assigned to the target customer service personnel, the target customer service personnel's status is released from the pre-occupancy state to the idle state. When re-determining the target customer service representative among the remaining candidate customer service representatives who are in an idle state for the remaining service work orders, if there are no remaining candidate customer service representatives in an idle state, the remaining service work orders are placed into the waiting queue. When a candidate customer service representative is found to be in an idle state, the process of determining the target customer service representative is re-executed for the service work orders in the waiting queue according to the processing priority.

9. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 7, characterized in that: The feedback update module calculates the service deviation and adjusts the statistical window by a positively correlated magnitude, specifically: Calculate the first relative deviation between the actual processing time of the service order and the historical average processing time of historical service orders of the same event type, and the second relative deviation between the customer evaluation score of the service order and the average customer evaluation score of historical service orders. Then, perform a weighted summation of the first relative deviation and the second relative deviation to obtain the service deviation amount of the service order. The correction amount of the statistical window is determined based on the average service deviation of each service work order. The correction amount is then added to the original length of the statistical window. The correction amount is positively correlated with the average service deviation.

10. The e-commerce intelligent order dispatching and customer lifecycle management system according to claim 8, characterized in that: The feedback update module also transfers service resource quotas based on the processing result of the wake-up work order, specifically: The number of wake-up work orders that did not trigger a return flow is counted among the wake-up work orders corresponding to the same dormant customer to obtain the wake-up failure rate of that customer. The "not triggering a return flow" means that the customer did not generate transaction data in the statistics window after the wake-up work order was processed. The service resource quota corresponding to the customer is reduced according to the number of wake-up failures, and the reduced service resource quota is transferred to the service work orders of the growth customers. The service resource quota includes the quota for the number of work orders generated for the customer in the unit statistics window and the quota for the work order dispatch priority.

Images