CRM marketing management system
The CRM marketing management system addresses the challenge of managing marketing activities by visualizing funnel stages and optimizing promotions/campaigns through a matrix interface, enhancing customer-specific marketing strategies.
Patent Information
- Application Number
- KR1020250005595
- Authority / Receiving Office
- KR · KR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2026-07-21
AI Technical Summary
Existing CRM marketing systems lack a clear and efficient way to visualize and manage the aggregate status of marketing activities across different stages of the product purchase process, and to optimize promotions or campaigns based on customer behavior and funnel stages.
A CRM marketing management system that displays a matrix interface with funnel stages as row elements and promotion/campaign types as column elements, using a processing server to predict and execute optimized promotions or campaigns based on customer behavior history, and manage content loading and deployment states.
Enables clear visualization and management of marketing activities across product purchase stages, optimizing promotions/campaigns for each customer, and ensuring personalized and efficient marketing strategies.
Smart Images

Figure P1020250005595_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a customer relationship management (CRM) marketing management system. Background Technology
[0002] Customer Relationship Management (CRM) refers to the process by which a company analyzes and integrates internal and external data related to customers to maximize customer-centric support, and based on this, plans, supports, and evaluates marketing activities tailored to customer characteristics. It can also refer to a series of activities aimed at planning and managing CRM marketing activities centered around apps or websites designed to implement e-commerce. The problem to be solved
[0003] One embodiment of the present invention includes a CRM marketing management system in which each member can clearly grasp the aggregate status of CRM marketing from an interface screen in which a plurality of divided section areas are displayed in a matrix form according to the row elements and column elements, wherein each funnel stage from the initial stage to the final stage of a product purchase process is represented as a row element and different promotion or campaign types unfolded in each funnel stage are represented as column elements.
[0004] One embodiment of the present invention includes a CRM marketing management system in which, each funnel stage regarding a product purchase process is a row element, customers belonging to each row element are a population, and an optimized promotion or campaign for each customer is predicted and assigned as a column element to a different promotion or campaign, and the optimized promotion or campaign for each customer can be executed collectively according to a member's trigger from an interface screen. means of solving the problem
[0005] In order to achieve the above-mentioned objectives of the present invention and other objectives, the CRM marketing management system of the present invention comprises:
[0006] A processing server that transmits a web document toward a member's terminal to render an interface screen in which a plurality of divided section areas are expressed in a matrix form according to the row elements and column elements, wherein each funnel stage from the initial stage to the final purchase stage regarding the product purchase process is a row element and different promotion or campaign types unfolding in each funnel stage are column elements.
[0007] The processing server above, for each of the partitioned section areas,
[0008] Exclusively display one of three distinct states under the funnel stage as the row element to which the corresponding split section area belongs, wherein the column element to which the corresponding split section area belongs is unable to deploy a promotion or campaign, wherein the column element to which the corresponding split section area belongs is unable to deploy a promotion or campaign, wherein the column element to which the corresponding split section area belongs is unable to deploy a promotion or campaign due to the absence of content stored in association with the corresponding split section area, and wherein the column element to which the corresponding split section area belongs is capable of deploying a promotion or campaign due to the presence of content stored in association with the corresponding split section area.
[0009] The processing server described above can display the number of customers assigned to different partitioned section areas partitioned according to the column elements, based on the result of a classification task to assign customers belonging to the population to different promotion or campaign types as column elements, using customers belonging to each funnel stage as a population as a row element.
[0010] For example, the processing server represents three different states regarding each of the partitioned section areas using different display elements so that they are identifiable from one another, and
[0011] The indication regarding the above unexpandable state and the indication regarding the above content unloading state and content loading state render the above divided section area with different background colors, and
[0012] The above content unloading state and content loading state can be rendered with different colors for different special characters under the same background color.
[0013] For example, the processing server above, in the classification task,
[0014] Optimized promotions or campaigns can be predicted based on the online behavior history of each customer forming the above population and the history of promotions or campaigns conducted targeting those customers.
[0015] For example, the processing server may include a sequence model for predicting an optimized promotion or campaign from the time-series data, using the customer's online behavior history as time-series data with sequential relationships.
[0016] For example, the above processing server, despite the prediction of the sequence model inferred as optimization for the promotion or campaign previously deployed for the customer,
[0017] Among the next-ranked predicted promotions or campaigns or the first promotion or campaign in the promotion or campaign history, the one with the highest predicted probability is determined to be the optimization and can be reflected on the interface screen.
[0018] For example, the processing server may execute a promotion or campaign as a column element to which each divided section area belongs on the interface screen, depending on the trigger input of a member who has checked the interface screen.
[0019] For example, the processing server above, depending on the type of message service to which the promotion or campaign to be executed is to be transmitted,
[0020] On the interface screen above, it is possible to implement a summary of the content, deletion of images forming the content, or image-to-text generation for images forming the content from editing the content stored in conjunction with each divided section area.
