A store operation diagnosis system based on causal reasoning

Abstract

The application discloses a store operation diagnosis system based on causal reasoning, and relates to the technical field of computer data processing.The store operation diagnosis system generates candidate repurchase abnormal factors and candidate improvement actions through repurchase risk instances, and matches the repurchase risk instances, the candidate repurchase abnormal factors and a store repurchase causal graph by a causal path processing module to form a causal path set.According to edge attribute data, customer behavior results corresponding to formed task intervention records and the number of effective feedback samples, a causal contribution label is generated, so that the old customer repurchase risk identification result can establish a data link with executable employee action tasks, and unreachable tasks caused by directly assigning tasks according to only static labels, threshold rules or prediction probabilities are reduced.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and in particular to a store operation diagnostic system based on causal reasoning. Background Technology

[0002] With the widespread adoption of digital operation systems in chain stores, store operation diagnosis is gradually shifting from manual store visits and store manager experience-based judgment to automated analysis based on business data such as sales, customer traffic, membership, appointments, services, and employee tasks. In beauty retail stores, repeat purchases by existing customers are strongly correlated with store operation continuity. Existing systems typically identify customer churn risk or repeat purchase decline risk based on member segmentation, inactive days, repurchase cycle, RFM tags, contact records, or customer response probability, and generate follow-up calls, discount reminders, project matching, or employee to-do lists accordingly.

[0003] However, in customer retention and repurchase scenarios, a customer's failure to repurchase does not necessarily correspond to the same manageable reason. When a customer exceeds their expected repurchase cycle, there may be multiple reasons, such as the original service staff leaving their post, no available appointments during their preferred time slot, excessive recent contact, or the service project not yet being ready to be undertaken. Similarly, when a customer's inactivity reaches a threshold, they may not yet have entered their historical repurchase window. If employee tasks are generated solely based on static tags, threshold rules, or predicted probabilities, it's easy to convert customers who cannot be intervened with, those unsuitable for intervention, or those lacking the necessary conditions into employee tasks, leading to repeated contact, task backlog, and frequent switching of employee follow-up targets.

[0004] Existing systems typically mitigate these issues through manual review, contact frequency limits, blacklists, or task priority adjustments. However, these processes often occur after the task is generated, making it difficult to determine before task generation whether there is an executable path between abnormal repeat purchases, candidate improvement actions, employee execution conditions, and store acceptance conditions. After task execution, the system usually records the task status, number of contacts, or order results. However, customer repeat purchases may stem from various factors such as natural store visits, marketing outreach, employee follow-ups, and changes in appointment resources. If the execution data lacks a coherent link with subsequent customer behavior, store management cannot reconstruct the reasons for task triggering or invalidation.

[0005] Therefore, existing store operation diagnostic systems still need to address the technical issues of lacking causal judgment, executable condition verification, and execution feedback feedback in the context of maintaining repeat purchases from existing customers. Summary of the Invention

[0006] This application provides a store operation diagnosis system based on causal reasoning, which solves the problems of unachievable tasks, difficulty in tracing the triggering reasons, and difficulty in feedback after identifying the risk of repeat purchases by old customers.

[0007] This invention provides a store operation diagnosis system based on causal reasoning, including a data access module, a repeat purchase risk instance generation module, an anomaly factor processing module, a causal path processing module, an intervention accessibility determination module, an employee action task generation module, and a feedback update module. The data access module is used to acquire data on existing customer operations. The repeat purchase risk instance generation module is used to generate repeat purchase risk instances based on existing customer management data when a customer meets the repeat purchase risk triggering conditions. The abnormal factor processing module is used to generate candidate abnormal factors and candidate improvement actions based on repeat purchase risk instances; The causal path processing module is used to match repeat purchase risk instances and candidate repeat purchase abnormal factors with the store repeat purchase causal graph to obtain a set of causal paths from candidate repeat purchase abnormal factors to the target repeat purchase indicator, and generate causal contribution markers based on the edge attribute data in the causal path set, the customer behavior results corresponding to the task intervention records formed under the repeat purchase risk instances, and the number of effective feedback samples. The intervention accessibility determination module is used to generate the intervention accessibility status of candidate improvement actions based on customer repurchase status, service relationship status, employee execution status, store appointment capacity status, and reach history status. The employee action task generation module is used to generate employee action tasks that are bound to the causal path set and causal contribution marker when the intervention accessibility status is an accessible intervention status; The feedback update module is used to generate feedback samples based on the execution data of employee action tasks and the customer behavior results after the status of employee action tasks is formed. Based on the feedback samples, the edge attribute data of the store repurchase cause-effect graph and the task priority of candidate improvement actions are updated.

[0008] In some embodiments, the customer retention data includes customer visit records, service item records, membership benefit records, appointment records, contact records, employee service relationship records, employee shift records, and employee task execution records. The data access module includes an object alignment unit and a time sequence sorting unit. The object alignment unit is used to map customer objects, service items, employees, appointment orders, contact records, and task records to a unified object identifier. The time sequence sorting unit is used to sort the same customer object's arrival, service, contact, appointment, task assignment, task completion, and repeat purchase results according to the time of business occurrence, forming a customer repeat purchase process sequence.

[0009] In some embodiments, the repurchase risk instance generation module includes a repurchase status identification unit and an instance splitting unit; The repurchase status identification unit is used to determine the customer's repurchase status based on the customer's historical service items, last visit record, membership rights record, and service item configuration. The customer's repurchase status includes not reaching the repurchase window, reaching the repurchase window, exceeding the repurchase window, service relationship change, and no response after being reached. The instance splitting unit is used to generate multiple repurchase risk instances according to the correspondence between service items, service employees and member rights when there are multiple historical service items or multiple service relationships for the same customer object, and to write the instance source for each repurchase risk instance.

[0010] In some embodiments, the abnormal factor processing module includes a factor library unit, a factor assembly unit, and an action mapping unit; the factor library unit is used to store customer cycle factors, service relationship factors, employee actionable factors, contact response factors, and appointment acceptance factors; The factor assembly unit is used to extract candidate abnormal repurchase factors based on the customer repurchase status, service relationship status, reach history status and store appointment capacity status in the repurchase risk instance, and write the evidence data source for each candidate abnormal repurchase factor. The action mapping unit is used to generate candidate improvement actions according to the mapping relationship between candidate repurchase anomaly factors and action templates.

[0011] In some embodiments, the store repeat purchase causal graph includes a node table, an edge table, and an edge attribute table. The node table records customer lifecycle stage nodes, membership benefit nodes, service item nodes, service relationship nodes, employee scheduling nodes, contact method nodes, employee action task nodes, appointment behavior nodes, in-store behavior nodes, and repeat purchase result nodes. The edge table records the directed influence relationships between the aforementioned nodes. The edge attribute table records the edge identifier, direction, applicable stores, applicable service items, intervention action identifier, sample size, and intervention response status.

