Dynamic real-time rating and tracking method for pre-sales full-life-cycle clue arrival and ordering intentions

Abstract

The invention discloses a pre-sales full-life-cycle clue arrival and ordering intention dynamic prediction and tracking method, and the method comprises the following steps: obtaining client feature data of a target client, inputting the data into a pre-trained dynamic rating model, outputting an arrival intention level of the target client, and for a high-intention client, carrying out the dynamic rating of the target client. And the optimal store and the optimal telemarketing consultant are allocated, and the follow-up strategy is rearranged, so that the store arrival rate and the transaction conversion rate of the customer are improved. According to the method, the client feature data of the target client with the reserved capital to reach the store is mined, the return visit rate of the client is improved, the client feature data is input into the high-fitting-degree dynamic rating model, the high-intention client is predicted, the high-intention client is redistributed and issued to the proper store to follow up with the telemarketing consultant, and the in-store / transaction conversion rate is improved. The dynamic rating model is used for rating the clues for multiple times, the customer condition is predicted and analyzed in time, and the customer conversion rate is higher.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically is a method for dynamically predicting and tracking the arrival of leads at a store and the intention to place an order during the entire pre-sale life cycle. Background Art

[0002] In the sales industry, the factory collects information about potential customers through different channels, and after cleaning, sends this information to the corresponding stores for follow-up. This process involves judging the customer's intention to buy a car, and aims to convert potential customers to test drive (in-store) / car purchase (order / deal).

[0003] The current lead-to-store / transaction conversion rate is low, and the lack of follow-up has led to a large number of potential customer losses. Currently, we rely on rule screening to retrieve customer leads, and the conversion rate of leads that are inaccurately retrieved is also low. Summary of the invention

[0004] In order to solve the defects and shortcomings of the prior art, the present invention provides a method for dynamically predicting and tracking the store arrival and ordering intention of pre-sales leads throughout the entire life cycle, predicting high-intent customers so as to re-send follow-ups and improve the store arrival / deal conversion rate.

[0005] The present invention adopts the following technical solutions: The method for dynamically predicting and tracking the lead's store arrival and order intention during the pre-sales life cycle includes the following steps: Obtain characteristic data of target customers; Establish and train a dynamic rating model, input customer feature data into the trained dynamic rating model, and predict the target customer's intention to visit the store; Determine the target customers' store visit intention level based on store visit intention, and predict customers with high order intention; For customers with high order intention, the best stores and telesales consultants will be assigned and follow-up will be rearranged.

[0006] Furthermore, the training method of the dynamic rating model comprises the following steps: Step 21: Acquire multiple customer information as sample information; Step 22: Perform feature engineering on the obtained sample data to construct the features required for training and classify customer types; Step 23, dividing the sample information data into a training set and a test set according to proportion; Step 24: Train different types of customers in different scenarios separately and build multiple sub-models; Step 25: Incorporate customer types into features, combine all types of customer training sets for training, and establish panoramic sub-modeling; Step 26: Evaluate the fit between customer characteristics and each sub-model, and optimize the model based on AUC to improve model accuracy.

[0007] Furthermore, the sample information includes basic information of customer type, number of transactions, intended car series, minimum retention days, minimum follow-up days, number of follow-up stores and last follow-up date, last follow-up store, number of followed-up, number of unfollowed-up, total call duration, maximum call duration, number of effective follow-ups, number of retention channels, first retention channel, last retention channel, first follow-up rating, last follow-up rating, number of unsent leads, number of unassigned leads, number of unfollowed-up leads, number of leads being followed up, number of overdue leads, number of leads not being followed up for a long time ... Number of follow-up leads, number of defeated leads, whether there are advertising leads, whether there are leads from Yiche.com, whether there are leads from Autohome, whether there are leads from Dochedi, whether there are other leads, number of follow-up ratings, number of leads from the store in N days, number of orders from leads from the store in N days, order rate of leads from the store in N days, number of leads from telemarketing consultants in N days, number of orders from leads from telemarketing consultants in N days, order rate of leads from telemarketing consultants in N days, number of days the app was launched in the past month, number of days the mini program was launched in the past month, number of days the app was launched in the past week, number of days the mini program was launched in the past week.