[0021] For example, the above processing server,
[0022] Around the trigger point when a promotion or campaign is executed,
[0023] Individual data regarding customers who have taken a reaction or action after the above-mentioned trigger point moves to the next funnel stage along the column direction, and
[0024] Individual data regarding customers for whom no response or action was taken even after the trigger point can be rendered to move to different promotions or campaigns along the row direction, thereby rendering a modified interface screen.
[0025] Meanwhile, a CRM marketing management system according to another aspect of the present invention is,
[0026] A processing server that transmits a web document toward a member's terminal to render an interface screen in which a plurality of divided section areas are expressed in a matrix form according to the row elements and column elements, wherein the row elements and column elements or column elements and row elements are mutually exclusively selected from among the row elements and column elements, and the different promotion or campaign types unfolded in each funnel stage and the product purchase process from the initial stage to the final purchase stage are represented.
[0027] The processing server above, for each of the partitioned section areas,
[0028] Exclusively displaying one of three distinct states under the funnel stage as the row or column element to which the corresponding segmented section area belongs, wherein the deployment impossible state where the deployment of a promotion or campaign is impossible as the column or row element to which the corresponding segmented section area belongs, the content unloading state where the deployment of a promotion or campaign is impossible as the column or row element to which the corresponding segmented section area belongs due to the absence of content stored in association with the corresponding segmented section area, and the content loading state where the deployment of a promotion or campaign is possible as the column or row element to which the corresponding segmented section area belongs due to the content stored in association with the corresponding segmented section area.
[0029] The processing server described above can display the number of customers assigned to different partitioned section areas partitioned according to the column elements or row elements, based on the result of a classification task for assigning customers belonging to the population to different types of promotions or campaigns as column elements or row elements, using customers belonging to each funnel stage as row elements or column elements as a population. Effects of the invention
[0030] According to a CRM marketing management system according to one embodiment of the present invention, each member can clearly grasp the aggregate status of CRM marketing from an interface screen in which a plurality of divided section areas are displayed in a matrix form according to the row elements and column elements, wherein each funnel stage from the initial stage to the final stage of the product purchase process is represented as a row element and different promotion or campaign types unfolded in each funnel stage are represented as column elements.
[0031] According to a CRM marketing management system according to one embodiment of the present invention, each funnel stage regarding a product purchase process is used as a row element, and customers belonging to each row element are used as a population. An optimized promotion or campaign for each customer is predicted and assigned as a column element, and an optimized promotion or campaign for each customer can be executed collectively according to a member's trigger from an interface screen. Brief explanation of the drawing
[0032] FIG. 1 illustrates a diagram for explaining the overall configuration of a CRM marketing management system according to one embodiment of the present invention. Figure 2 shows a diagram illustrating a funnel model regarding the product purchasing process. FIG. 3 illustrates a diagram for explaining one form of an interface screen provided by a CRM marketing management system according to one embodiment of the present invention. FIG. 4 illustrates a different form of an interface screen provided by a CRM marketing management system according to various embodiments of the present invention. FIG. 5 illustrates another form of an interface screen provided by a CRM marketing management system according to various embodiments of the present invention. FIG. 6 illustrates a diagram showing an exemplary form of a screen that shows a form of promotion or campaign developed from a CRM marketing management system according to one embodiment of the present invention. Figure 7 illustrates a diagram explaining the architecture of a sequence model for predicting an optimized promotion or campaign for each customer using time-series data regarding customer behavior as input. FIG. 8 illustrates a diagram for explaining the application of a weight (Wh) for a hidden state (ht-1) at a previous time step, a weight (Wx) for an input (xt) at a current time step, an activation function (tan h), and a weight (Wy) for generating a hidden state (yt) or output at a current time step in a sequence model that can be applied to one embodiment of the present invention. FIG. 9 illustrates a diagram for explaining the prediction of an optimized promotion or campaign inferred at each time step from the current input regarding the hidden state at the previous time step and the customer behavior (funnel step at each time step) at each time step in a sequence model applicable to one embodiment of the present invention. FIG. 10 illustrates an architecture that forms a sequence-to-sequence including an encoder and a decoder, wherein each encoder and decoder includes an RNN neural network to perform a translation task of outputting an English sentence from an input German sentence, and a diagram is shown to explain the extraction of a context vector (v) from the text of a promotion or campaign through the RNN neural network forming the encoder. Figures 11 and 12 illustrate different diagrams for explaining a BERT model that can be applied to generate a summary of content loaded from a member or to extract a context vector regarding the entire context of content loaded from a member. FIG. 13 illustrates a contrastive pre-training (contrastive language image pre-training, CLIP) that takes a plurality of paired images and text corresponding to captioning describing the plurality of images as input, and trains text encoders and image encoders that match each other by maximizing cosine similarity between pairs that match between text embeddings output from each text encoder and image embeddings output from each image encoder, and minimizing cosine similarity between pairs that do not match each other. Specific details for implementing the invention
[0033] Hereinafter, a CRM marketing management system according to a preferred embodiment of the present invention will be described with reference to the attached drawings.