[0012] In some embodiments, the intervention accessibility determination module includes a customer accessibility determination unit; the customer accessibility determination unit is used to generate customer-side accessibility results based on customer repurchase cycle status, contact frequency status, customer response status and membership rights status, and the customer-side accessibility results include customer-side reachability results, customer-side delayed results, customer-side restricted results and customer-side prohibited results; The intervention accessibility status includes an accessible intervention status, a delayed intervention status, a reassigned intervention status, and a prohibited intervention status. The customer-side delayed result is used to correspond to the delayed intervention state, the customer-side prohibited result is used to correspond to the prohibited intervention state, the customer-side restricted result is used to restrict candidate improvement actions that rely on member rights, and the customer-side reachability result is used to output to the subsequent determination process of the intervention reachability determination module.

[0013] In some embodiments, the intervention accessibility determination module further includes an employee accessibility determination unit and a store acceptance accessibility determination unit; The employee accessibility determination unit is used to generate employee-side accessibility results based on the on-duty status, available appointment time slots, service skill tags, and historical service relationships of the responsible employees; the store accessibility determination unit is used to generate store-side accessibility results based on the store's appointment capacity status, service item time configuration, and the set of employees who can accept the appointment. The intervention accessibility status is synthesized based on customer-side accessibility results, employee-side accessibility results, and store-side accessibility results, and written into candidate improvement actions.

[0014] In some embodiments, the employee action task generation module includes an action selection unit, a task conflict verification unit, and a task object generation unit; The action selection unit is used to select candidate improvement actions corresponding to employee action tasks from candidate improvement actions based on causal contribution markers and intervention accessibility status. The task conflict verification unit is used to identify task conflict results based on the incomplete tasks, completed tasks, and prohibited intervention status of the same customer object. The task object generation unit is used to generate employee action tasks based on the selected candidate improvement actions when the task conflict result is no conflict, and write customer object, responsible employee, task type, execution window, reach method, candidate repurchase anomaly factor, causal path set and intervention reachability status.

[0015] In some embodiments, the feedback update module includes a feedback sample generation unit, a graph data update unit, and a task priority update unit; The feedback sample generation unit is used to record the task status, executing employee, contact method, customer response, appointment result, in-store result and repeat purchase result after an employee's action task is executed, canceled, reassigned or overdue, and to synthesize the aforementioned records with the causal path set into a feedback sample; The graph data update unit is used to update the edge attribute table according to the correspondence between task status and customer behavior results in the feedback samples; the task priority update unit is used to adjust the task priority of candidate improvement actions under similar repurchase risk instances according to the updated edge attribute table.

[0016] In some embodiments, a diagnostic link recording module is also included, which is used to write a link identifier for the same repurchase risk instance and write the link identifier into the candidate repurchase anomaly factor, the causal path set, the intervention accessibility status, the employee action task and the feedback sample respectively. The diagnostic link recording module is also used to output task triggering reason records and task invalidation reason records according to the link identifier. The task triggering reason records include repurchase risk triggering conditions, candidate repurchase abnormal factors, causal contribution markers and reachable intervention status. The task invalidation reason records include task status, customer response, appointment result, in-store result and repurchase result.

[0017] Through the above technical solution, the present invention can achieve at least the following beneficial effects: This invention generates candidate repurchase anomaly factors and candidate improvement actions by generating repurchase risk instances. The causal path processing module then matches the repurchase risk instances and candidate repurchase anomaly factors with the store's repurchase causal graph to form a set of causal paths. Based on edge attribute data, customer behavior results corresponding to the established task intervention records, and the number of valid feedback samples, a causal contribution marker is generated. This enables the identification results of repeat customer repurchase risks to establish a data link with executable employee action tasks, reducing unreachable tasks caused by directly assigning tasks based solely on static labels, threshold rules, or predicted probabilities.

[0018] Furthermore, the intervention accessibility determination module generates the intervention accessibility status of candidate improvement actions based on customer repurchase status, service relationship status, employee execution status, store appointment capacity status, and reach history status. This allows employee action tasks to be generated when the candidate improvement actions reach the accessible intervention status, thus ensuring that the task generation results correspond to the customer repurchase status, employee execution status, and store appointment capacity status.

[0019] Furthermore, employee action tasks are bound to a set of causal paths and causal contribution markers. The feedback update module generates feedback samples based on the execution data of employee action tasks and the customer behavior results after the formation time of the employee action task status, and updates the side attribute data and task priority, so that the task execution results and subsequent customer behaviors can flow back along the same data link.

[0020] Furthermore, the diagnostic link recording module associates repeat purchase risk instances, candidate repeat purchase anomaly factors, causal path sets, intervention accessibility status, employee action tasks, and feedback samples with link identifiers, enabling task triggering reason records and task invalidation reason records to be output according to the same link identifier. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0022] Figure 1 This is a framework diagram of the store operation diagnosis system based on causal reasoning in the embodiment. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

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

[0025] To facilitate understanding, the terminology used in the embodiments of this application will be introduced first: Customer retention data refers to the business records generated by a target store during customer visits, service items, membership benefits, appointments, outreach, employee service relationships, employee scheduling, and employee task execution.

[0026] Repeat purchase risk instances refer to diagnostic data units for repeat purchase anomalies, formed by processing customer objects, service items, and service relationships. Candidate repeat purchase anomaly factors are data factors that characterize the source of repeat purchase anomalies. Candidate improvement actions are employee-executable actions corresponding to the candidate repeat purchase anomaly factors. A store repeat purchase cause-and-effect graph is a graph data structure that records the various business nodes and the directed influence relationships between nodes in the process of repeat purchases by existing customers.

[0027] Intervention accessibility status refers to the feasibility assessment result of candidate improvement actions based on customer repurchase status, service relationship status, employee execution status, store appointment capacity status, and reach history. Feedback sample refers to data records synthesized from employee action task execution data and customer behavior results after the time the employee action task status was established. Employee execution status refers to the employee-side execution condition status formed by employee on-duty status, available appointment time slots, service skill tags, and historical service relationships. Store appointment capacity status refers to the store-side acceptance condition status formed by store appointment capacity, service item time configuration, and the set of available employees.

[0028] Customer behavior outcomes refer to records of customer responses, appointments, visits, and repeat purchases after the status formation time of an employee's action task. Task intervention records refer to task status records formed during the generation, execution, cancellation, reassignment, and overdue processes of employee action tasks. Status formation time refers to the business time corresponding to when an employee's action task enters the status of execution completion, cancellation confirmation, reassignment confirmation, or overdue deadline. Path response support value refers to the path-side response quantification result obtained by transforming the edge attribute data in the causal path set. Feedback behavior value refers to the behavior-side quantification result obtained by transforming the task status, customer response, appointment result, visit result, and repeat purchase result in the feedback sample. Valid feedback sample count refers to the number of feedback samples that match the candidate repeat purchase anomaly factor, causal path set, and contribution calculation version identifier within the same statistical window, and whose sample validity coefficient is greater than 0.