[0008] Furthermore, after the store-side telemarketing consultant follows up, based on the ratings given during the communication process, the customer intention levels include: Class H indicates willingness to purchase a car within a week; Class A indicated willingness to purchase a car within two weeks; Class B indicates the willingness to purchase a car within one month; Class C indicates willingness to purchase a car within N days; N level refers to customers who have not yet been classified; F level refers to defeated customers; F0: The defeat application has been submitted but has not yet been approved; O level is for car ordering customers; Class D refers to vehicle delivery customers.

[0009] Furthermore, the number of follow-up ratings includes the number of follow-up ratings A, the number of follow-up ratings B, the number of follow-up ratings C, the number of follow-up ratings F, the number of follow-up ratings H, the number of follow-up ratings N, and the number of follow-up ratings O.

[0010] Furthermore, the customer types include eight types, namely, not issued, not allocated, not followed up, being followed up, overdue for follow-up, long-term non-follow-up, defeated, and others; among which: Unissued means a new customer, a customer ID that is not in the lead table. Unissued new customers are generally factory leads that have not yet been issued to stores. Unassigned means the leads in the lead table but not associated with any follow-up records; Not followed up means that the customer has been assigned to a telesales consultant, but the telesales consultant has not yet made the first telephone follow-up call to the customer. The rating of the customer not followed up in the lead table is H, A, B, C, N and is valid, and there are follow-up records associated with it. The actual follow-up time of all follow-up records is empty; "In Follow-up" means that the customer who has been followed up for the first time has not failed or the lead has expired, there is a plan for the next follow-up and the planned follow-up time has not yet arrived. The customer who is being followed up is rated H, A, B, C, N in the lead table and is valid, and the planned follow-up time has not expired; Overdue follow-up means that the customer who has been followed up for the first time has not failed or the lead has expired, there is a follow-up plan for the next time, and the follow-up has not been carried out after the planned follow-up time has passed. The customer who is overdue for follow-up is rated as H, A, B, C, N in the lead table and is valid, and the planned follow-up time has expired; Overdue follow-up means that the customer who has been followed up for the first time has not failed or the lead has expired, and it has been more than N days since the last follow-up date. The customer who has not been followed up for a long time is rated as H, A, B, C, N in the lead table and is valid, has a follow-up plan, and is more than N days away from the last actual follow-up date; Defeated means that the telesales consultant rated the lead as F or FO or invalid lead after follow-up, indicating that the customer has no intention to buy a car. The defeated customer is rated as F, O or invalid lead in the lead table; Other means that the customer information is collected within one day and there is no planned follow-up, actual follow-up, or the lead cannot be classified into the above categories.

[0011] Furthermore, the sources of the lead list include: information left by customers offline in stores, showrooms, etc., information left on APPs, mini-programs, online, or on third-party platforms, as well as some activities to attract customers to leave information.

[0012] Furthermore, the unissued, unallocated, and unfollowed-up items are integrated into the unissued scenario model; the overdue follow-up, follow-up in progress, and long-term non-follow-up are integrated into the follow-up in progress scenario model; and the defeat is a separate defeat scenario model, which is used to predict and analyze different types of customer situations.

[0013] Furthermore, the dynamic rating model is a LightGBM algorithm model based on a gradient boosting decision tree, which obtains the final customer intention rating by weighted averaging and fusing the prediction results of each sub-model.

[0014] Furthermore, the relationships between customers and stores are sorted, and the stores with the highest likelihood of placing orders and the telesales consultants with the highest store order rates are output.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention improves the customer return rate by mining the customer characteristic data of target customers who have left information and are waiting to visit the store, inputs the customer characteristic data into a high-fit dynamic rating model, predicts high-intent customers, and reallocates high-intent customers to appropriate stores and telemarketing consultants for follow-up and tracking, thereby improving the store visit / deal conversion rate. The present invention uses a dynamic rating model to rate leads multiple times, making the marking more accurate, the second connection more accurate, and analyzing the customer situation and arranging follow-up, thereby increasing the customer conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a flow chart of dynamic rating model training according to an embodiment of the present invention; Figure 2 A flow chart for following up with a customer for an embodiment of the present invention; Figure 3 It is a flowchart of a method for dynamically predicting and tracking the store arrival and ordering intention of pre-sales full life cycle leads in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a dynamic rating model according to an embodiment of the present invention; Figure 5 The following is a flow chart of customer classification according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the technical solution of the present invention, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments described are only a part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work should belong to the scope of protection of the present invention.