[0029] The statistical window refers to the range of store operating time used to summarize feedback samples. The contribution calculation version identifier refers to the data caliber identifier bound to the state dictionary, weight configuration, threshold configuration, edge attribute table snapshot, and statistical window. The sample validity coefficient is a coefficient used to indicate whether a feedback sample participates in the causal contribution value calculation. The feedback sample version identifier refers to the version record corresponding to the state dictionary, weight configuration, and threshold configuration written when the feedback sample is generated.

[0030] Diagnostic link records refer to data records used to record the relationships between repeat purchase risk instances, candidate repeat purchase anomaly factors, causal path sets, intervention accessibility status, employee action tasks, and feedback samples.

[0031] Example 1: like Figure 1 As shown, this embodiment employs a store operation diagnosis system based on causal reasoning, including a data access module, a repeat purchase risk instance generation module, an anomaly factor processing module, a causal path processing module, an intervention accessibility determination module, an employee action task generation module, and a feedback update module. The data access module is used to acquire customer management data for the target store. The repeat purchase risk instance generation module is used to generate repeat purchase risk instances based on existing customer management data when a customer meets the repeat purchase risk triggering conditions. The abnormal factor processing module is used to generate candidate abnormal factors and candidate improvement actions based on repeat purchase risk instances; The causal path processing module is used to match repeat purchase risk instances and candidate repeat purchase abnormal factors with the store repeat purchase causal graph to obtain a set of causal paths from candidate repeat purchase abnormal factors to target repeat purchase indicators. Based on the edge attribute data in the causal path set, the customer behavior results corresponding to the established task intervention records matched with repeat purchase risk instances, and the number of effective feedback samples, a causal contribution label is generated to distinguish between intervened abnormal factors, observed abnormal factors, and non-priority intervention abnormal factors. The intervention accessibility determination module is used to generate the intervention accessibility status of candidate improvement actions based on customer repurchase status, service relationship status, employee execution status, store appointment capacity status, and reach history status. The employee action task generation module is used to generate employee action tasks that are bound to the causal path set and causal contribution marker when the intervention accessibility state is an accessible intervention state that indicates that the candidate improvement action has the conditions for execution. The feedback update module is used to update the edge attribute data of the store repeat purchase causal graph and the task priority of candidate improvement actions based on the execution data of employee action tasks and the feedback samples formed by customer behavior results after the status of employee action tasks.

[0032] Edge attribute data refers to data items read from the edge attribute table based on the edge identifiers in the causal path set, including at least the sample size, intervention response status, applicable stores, applicable service items, and intervention action identifier. Established task intervention records matching repeat purchase risk instances refer to task intervention records where the link identifier, customer object, service item, and candidate improvement action all correspond to the current repeat purchase risk instance, and the task status has progressed to completion, cancellation confirmation, reassignment confirmation, or overdue deadline. When matching customer behavior results with established task intervention records, the link identifier and customer object are used as indexes, and the observation window after the status formation time is used as a time constraint.

[0033] In one implementation, the state dictionary refers to a configuration record that maps task status, customer response, appointment result, in-store result, repeat purchase result, and intervention response status to transformation values. The weight configuration refers to a configuration record that records the proportion of task status, customer response, appointment result, in-store result, and repeat purchase result in the feedback behavior value. The threshold configuration refers to a configuration record that records the positive contribution threshold, low contribution threshold, and minimum number of valid feedback samples. The edge attribute table snapshot refers to the set of edge attribute data read and frozen for this calculation under the same contribution calculation version identifier. The state dictionary, weight configuration, threshold configuration, and edge attribute table snapshot all have associated version identifiers and, together with the contribution calculation version identifier, are used to determine the data caliber for the same causal contribution label generation process.

[0034] In one implementation, the customer retention data includes customer visit records, service item records, membership benefit records, appointment records, contact records, employee service relationship records, employee scheduling records, and employee task execution records. The data access module includes an object alignment unit and a time-series processing unit. The object alignment unit maps customer objects, service items, employees, appointment orders, contact records, and task records to a unified object identifier. The time-series processing unit sorts the visit, service, contact, appointment, task assignment, task completion, and repeat purchase results of the same customer object according to the time of business occurrence, forming a customer repeat purchase process sequence.

[0035] A unified object identifier refers to an identifier field that establishes associations between customer objects, service items, employees, appointment orders, contact records, and task records. The customer repeat purchase process sequence must include at least the customer identifier, store identifier, service item identifier, employee identifier, transaction occurrence time, transaction record type, and transaction status. During data access, records lacking a transaction occurrence time are written to the chronological order position according to the record generation time; for records lacking a unified object identifier, a pending confirmation association status is generated based on the customer's mobile phone number, member account, appointment order number, or task number, and this pending confirmation association status is written to the customer repeat purchase process sequence.

[0036] In one implementation, the repeat purchase risk instance generation module includes a repeat purchase status identification unit and an instance splitting unit. The repeat purchase status identification unit is used to determine the customer's repeat purchase status based on the customer's historical service items, last visit record, membership benefit record, and service item configuration. The customer's repeat purchase status includes not having reached the repeat purchase window, having reached the repeat purchase window, exceeding the repeat purchase window, service relationship change, and no response after being reached. The instance splitting unit is used to generate multiple repeat purchase risk instances according to the correspondence between service items, service staff, and membership benefits when the same customer has multiple historical service items or multiple service relationships, and write the instance source for each repeat purchase risk instance.

[0037] Repeat purchase risk triggering conditions include at least one of the following: the repeat purchase window has expired, the repeat purchase window has been exceeded, the service relationship has changed, there has been no response after being contacted, and the membership benefits status has changed. A repeat purchase risk instance must include at least the following: customer identifier, store identifier, service item identifier, service employee identifier, membership benefits identifier, customer repeat purchase status, service relationship status, risk trigger source, and instance source. The instance source is used to record which set of correspondences between the service item, service employee, and membership benefits generated the repeat purchase risk instance.

[0038] In one implementation, the abnormal factor processing module includes a factor library unit, a factor assembly unit, and an action mapping unit. The factor library unit stores customer cycle factors, service relationship factors, employee actionable factors, outreach response factors, and appointment acceptance factors. The factor assembly unit extracts candidate abnormal repurchase factors based on the customer repurchase status, service relationship status, outreach history status, and store appointment capacity status in the repurchase risk instance, and writes the evidence data source for each candidate abnormal repurchase factor. The action mapping unit generates candidate improvement actions according to the mapping relationship between candidate abnormal repurchase factors and action templates.

[0039] Evidence data sources refer to the sources of business records that trigger candidate repurchase anomaly factors, including at least in-store visit records, appointment records, outreach records, employee scheduling records, employee service relationship records, and membership benefit records. Action templates include an action template identifier, applicable candidate repurchase anomaly factors, task type, default responsible role, outreach method, execution window type, and action disabling conditions. Candidate improvement actions are generated based on candidate repurchase anomaly factors, action templates, and evidence data sources, and are written into the candidate improvement action identifier, candidate repurchase anomaly factor, candidate responsible employee, and candidate execution window.

[0040] In one implementation, the store repeat purchase causal graph includes a node table, an edge table, and an edge attribute table. The node table records customer lifecycle stage nodes, membership benefit nodes, service item nodes, service relationship nodes, employee scheduling nodes, contact method nodes, employee action task nodes, appointment behavior nodes, in-store behavior nodes, and repeat purchase result nodes. The edge table records the directed influence relationships between the aforementioned nodes. The edge attribute table records the edge identifier, direction, applicable stores, applicable service items, intervention action identifier, sample size, and intervention response status.

[0041] The nodes in the node table correspond to customer repurchase status, service items, service relationships, outreach behaviors, appointment behaviors, employee action tasks, and repurchase results in the customer retention data. The directed influence relationships in the edge table are generated based on historical business records, store operation rules, and task intervention records. The edge attribute table also records the edge source type, edge update time, applicable customer repurchase status, and result recording scope. Target repurchase metrics refer to business indicators used to represent customer repurchase results, including at least one of the following: appointment generation result, in-store result, and service item repurchase result.

[0042] An edge identifier is a data identifier used to uniquely represent a directed influence relationship in a causal graph of store repeat purchases. The direction records the start and end nodes of the directed influence relationship. Applicable stores and applicable services limit the range of stores and services that the edge attribute data can match. The intervention action identifier records the candidate improvement actions associated with this directed influence relationship. The sample size records the number of feedback samples generated from the intervention response status of this directed influence relationship.

[0043] In one implementation, the feedback update module includes a feedback sample generation unit, a graph data update unit, and a task priority update unit. The feedback sample generation unit is used to record the task status, executing employee, contact method, customer response, appointment result, in-store result, and repeat purchase result after an employee's action task is executed, canceled, reassigned, or overdue, and to synthesize the aforementioned records with the causal path set into a feedback sample. The graph data update unit is used to update the edge attribute table according to the correspondence between the task status and customer behavior result in the feedback sample. The task priority update unit is used to adjust the task priority of candidate improvement actions under similar repeat purchase risk instances according to the updated edge attribute table.

[0044] Similar repeat purchase risk instances refer to a set of repeat purchase risk instances that match in terms of store identifier, service type, customer repeat purchase status, and candidate repeat purchase anomaly factors. When updating the side attribute table, the task status, executing employee, contact method, and execution window of the employee action task are used as intervention variables, and customer response, appointment result, in-store result, and repeat purchase result are used as outcome variables. When updating the task priority, the side attribute data and intervention accessibility status corresponding to similar repeat purchase risk instances are read, and the updated task priority is written into the candidate improvement actions.

[0045] Task priority refers to the data status of candidate improvement actions in the task selection order under similar repurchase risk instances. Task priority is jointly determined by edge attribute data, causal contribution markers, and intervention accessibility status, and is written along with the candidate improvement actions. After the edge attribute table is updated, candidate improvement actions under similar repurchase risk instances enter the action selection process according to the updated task priority; candidate improvement actions with causal contribution markers indicating interveneable abnormal factors and intervention accessibility status indicating an reachable intervention state enter the selection range for generating employee action tasks.

[0046] In one implementation, a diagnostic link recording module is also included. This module is used to write a link identifier for the same repeat purchase risk instance and write the link identifier into the candidate repeat purchase anomaly factor, the causal path set, the intervention accessibility status, the employee action task, and the feedback sample, respectively. The diagnostic link recording module is also used to output a task triggering reason record and a task invalidation reason record according to the link identifier. The task triggering reason record includes the repeat purchase risk triggering condition, the candidate repeat purchase anomaly factor, the causal contribution marker, and the accessibility intervention status. The task invalidation reason record includes the task status, customer response, appointment result, in-store result, and repeat purchase result.

[0047] Link identifiers refer to data association identifiers that connect repeat purchase risk instances, candidate repeat purchase anomaly factors, causal path sets, intervention attainability status, employee action tasks, and feedback samples. Task trigger reason records also include link identifiers, target repeat purchase metrics, evidence data sources, and candidate improvement action identifiers. Task invalidity reason records also include link identifiers, intervention attainability status, task conflict results, cancellation reasons, reassignment reasons, and overdue reasons. The diagnostic link record module outputs task trigger reason records and task invalidity reason records according to the link identifiers.

[0048] In a preferred embodiment of Example 1, when the causal path processing module generates a causal contribution marker for the candidate repurchase anomaly factor, it determines the applicable store, applicable service item, service employee, historical service relationship and membership rights status according to the instance source of the repurchase risk instance, and uses the aforementioned information as path filtering conditions to retain the directed path under the same business scope as the repurchase risk instance from the causal path set.

[0049] For each retained directed path, the causal path processing module reads the sample count, intervention response status, applicable service items, and intervention action identifier from the edge attribute table, and associates the employee action tasks and customer behavior results that have been formed on the same directed path.

[0050] Customer behavior outcomes are taken from the observation window following the time the employee's action task status is established. The observation window is determined according to the repurchase cycle, appointment lead time, and membership validity period in the service item configuration; when the same customer has multiple customer behavior outcomes within the observation window, a feedback sample sequence is formed according to the business occurrence time of customer response, appointment result, in-store result, and repurchase result.

[0051] When multiple employee action tasks exist for the same customer within the observation window, the customer behavior result is preferentially matched with the employee action task that matches the link identifier. If the link identifier is missing, matching is performed according to the chronological order of customer, service item, candidate improvement action, and status formation time, and the matching criteria are written into the feedback sample. Once a customer behavior result has been bound to a feedback sample, it will not be bound to other feedback samples under the same contribution calculation version identifier. When the same customer behavior result meets the matching conditions of multiple employee action tasks, it is preferentially bound to the employee action task with the closest status formation time and consistent service item.

[0052] For any directed path, the causal path processing module generates a path response support value based on the number of samples on each edge of the directed path and the intervention response status. The calculation method is as follows: , in, Represents a directed path Path response support value; This represents a directed path from the node corresponding to the candidate repurchase anomaly factor to the node corresponding to the target repurchase indicator and passing through the node corresponding to the candidate improvement action; Represents a directed path The set of edges contained therein; Represents the set of edges One of the edges in; Representing an edge The number of samples recorded in the edge attribute table; Representing an edge The intervention response state transition value; This represents the path support correction constant used to avoid a denominator of zero. The range of values ​​is , The larger the value, the greater the influence of the edge on the path response support value, and the logarithmic function limits the dominance of a single high-sample edge on the entire directed path.

[0053] Intervention response state transition value The positive response state is obtained by mapping the intervention response state from the edge attribute table. Greater than 0, no stable response state corresponds to Equal to 0, the negative response state corresponds to Less than 0. Edge When the intervention response status is empty, the snapshot of the side attribute table is missing, or the applicable store or applicable service item is inconsistent with the repeat purchase risk instance, Set to 0; edge Number of samples When empty, Set to 0.