[0019] like Figure 1~Figure 5 As shown, a method for dynamically predicting and tracking the store arrival and ordering intention of pre-sales leads throughout the entire life cycle of this embodiment includes the following steps: (1) Obtain characteristic data of target customers and pre-process the data; The preprocessing includes performing feature engineering on the acquired data to construct features required for training, specifically including: Fill missing values, fill continuous value features with 0, fill discrete value features with empty strings, etc.; Continuous value binning: For features with more than 10 continuous value categories, equal frequency binning is performed; Feature combination, calculate the lead order rate of stores and telemarketing consultants through the store's historical follow-up records as model features.

[0020] (2) Establish and train a dynamic rating model; The dynamic rating model in this embodiment is a LightGBM algorithm model based on a gradient boosting decision tree. The dynamic rating model includes multiple sub-models, and the final customer intention rating is obtained by weighted averaging and fusing the prediction results of each sub-model.

[0021] The model of this embodiment uses gradient boosting decision tree as the boosting method, the maximum number of iterations is 1000, the learning rate is set to 0.08, the maximum number of leaf nodes for each tree is 36, the maximum depth of the tree is not limited, the data ratio used in each iteration is 90%, data sampling is performed once every 3 iterations, the feature ratio used in each iteration is 60%, the L1 regularization parameter is set to 0.2, the L2 regularization parameter is set to 0, the smoothing parameter of the category feature is set to 8, the maximum number of bins is 255, the logarithmic loss of binary classification is used as the evaluation indicator, and the imbalanced data set adjustment is set to true.

[0022] The training method of the dynamic rating model comprises the following steps: Step (21), obtaining multiple customer information as sample information; The sample information includes basic information of customer type (including customer ID, customer city, customer province), number of transactions (including corresponding dealership city, dealership province), intended car series, minimum retention days, minimum follow-up days, number of follow-up stores (including last follow-up date, last follow-up store), number of followed-up, number of unfollowed-up, total call duration, maximum call duration, number of effective follow-ups, number of retention channels, first retention channel, last retention channel, first follow-up rating, last follow-up rating, number of unsent leads, number of unassigned leads, number of unfollowed-up leads, number of leads in follow-up, number of overdue leads, number of leads not followed up for a long time, and number of defeated leads. Number of requests, whether there are leads from advertising, whether there are leads from Bitauto, whether there are leads from Autohome, whether there are leads from Autohome, whether there are leads from Dochedi, whether there are other leads, number of follow-up ratings A, number of follow-up ratings B, number of follow-up ratings C, number of follow-up ratings F, number of follow-up ratings H, number of follow-up ratings N, number of follow-up ratings O, number of leads from stores in 90 days, number of orders from leads from stores in 90 days, ordering rate of leads from stores in 90 days, number of leads from telemarketing consultants in 90 days, number of orders from leads from telemarketing consultants in 90 days, ordering rate of leads from telemarketing consultants in 90 days, number of days the app was launched in the past month, number of days the mini program was launched in the past month, etc.

[0023] Step (22): perform feature engineering on the obtained sample data, construct the features required for training, divide customer types, and classify customers.

[0024] There are eight types of customers, namely, not issued, not assigned, not followed up, being followed up, overdue for follow up, long-term non-follow up, defeated, and others; among them: Unsent means new customers, leads (customer IDs) that are not in the lead list. Unsent new customers are generally factory leads that have not been sent to stores. The sources of the lead list include: customer information left offline in stores, showrooms, etc., online information left in APPs, mini programs, official websites, etc., or information left on third-party platforms, and some activities to attract customers to leave information, etc. Unassigned means the leads are in the lead table but are not associated with any follow-up records; for example, the leads have been sent to the store but have not yet been assigned to a specific telesales consultant.