[0054] Intervention response status includes positive response status, unstable response status, and negative response status. A positive response status refers to the percentage of customer behavior results achieving the target repurchase indicator after the employee's action task is completed, meeting the positive response criteria configured by the store, under similar repurchase risk instances. An unstable response status refers to the percentage of customer behavior results that do not form a positive response status or a negative response status under similar repurchase risk instances. A negative response status refers to the percentage of customer records that do not respond, fail to book, do not come to the store, or do not repurchase after the employee's action task is completed, meeting the negative response criteria configured by the store. Intervention response status is synthesized from feedback samples and written into the edge attribute table, and enters the causal contribution label generation process along with a snapshot of the edge attribute table. The positive and negative response criteria are the record percentage judgment conditions used in the store configuration to determine the intervention response status. The record percentage judgment conditions include the applicable store, applicable service item, statistical window, target repurchase indicator, record percentage threshold, and output intervention response status.

[0055] For any feedback sample associated with a candidate repurchase anomaly factor, the causal path processing module generates a feedback behavior value based on task status, customer response, appointment result, in-store result, and repurchase result. The calculation method is as follows: , in, Indicates feedback sample Feedback behavior value; This represents a feedback sample that is bound to the candidate repurchase anomaly factor and its causal path set; Indicates the weight of the task state transition value; Indicates feedback sample Task state transition values; Indicates the weight of the customer response conversion value; Indicates feedback sample Customer response conversion value in the data; Indicates the weight of the converted value of the reservation result; Indicates feedback sample The converted value of the appointment result in the middle; Indicates the weight of the conversion value of the in-store result; Indicates feedback sample The conversion value of the in-store result; Indicates the weight of the conversion value of repeat purchase results; Indicates feedback sample The conversion value of repeat purchase results in the middle.

[0056] The range of values ​​for each conversion is: The value range of each weight is [missing information]. Furthermore, the sum of all weights is 1; the conversion values ​​corresponding to task completion, customer response, successful appointment, in-store completion, and repeat purchase completion are higher than the conversion values ​​corresponding to task overdue, customer no response, appointment failure, no in-store visit, and no repeat purchase.

[0057] Each conversion value is mapped from the status dictionary configured by the store and remains consistent within the same store's operating cycle; the task status conversion values ​​corresponding to task cancellation, task reassignment, and task overdue are lower than the task status conversion values ​​corresponding to task completion. Each weight is obtained from the store's operating configuration or calibrated using historical feedback samples. When using historical feedback samples for calibration, the weight of the repurchase result conversion value is used as the weight on the business result side, and the weights of task status conversion value, customer response conversion value, appointment result conversion value, and in-store result conversion value remain fixed within the same statistical window.

[0058] The status dictionary must include at least the status type, status name, conversion value, applicable stores, applicable services, version identifier, and effective date. Under the same version identifier, the same status name within the same status type corresponds to a unique conversion value. The conversion values ​​in the status dictionary can be obtained based on historical feedback sample quantiles or validation sets, and are used as a fixed metric in calculations within the same store's operating cycle. When the status dictionary changes, existing feedback samples retain their original version identifier, while newly generated feedback samples are written with the changed version identifier.

[0059] In order to incorporate historical support information from the edge attribute table into the judgment along with current feedback samples, the causal path processing module adjusts the fusion ratio of path response support values ​​and feedback behavior values ​​based on the number of feedback samples. The calculation method is as follows: , in, Indicates candidate repurchase anomaly factors The corresponding path supports the fusion ratio; This represents the candidate repurchase anomaly factors currently to be labeled; Indicates candidate repurchase anomaly factors The corresponding number of feedback sample calibrations; Indicates candidate repurchase anomaly factors The number of valid feedback samples associated within the current statistical window. The current statistical window can be set to the past 30 days, past 60 days, or past 90 days according to the store's operating cycle; The larger, The smaller the value, the higher the proportion of the feedback behavior value in the subsequent causal contribution judgment; when When the number of feedback samples is insufficient, the system uses the path response support value in the edge attribute table for conservative judgment.

[0060] The causal contribution value of candidate repeat purchase anomalies is generated according to the following formula: , in, Indicates candidate repurchase anomaly factors causal contribution value; This indicates that the calculation result within the parentheses is restricted to... Within the range; The aforementioned definition will be used; Indicates candidate repurchase anomaly factors The corresponding set of causal paths; The aforementioned definition will be used; Represents the set of causal paths The number of directed paths in; This represents the constant used to correct the denominator of the path count; Indicates candidate repurchase anomaly factors The corresponding set of feedback samples; The aforementioned definition will be used; Indicates feedback sample The sample effective coefficient; The aforementioned definition will be used; This represents a constant used to correct the denominator of the feedback sample weights.

[0061] Sample efficiency coefficient Determined according to the following formula: , in, The aforementioned definition will be used; Indicates feedback sample The path consistency identifier value, feedback sample The value is 1 if the bound employee action task matches the directed path in the causal path set, and 0 otherwise. Indicates feedback sample The observation window consistency flag value is set to 1 if the customer behavior result occurs within the observation window, and 0 otherwise. Indicates feedback sample The employee consistency identifier value is 1 if the employee performing the task is a responsible employee or a transferred employee, and 0 otherwise. Indicates feedback sample The actual intervention indicator value is 1 when the employee's action task has been completed or has been completed by the reassigning employee after being reassigned, and 0 when the task is canceled, the task is overdue and not reached, or the task is not accepted after being reassigned. , , and The range of values ​​is .when At that time, feedback samples It is retained in the feedback sample set for link recording, but does not participate in the calculation of causal contribution value.

[0062] When the causal path set When not empty, the causal path processing module generates causal contribution labels based on the causal contribution value, the number of valid feedback samples, and the preset labeling threshold. The generation method is as follows: like and ,but , like or ,but , like and ,but ; in, Indicates candidate repurchase anomaly factors causal contribution markers; The aforementioned definition will be used; Indicates candidate repurchase anomaly factors The corresponding positive contribution threshold; The aforementioned definition will be used; Indicates candidate repurchase anomaly factors The corresponding minimum number of valid feedback samples; Indicates candidate repurchase anomaly factors The corresponding low contribution threshold. The positive contribution threshold and low contribution threshold are set according to the applicable service items, historical service relationships, and store operating cycles, and must meet the following requirements. .

[0063] , and All are correction constants greater than 0; The number of feedback samples greater than 0 is used for calibration; Candidate repurchase anomaly factors The number of feedback samples with a valid coefficient greater than 0 within the current statistical window, and the number of such samples is an integer greater than or equal to 0. It is an integer greater than 0; It is an integer greater than or equal to 0. When At that time, the causal path processing module will identify candidate repurchase anomaly factors. Record this as an anomaly to be observed, and pause the output of this candidate repurchase anomaly as an interveneable anomaly; when and At that time, the causal path processing module will identify candidate repurchase anomaly factors. This is denoted as the abnormal factor to be observed.