[0025] Not followed up means that the customer has been assigned to a telemarketing consultant, but the telemarketing consultant has not yet made the first telephone follow-up call to the customer. The customer who has not been followed up is rated as H, A, B, C, or N in the lead table and is valid and has associated follow-up records, but the actual follow-up time of all follow-up records is empty; "In Follow-up" means that the customer who has been followed up for the first time has not failed or the lead has expired, there is a plan for the next follow-up and the planned follow-up time has not yet arrived. The customer who is being followed up is rated H, A, B, C, N in the lead table and is valid, and the planned follow-up time has not expired; Overdue follow-up means that the customer who has been followed up for the first time has not failed or the lead has expired, there is a follow-up plan for the next time, and the follow-up has not been carried out after the planned follow-up time has passed. The customer who is overdue for follow-up is rated as H, A, B, C, N in the lead table and is valid, and the planned follow-up time has expired; Overdue follow-up means that the customer who has been followed up for the first time has not failed or the lead has expired, and it has been more than 90 days since the last follow-up date. The customer who has not been followed up for a long time is rated as H, A, B, C, N in the lead table and is valid, has a follow-up plan, and is more than 90 days away from the last actual follow-up date; Defeated means that the telesales consultant rated the lead as F or FO or invalid lead after follow-up. It means that the customer has no intention to buy a car. The defeated customer is rated as F, O or invalid lead in the lead table; Others means the leads were created within one day and have no planned follow-up, actual follow-up, or the leads cannot be classified into the above categories.

[0026] Furthermore, after the store-side telemarketing consultant follows up, the customer intention level is described based on the rating given during the communication process: Level H means that they are willing to buy a car within a week (return visit within 3 days); Class A indicates willingness to purchase a car within two weeks (return visit within 7 days); Class B indicates the willingness to purchase a car within a month (return visit within 15 days); Class C indicates willingness to purchase a car within 90 days (return visit within 30 days); N level refers to customers who have not yet been classified; F level refers to defeated customers; F0: The defeat application has been submitted but has not yet been approved; O level is for car ordering customers; Class D refers to vehicle delivery customers.

[0027] Step (23), dividing the sample information data into a training set and a test set in proportion; Step (24), train different types of customers in different scenarios using training sets separately to establish multiple sub-models; in this embodiment, Figure 4 As shown in the figure, the corresponding sub-models trained for the eight customer types are fused into three sub-models through weighted average, namely, the non-delivery scenario model, the follow-up scenario model, and the defeat scenario model. The prediction results of each sub-model are fused to obtain the final model result.

[0028] Specifically, the above-mentioned non-issued, non-allocated, and non-followed-up are integrated into a non-issued scenario model, the follow-up overdue, follow-up in progress, and long-term non-follow-up are integrated into a follow-up in progress scenario model, and the defeat is a separate defeat scenario model, which is used to predict and analyze different types of customer situations.

[0029] Step (25): Incorporate customer type into features, combine all types of customer training sets for training, and build a panoramic sub-model for predicting and analyzing all customer situations separately.

[0030] Step (26): Evaluate the fit between customer characteristics and each sub-model, use the test set to evaluate the generalization ability of the trained model, and optimize the model based on AUC to improve model accuracy.

[0031] (3) Input the target customer data to be predicted into the trained dynamic rating model to obtain the target user's store visit intention rating score and personal basic information, determine the target customer's store visit intention level based on the store visit intention rating score, and predict customers with high order intention; The target customer data to be predicted in this embodiment includes customers who have left information within six months, excluding customers who visited the store / placed an order within six months and completed transactions within one year; Finally, we get customer ID, customer type, overall prediction score, individual prediction score, prediction score fusion result, rating, customer city, customer province, intended car series, last follow-up date, last follow-up store, dealership city, dealership province, clue content, backup field 1, backup field 2, backup field 3, and prediction date.

[0032] The target customers get the predicted probability through the model, such as Figure 3 As shown in the figure, users are divided into five levels of store intention according to the predicted store probability (1 point: predicted score is less than or equal to 0.2; 2 points: predicted score is greater than 0.2 and less than or equal to 0.4; 3 points: predicted score is greater than 0.4 and less than or equal to 0.55; 4 points: predicted score is greater than 0.55 and less than or equal to 0.7; 5 points: predicted score is greater than 0.7). The higher the store intention rating score, the higher the store intention. For customers with high order intention, the best store and telemarketing consultant are assigned, and follow-up is rearranged.

[0033] The dynamic rating model of this embodiment is updated regularly to predict the probability of a customer visiting the store. The update optimization time of the dynamic rating model can be set, for example, the update training is at 9:00 on the 1st of each month, and the prediction update is at 9:00 every day. The operation status monitoring and operation effect of the model can be notified by email or other means.

[0034] Within the set time, the data operation center (BI) obtains the customer leads and follow-up forms created on the day, and the dynamic rating model makes predictions based on the new follow-up forms every day (T). Figure 2 As shown in the figure, the high-intent leads predicted by the dynamic rating model go to the corresponding stores through the following three paths: 1. The high-intention rating output by the model is transmitted to the business middle office lead center (A2) through the interface middle office, covering the ratings manually filled in by sales consultants in the past, completing the lead rating adjustment and issuing it; 2. Aggregate the corresponding customers of the leads that the sales store has not followed up on within the deadline, and send them to the customer data platform (CDP&MA) for intelligent operation two days later. Then, send them to the lead center (A2) as new leads to match the related stores and telesales consultants for follow-up; 3. Similarly, if a high-intent lead identified by the model is defeated by the consultant, the information of the defeated customer will also be sent to the customer data platform (CDP&MA), and the defeated customer will be sent to the business middle office lead center (A2) as a new lead. Finally, the defeated customer will continue to land in the store for follow-up.

[0035] Specifically, customers are ranked by their relationship with stores (one store one code > store with information / follow-up > store closest to work / residence > store with the highest sales volume in the same city as information / residence > store with the highest conversion rate in the same city as information / residence), and the best store and telemarketing consultant are selected for follow-up, and finally the transaction is completed at the store. One store one code refers to the store that the customer scans and binds.

[0036] It should be understood that the various forms of processes described above may be reordered, with steps added or deleted. For example, the steps described in the present invention may be performed sequentially, in parallel, or in different orders, and the present invention is not limited thereto.

[0037] The above embodiments do not limit the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for dynamically predicting and tracking the store arrival and order intention of pre-sales leads throughout the entire life cycle, characterized in that: The following steps are involved: Obtain characteristic data of target customers; Establish and train a dynamic rating model, input customer feature data into the trained dynamic rating model, and predict the target customer's intention to visit the store; Determine the target customers' store visit intention level based on store visit intention, and predict customers with high order intention; For customers with high order intention, the best stores and telesales consultants will be assigned and follow-up will be rearranged.

2. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival and order intention according to claim 1 is characterized in that: The training method of the dynamic rating model comprises the following steps: Step 21: Acquire multiple customer information as sample information; Step 22: Perform feature engineering on the obtained sample data to construct the features required for training and classify customer types; Step 23, dividing the sample information data into a training set and a test set according to proportion; Step 24: Train different types of customers in different scenarios separately and build multiple sub-models; Step 25: Incorporate customer types into features, combine all types of customer training sets for training, and establish panoramic sub-modeling; Step 26: Evaluate the fit between customer characteristics and each sub-model, and optimize the model based on AUC to improve model accuracy.

3. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival and order intention according to claim 2 is characterized in that: The sample information includes basic information of customer type, number of transactions, intended car series, minimum retention days, minimum follow-up days, number of follow-up stores and last follow-up date, last follow-up store, number of followed-up, number of unfollowed-up, total call duration, maximum call duration, number of effective follow-ups, number of retention channels, first retention channel, last retention channel, first follow-up rating, last follow-up rating, number of unsent leads, number of unassigned leads, number of unfollowed-up leads, number of leads being followed up, number of overdue leads, and number of leads not followed up for a long time. Number of leads, number of defeated leads, whether there are leads from advertising, whether there are leads from Bitauto, whether there are leads from Autohome, whether there are leads from Autohome, whether there are other leads, number of follow-up ratings, number of leads from the store in N days, number of orders from leads from the store in N days, order rate of leads from the store in N days, number of leads from telemarketing consultants in N days, number of orders from leads from telemarketing consultants in N days, order rate of leads from telemarketing consultants in N days, number of days the app was launched in the past month, number of days the mini program was launched in the past month, number of days the app was launched in the past week, number of days the mini program was launched in the past week.

4. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival at the store and order intention according to claim 3 is characterized in that: After the store-side telemarketing consultant follows up, based on the ratings given during the communication process, the customer intention levels include: Class H indicates willingness to purchase a car within a week; Class A indicated willingness to purchase a car within two weeks; Class B indicates the willingness to purchase a car within one month; Class C indicates willingness to purchase a car within N days; N level refers to customers who have not yet been classified; F level refers to defeated customers; F0: The defeat application has been submitted but has not yet been approved; O level is for car ordering customers; Class D refers to vehicle delivery customers.

5. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival and order intention according to claim 3 is characterized in that: The number of follow-up ratings includes the number of follow-up ratings A, the number of follow-up ratings B, the number of follow-up ratings C, the number of follow-up ratings F, the number of follow-up ratings H, the number of follow-up ratings N, and the number of follow-up ratings O.

6. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival and order intention according to claim 3 is characterized in that: There are eight types of customers, namely, not issued, not allocated, not followed up, being followed up, overdue for follow up, long-term non-follow up, defeated, and others; among them: Unissued means a new customer, a customer ID that is not in the lead table. Unissued new customers are generally factory leads that have not yet been issued to stores. Unassigned means the leads in the lead table but not associated with any follow-up records; Not followed up means that the customer has been assigned to a telesales consultant, but the telesales consultant has not yet made the first telephone follow-up call to the customer. The rating of the customer not followed up in the lead table is H, A, B, C, N and is valid, and there are follow-up records associated with it. The actual follow-up time of all follow-up records is empty; "In Follow-up" means that the customer who has been followed up for the first time has not failed or the lead has expired, there is a plan for the next follow-up and the planned follow-up time has not yet arrived. The customer who is being followed up is rated H, A, B, C, N in the lead table and is valid, and the planned follow-up time has not expired; Overdue follow-up means that the customer who has been followed up for the first time has not failed or the lead has expired, there is a follow-up plan for the next time, and the follow-up has not been carried out after the planned follow-up time has passed. The customer who is overdue for follow-up is rated as H, A, B, C, N in the lead table and is valid, and the planned follow-up time has expired; Overdue follow-up means that the customer who has been followed up for the first time has not failed or the lead has expired, and it has been more than N days since the last follow-up date. The customer who has not been followed up for a long time is rated as H, A, B, C, N in the lead table and is valid, has a follow-up plan, and is more than N days away from the last actual follow-up date; Defeated means that the telesales consultant rated the lead as F or FO or invalid lead after follow-up, indicating that the customer has no intention to buy a car. The defeated customer is rated as F, O or invalid lead in the lead table; Other means that the customer information is collected within one day and there is no planned follow-up, actual follow-up, or the lead cannot be classified into the above categories.

7. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival and order intention according to claim 6 is characterized in that: The sources of the lead list include: information left by customers offline in stores, showrooms, etc., information left on APPs, mini-programs, online, or on third-party platforms, as well as some activities to attract customers to leave information.

8. The method for dynamically predicting and tracking the store arrival and ordering intention of pre-sales leads throughout the entire life cycle according to claim 6 is characterized in that: The unissued, unallocated, and unfollowed-up are integrated into the unissued scenario model; the overdue follow-up, follow-up in progress, and long-term non-follow-up are integrated into the follow-up in progress scenario model; and the defeat is a separate defeat scenario model, which is used to predict and analyze different types of customer situations.

9. The method for dynamically predicting and tracking the pre-sale full life cycle lead arrival at the store and order intention according to claims 1 to 8 is characterized in that: The dynamic rating model is a LightGBM algorithm model based on the gradient boosting decision tree, which obtains the final customer intention rating by weighted averaging and fusing the prediction results of each sub-model.

10. The method for dynamic prediction and tracking of pre-sale full life cycle lead arrival and order intention according to claim 9, characterized in that: Sort by the relationship between customers and stores, and output the stores with the highest possibility of placing orders and the telesales consultants with the highest store ordering rate.

Images