[0064] When generating the same candidate repurchase anomaly factor When marking causal contributions, the causal path processing module writes the contribution calculation version identifier. Contribution calculation version identifier It is only used to bind the state dictionary, weight configuration, threshold configuration, edge attribute table snapshot and current statistics window used in this calculation. Each conversion value, each weight, , , Both the intervention response status and the contribution calculation version identifier are taken from the same source. Corresponding data caliber; when the version identifier of any of the aforementioned data is... In case of inconsistency, the corresponding edge or feedback sample will not participate in the current causal contribution value calculation, and the reason for the version inconsistency will be written into the diagnostic link record. The contribution calculation version identifier remains fixed within the current statistical window. When the state dictionary, weight configuration, or threshold configuration changes, a new contribution calculation version identifier will be used from the next statistical window.

[0065] Used for calculation A valid feedback sample should simultaneously satisfy the sample validity coefficient. Greater than 0 and feedback sample version identifier and contribution calculation version identifier Consistent; feedback samples that do not meet the aforementioned conditions are retained in the feedback sample set for link recording, but are not counted. .

[0066] The causal contribution marker is used to classify candidate repeat purchase anomalies into interveneable anomalies, unobservable anomalies, or non-priority intervention anomalies based on their causal contribution value and the number of valid feedback samples.

[0067] If the task intervention record and customer behavior outcome cannot be matched by customer object and link identifier, or if the customer behavior outcome occurs outside the observation window, the causal path processing module sets the sample validity coefficient of the corresponding feedback sample to zero. The above marking results, together with the link identifiers of candidate repurchase anomaly factors, causal path sets, candidate improvement actions, and repurchase risk instances, are written into the diagnostic link record for subsequent action selection by the employee action task generation module.

[0068] Example 2: Based on Example 1, this example provides a specific method for generating customer-side accessibility results in a store operation diagnostic system based on causal reasoning; The intervention accessibility determination module includes a customer accessibility determination unit; the customer accessibility determination unit is used to generate customer-side accessibility results based on the customer repurchase cycle status, contact frequency status, customer response status and membership rights status. The customer-side accessibility results include customer-side reachability results, customer-side delayed results, customer-side restricted results and customer-side prohibited results. Intervention accessibility status includes accessible intervention status, delayed intervention status, reassigned intervention status, and prohibited intervention status; the customer-side delayed results are used for the corresponding delayed intervention status, the customer-side prohibited results are used for the corresponding prohibited intervention status, the customer-side restricted results are used to restrict candidate improvement actions that rely on member rights, and the customer-side accessible results are used to output to the subsequent judgment process of the intervention accessibility judgment module.

[0069] Customer response status refers to the state in which a customer responds to a contact record, appointment invitation, or employee action task within the contact statistics window. A valid appointment behavior refers to an appointment record generated by a customer within the contact statistics window that has not been cancelled, or a proactive appointment record formed by a customer under the same service item. When a valid appointment behavior exists in the customer response status, it is considered as a responded status in the customer-side accessibility result determination. When contact history is missing, the contact frequency status is treated as not reaching the store's configured threshold, and the missing contact history status is written into the diagnostic link record; when customer response records are missing and there is no valid appointment behavior, the customer response status is treated as a non-response status.

[0070] In one implementation, the intervention accessibility determination module further includes an employee accessibility determination unit and a store acceptance accessibility determination unit; the employee accessibility determination unit is used to generate employee-side accessibility results based on the on-duty status, available appointment time slots, service skill tags, and historical service relationships of the responsible employee; the store acceptance accessibility determination unit is used to generate store-side accessibility results based on the store's appointment capacity status, service item time configuration, and available employee set; the intervention accessibility status is synthesized based on the customer-side accessibility results, employee-side accessibility results, and store-side accessibility results, and written into candidate improvement actions.

[0071] Responsible employees refer to the employees who will perform the candidate improvement actions when the employee action tasks are generated. Candidate responsible employees are the set of assignable employees obtained based on historical service relationships, service skill tags, and employee scheduling records. The set of employees eligible to take on additional tasks refers to the set of employees within the target store who simultaneously meet the requirements of being on duty, available appointment slots, and service skill tags.

[0072] When the responsible employee has a historical service relationship and has an available appointment time slot, the employee-side accessibility results are used to synthesize the accessibility intervention status; when the responsible employee is off duty or has no appointment time slot but the set of employees who can take over has an employee, the employee-side accessibility results are used to synthesize the transfer intervention status; when the responsible employee is off duty or has no appointment time slot and the set of employees who can take over is empty, the employee-side accessibility results are used to synthesize the delayed intervention status.

[0073] In one implementation, the employee action task generation module includes an action selection unit, a task conflict verification unit, and a task object generation unit. The action selection unit is used to select candidate improvement actions from candidate improvement actions to generate employee action tasks based on causal contribution markers and intervention accessibility status. The task conflict verification unit is used to identify task conflict results based on the incomplete tasks, completed tasks, and prohibited intervention status of the same customer object. The task object generation unit is used to generate employee action tasks based on the selected candidate improvement actions when the task conflict result is non-conflicting, and writes the target customer, responsible employee, task type, execution window, reach method, candidate repeat purchase anomaly factor, causal path set, and intervention accessibility status.

[0074] Task conflict results include no conflict, conflict with incomplete tasks, conflict with recently completed tasks, and conflict with prohibited intervention. Employee action tasks also include task identifier, repeat purchase risk instance identifier, candidate improvement action identifier, task generation time, task deadline, and task status. During task object generation, the responsible employee is selected from the candidate responsible employees; if no available time slot is available for the candidate responsible employee, the employee action task is written to the responsible employee corresponding to the transfer intervention status.

[0075] An incomplete task conflict refers to an unfinished employee action task existing for the same customer under the same service item and the same candidate repurchase anomaly factor. A recently completed task conflict refers to a completed employee action task of the same type existing within the task cooldown window configured at the store. A prohibited intervention conflict refers to a prohibited intervention status existing for the same customer within the same outreach statistics window. When the task conflict result is no conflict, the task object generation unit generates an employee action task; when the task conflict result is an incomplete task conflict, a recently completed task conflict, or a prohibited intervention conflict, the task object generation unit writes the conflict reason into the diagnostic link record.

[0076] In a preferred embodiment of Example 2, when the customer accessibility determination unit generates customer-side accessibility results, it takes repurchase risk instances and candidate improvement actions as determination objects, reads the last visit time, service item repurchase cycle configuration, contact history, customer response record and membership rights record in the repurchase risk instance, and converts the customer repurchase cycle status, contact frequency status, customer response status and membership rights status into customer-side accessibility results under the same determination caliber.

[0077] The customer reachability determination unit calculates the repurchase window position value based on the start and end intervals of the repurchase window corresponding to the service item: , in, Indicates the object to be judged The repurchase window position value; This represents a customer-side assessment object consisting of a repeat purchase risk instance and a candidate improvement action; Indicates the object to be judged The corresponding current determination time; Indicates the object to be judged The last time the corresponding customer visited the store; Indicates the object to be judged The starting interval for repurchase windows for the corresponding service items; Indicates the object to be judged The deadline for repurchase of the corresponding service items; Indicates the object to be judged The corresponding repurchase window denominator correction constant. The repurchase window start interval and repurchase window end interval are derived from the service item configuration and satisfy the following conditions: ;when At that time, the customer's repurchase cycle status is "not yet at the repurchase window"; when At that time, the customer's repurchase cycle status is "repurchase window has arrived"; when At that time, the customer's repurchase cycle status is "exceeded the repurchase window".

[0078] The customer reachability assessment unit calculates the reach pressure value based on the reach statistics window and reach frequency threshold configured in the store: , in, Indicates the object to be judged The pressure value that is reached; Indicates the object to be judged The number of times the corresponding customer has been reached in the outreach statistics window; Indicates the object to be judged The threshold for the number of touches configured for each store within the touch statistics window; Indicates the object to be judged The corresponding contact count denominator correction constant. The contact statistics window is configured according to the store's operating cycle as the last 7 days, last 15 days, or last 30 days; when When the frequency of contact reaches the store configuration threshold; when At that time, the frequency of contact was not reached at the store configuration threshold.

[0079] The customer accessibility assessment unit calculates the available value of membership benefits based on the validity period of membership benefits and the dependency relationship between candidate improvement actions and membership benefits: , in, Indicates the object to be judged The available value of the equity; Indicates the object to be judged The corresponding candidate improvement action is identified by whether it depends on membership benefits. The value is 1 if it depends on membership benefits and 0 if it does not. Indicates the object to be judged The value indicates whether the corresponding customer has available membership benefits. It is 1 if there are available membership benefits and 0 if there are no available membership benefits. This indicates that the calculation result within the parentheses is restricted to... Within the range; Indicates the object to be judged The expiration date of the corresponding membership benefits; The aforementioned definition will be used; Indicates the object to be judged The lead time for the corresponding candidate improvement actions; Indicates the object to be judged The corresponding correction constant for the denominator of the equity lead time.

[0080] When the same customer has multiple available membership benefits The earliest expiration date of the membership benefit that matches the candidate improvement action is selected. When the candidate improvement action does not depend on the membership benefit, the benefit availability value is 1; when the candidate improvement action depends on the membership benefit and there is no available membership benefit for the customer, the benefit availability value is 0; when the candidate improvement action depends on the membership benefit and the membership benefit is within its validity period, the closer the remaining time of the membership benefit is to or beyond the benefit's early expiration date, the higher the benefit availability value.

[0081] The customer accessibility determination unit generates customer-side accessibility results based on the priorities of reach prohibition, repurchase postponement, benefit restriction, and customer accessibility: like ,but , like ,but , like ,but , like ,but ; in, Indicates the object to be judged Customer-side accessibility results; The aforementioned definition will be used; The aforementioned definition will be used; Indicates the object to be judged The corresponding identifier value indicates whether the customer is in a non-responsive state within the outreach statistics window. It is set to 1 when the customer is in a non-responsive state and to 0 when the customer is in a responsive state or has a valid appointment. Using the aforementioned definitions: When the reach frequency reaches the store's configured threshold and the customer's response status is non-response, the candidate improvement action enters the customer-side prohibited result; if the customer's repurchase cycle status is not yet at the repurchase window if the customer's prohibited result is not triggered, the candidate improvement action enters the customer-side delayed result, and is re-evaluated after the repurchase window start time arrives; if the customer's prohibited result is not triggered and the customer's repurchase cycle status is already at or beyond the repurchase window, and the membership benefits status does not meet the requirements of the candidate improvement action, the candidate improvement action enters the customer-side restricted result; if the reach frequency does not reach the store's configured threshold, or if the customer has already responded or made a valid appointment, and the membership benefits status meets the requirements of the candidate improvement action, the candidate improvement action enters the customer-side reachable result.

[0082] Delayed results on the customer side correspond to delayed intervention status; prohibited results on the customer side correspond to prohibited intervention status; restricted results on the customer side do not directly generate employee action tasks, and the action selection unit prioritizes candidate improvement actions that do not rely on membership benefits. If there are no candidate improvement actions that do not rely on membership benefits under the same candidate repurchase abnormal factor, then it corresponds to delayed intervention status; reachable results on the customer side enter the combined judgment process of reachable results on the employee side and reachable results on the store side.

[0083] The threshold for the number of touches, the touch statistics window, the start interval of the repeat purchase window, the end interval of the repeat purchase window, and the benefit advance period are all derived from the store configuration or service item configuration. When a customer's last visit time is missing, service repurchase cycle configuration is missing, contact history is missing, customer response status cannot be identified, or candidate improvement actions rely on membership benefits and the validity period of membership benefits cannot be identified, the customer accessibility determination unit adopts a conservative strategy, recording the customer-side accessibility result as a customer-side restriction result and writing the missing field into the diagnostic link record; however, if the customer already has a valid appointment record or proactive visit record in the contact statistics window, the customer response status is treated as a responded status. The customer-side accessibility result is recalculated each time a repurchase risk instance is generated, a contact record is added, a customer response record is added, an appointment record is added, a visit record is added, or the membership benefit status changes; if the same customer already has a prohibited intervention status in the same contact statistics window, subsequent candidate improvement actions will continue to use the prohibited intervention status until the re-determination conditions are triggered.

[0084] , and All are denominator correction constants that are greater than 0; It is an integer greater than or equal to 0; It is an integer greater than 0; A time interval greater than 0; Greater than The same judgment object corresponding , and Calculations are performed using the same unit of time.

[0085] When the customer reachability determination unit generates customer-side reachability results, it writes the customer-side determination configuration version identifier. The client side determines the configuration version identifier. Used only for binding the judgment object The store configuration, service item configuration, reach statistics window, reach frequency threshold, and membership benefits configuration used in this calculation are as follows: When customer-side reachability results are combined with employee-side and store-side reachability results to generate the intervention reachability status, the customer-side judgment configuration version identifier will continue to be used. The corresponding configuration criteria; when the configuration version changes before the task is generated, the unassigned employee action tasks will re-execute the client-side reachability determination, while the already assigned employee action tasks will retain the original client-side determination configuration version identifier and write it into the diagnostic link record.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0087] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A store operation diagnostic system based on causal reasoning, characterized in that, It includes a data access module, a repeat purchase risk instance generation module, an anomaly factor processing module, a causal path processing module, an intervention accessibility determination module, an employee action task generation module, and a feedback update module; The data access module is used to acquire data on existing customer operations. The repeat purchase risk instance generation module is used to generate repeat purchase risk instances based on existing customer management data when a customer meets the repeat purchase risk triggering conditions. The abnormal factor processing module is used to generate candidate abnormal factors and candidate improvement actions based on repeat purchase risk instances; The causal path processing module is used to match repeat purchase risk instances and candidate repeat purchase abnormal factors with the store repeat purchase causal graph to obtain a set of causal paths from candidate repeat purchase abnormal factors to the target repeat purchase indicator, and generate causal contribution markers based on the edge attribute data in the causal path set, the customer behavior results corresponding to the task intervention records formed under the repeat purchase risk instances, and the number of effective feedback samples. The intervention accessibility determination module is used to generate the intervention accessibility status of candidate improvement actions based on customer repurchase status, service relationship status, employee execution status, store appointment capacity status, and reach history status. The employee action task generation module is used to generate employee action tasks that are bound to the causal path set and causal contribution marker when the intervention accessibility status is an accessible intervention status; The feedback update module is used to generate feedback samples based on the execution data of employee action tasks and the customer behavior results after the status of employee action tasks is formed. Based on the feedback samples, the edge attribute data of the store repurchase cause-effect graph and the task priority of candidate improvement actions are updated.

2. The store operation diagnosis system based on causal reasoning according to claim 1, characterized in that, The customer retention data includes customer visit records, service records, membership benefit records, appointment records, contact records, employee service relationship records, employee shift records, and employee task execution records. The data access module includes an object alignment unit and a time sequence sorting unit. The object alignment unit is used to map customer objects, service items, employees, appointment orders, contact records, and task records to a unified object identifier. The time sequence sorting unit is used to sort the same customer object's arrival, service, contact, appointment, task assignment, task completion, and repeat purchase results according to the time of business occurrence, forming a customer repeat purchase process sequence.

3. The store operation diagnosis system based on causal reasoning according to claim 2, characterized in that, The repurchase risk instance generation module includes a repurchase status identification unit and an instance splitting unit; The repurchase status identification unit is used to determine the customer's repurchase status based on the customer's historical service items, last visit record, membership rights record, and service item configuration. The customer's repurchase status includes not reaching the repurchase window, reaching the repurchase window, exceeding the repurchase window, service relationship change, and no response after being reached. The instance splitting unit is used to generate multiple repurchase risk instances according to the correspondence between service items, service employees and member rights when there are multiple historical service items or multiple service relationships for the same customer object, and to write the instance source for each repurchase risk instance.

4. The store operation diagnosis system based on causal reasoning according to claim 1, characterized in that, The abnormal factor processing module includes a factor library unit, a factor assembly unit, and an action mapping unit; the factor library unit is used to store customer cycle factors, service relationship factors, employee actionable factors, contact response factors, and appointment acceptance factors; The factor assembly unit is used to extract candidate abnormal repurchase factors based on the customer repurchase status, service relationship status, reach history status and store appointment capacity status in the repurchase risk instance, and write the evidence data source for each candidate abnormal repurchase factor. The action mapping unit is used to generate candidate improvement actions according to the mapping relationship between candidate repurchase anomaly factors and action templates.

5. The store operation diagnosis system based on causal reasoning according to claim 1, characterized in that, The store repeat purchase causal graph includes a node table, an edge table, and an edge attribute table. The node table records customer lifecycle stage nodes, membership benefit nodes, service item nodes, service relationship nodes, employee scheduling nodes, contact method nodes, employee action task nodes, appointment behavior nodes, in-store behavior nodes, and repeat purchase result nodes. The edge table records the directed influence relationships between the aforementioned nodes. The edge attribute table records the edge identifier, direction, applicable stores, applicable service items, intervention action identifier, sample size, and intervention response status.

6. The store operation diagnosis system based on causal reasoning according to claim 1, characterized in that, The intervention accessibility determination module includes a customer accessibility determination unit; The customer reachability determination unit is used to generate customer-side reachability results based on customer repurchase cycle status, contact frequency status, customer response status, and membership rights status. The customer-side reachability results include customer-side reachability results, customer-side delayed results, customer-side restricted results, and customer-side prohibited results. The intervention accessibility status includes an accessible intervention status, a delayed intervention status, a reassigned intervention status, and a prohibited intervention status. The customer-side delayed result is used to correspond to the delayed intervention state, the customer-side prohibited result is used to correspond to the prohibited intervention state, the customer-side restricted result is used to restrict candidate improvement actions that rely on member rights, and the customer-side reachability result is used to output to the subsequent determination process of the intervention reachability determination module.

7. The store operation diagnosis system based on causal reasoning according to claim 6, characterized in that, The intervention accessibility determination module also includes an employee accessibility determination unit and a store accessibility determination unit; The employee accessibility determination unit is used to generate employee-side accessibility results based on the on-duty status, available appointment time slots, service skill tags, and historical service relationships of the responsible employee. The store accessibility determination unit is used to generate store-side accessibility results based on the store's reservation capacity status, service item time configuration, and the set of available employees. The intervention accessibility status is synthesized based on customer-side accessibility results, employee-side accessibility results, and store-side accessibility results, and written into candidate improvement actions.

8. The store operation diagnosis system based on causal reasoning according to claim 7, characterized in that, The employee action task generation module includes an action selection unit, a task conflict verification unit, and a task object generation unit. The action selection unit is used to select candidate improvement actions corresponding to employee action tasks from candidate improvement actions based on causal contribution markers and intervention accessibility status. The task conflict verification unit is used to identify task conflict results based on the incomplete tasks, completed tasks, and prohibited intervention status of the same customer object. The task object generation unit is used to generate employee action tasks based on the selected candidate improvement actions when the task conflict result is no conflict, and write customer object, responsible employee, task type, execution window, reach method, candidate repurchase anomaly factor, causal path set and intervention reachability status.

9. The store operation diagnosis system based on causal reasoning according to claim 1, characterized in that, The feedback update module includes a feedback sample generation unit, a graph data update unit, and a task priority update unit. The feedback sample generation unit is used to record the task status, executing employee, contact method, customer response, appointment result, in-store result and repeat purchase result after an employee's action task is executed, canceled, reassigned or overdue, and to synthesize the aforementioned records with the causal path set into a feedback sample; The graph data update unit is used to update the edge attribute table according to the correspondence between task status and customer behavior results in the feedback samples; the task priority update unit is used to adjust the task priority of candidate improvement actions under similar repurchase risk instances according to the updated edge attribute table.

10. The store operation diagnosis system based on causal reasoning according to any one of claims 1 to 9, characterized in that, It also includes a diagnostic link recording module, which is used to write a link identifier for the same repeat purchase risk instance, and write the link identifier into the candidate repeat purchase anomaly factor, causal path set, intervention accessibility status, employee action task and feedback sample respectively; The diagnostic link recording module is also used to output task triggering reason records and task invalidation reason records according to the link identifier. The task triggering reason records include repurchase risk triggering conditions, candidate repurchase abnormal factors, causal contribution markers and reachable intervention status. The task invalidation reason records include task status, customer response, appointment result, in-store result and repurchase result.

Images