[0034] Referring to FIG. 1, a CRM marketing management system or a processing server (100) for overall operation and management of a CRM marketing management system according to one embodiment of the present invention can transmit a web document for rendering an interface screen in response to a request from a member's terminal (10) (interface screen rendering unit (110)), and can generate a web document for rendering an interface screen that reflects the prediction of the promotion / campaign optimization unit (115) (sequence model described later) for predicting an optimized promotion or campaign for each customer from the customer's behavior (movement path or history in the customer's kernel stage and the customer's current kernel stage, etc.) stored in the database (150), and transmit it to the member's terminal (10). Meanwhile, according to a trigger signal transmitted from the member's terminal (10), the processing server (100) may execute a promotion or campaign for each member according to the aggregation of the interface screen displayed on the member's terminal (10), for example, may transmit a promotion or campaign predicted to be optimized for each member according to the aggregation on the interface screen, and may send a promotion or campaign predicted to be optimized through various message services such as different types of app push, SMS, Talk, etc. (promotion / campaign sending unit 125). At this time, according to the constraints of different message services such as app push, SMS, Talk, etc., it may include a content editing unit (120) for implementing editing of content loaded from the member, and the content editing unit (120) may implement the generation of a text summary as loaded content or the generation of text-to-image as loaded content.The above promotion or campaign may be carried out to promote purchases from customers (or customers' terminals (20)) through a commercial web for commercial mediation operated by a member (or member's terminal (10)), and various promotions or campaigns such as notifications of recommended products, notifications of special exhibitions, issuance of coupons, and price discounts may be carried out.
[0035] A CRM marketing management system according to one embodiment of the present invention may provide a platform for planning and managing customer relationship management (CRM) marketing, wherein CRM refers to customer relationship management, which may mean a process in which a company analyzes and integrates internal and external data related to customers to maximize customer-centric support, and based on this, plans, supports, and evaluates marketing activities tailored to customer characteristics. In one embodiment of the CRM management system according to one embodiment of the present invention, it may mean a system capable of planning and managing CRM marketing activities that are carried out centered around an app or website for implementing electronic commerce or e-commerce.
[0036] A CRM management system according to one embodiment of the present invention can plan and manage marketing activities that can be unfolded at each stage of a series of product purchasing processes for purchasing products (e.g., a funnel model for a product purchasing process illustrated in FIG. 2), and can implement a system capable of managing CRM marketing overall, such as planning promotions or campaigns that can be unfolded at each stage as the product purchasing process progresses, such as promotions to induce customer inflow, promotions to induce membership sign-up, promotions to induce product search, promotions to induce adding to a cart, and promotions to induce product purchase, as illustrated in FIG. 3, for example; triggering planned promotions or campaigns; performing and managing the planning, execution, and performance analysis of promotions or campaigns; establishing more efficient CRM marketing strategies (or rules); and deriving plans for promotions or campaigns from established CRM marketing strategies. Furthermore, the system can implement a system capable of managing overall CRM marketing targeting members registered on a platform (or a processing server (100) operating the platform) that provides the CRM marketing management system according to one embodiment of the present invention. An interface screen rendered in a table format can be provided so that it can be checked clearly.
[0037] Referring to FIG. 3, a CRM marketing management system according to one embodiment of the present invention may provide an interface screen rendered in a table form to comprehensively check promotions or campaigns for promoting marketing of each member, or promotions or campaigns for promoting marketing through an app or website operated or managed by the member. On the interface screen, different types of promotions or different types of campaigns may be represented as column elements, and different types of funnel stages (each funnel stage or funnel marketing stage expressed in the funnel model of FIG. 2) may be represented as row elements, and a plurality of divided section areas partitioned in a matrix form according to each column element and row element may be represented. Along with the plurality of divided section areas represented on the interface screen, the state of each divided section area may also be represented. In one embodiment of the present invention, the state of the partitioned section area may mean a total of three different states, including a state where the promotion or campaign cannot be developed as a column element to which the corresponding partitioned section area belongs, a state where content regarding the promotion or campaign is unloaded on a database (150) connected to a processing server (100) in conjunction with the corresponding partitioned section area, a state where the promotion or campaign cannot be developed as a row element to which the corresponding partitioned section area belongs due to the absence of content regarding the promotion or campaign, and a state where content regarding the promotion or campaign is loaded on a database (150) connected to a processing server (100) in conjunction with the corresponding partitioned section area, a state where the promotion or campaign can be developed as a column element to which the corresponding partitioned section area belongs from the content regarding the promotion or campaign.
[0038] For example, regarding the state of each of the multiple partitioned section areas partitioned according to each column element and row element, with different promotions or campaigns as column elements and different funnel stages (funnel marketing stages) as row elements on the interface screen above, the state of being unable to deploy a promotion or campaign may mean, for example, a state in which the deployment of a promotion or campaign is impossible as the row element to which the corresponding partitioned section area belongs when considering the funnel stage (or funnel marketing stage) as the row element to which the corresponding partitioned section area belongs; for example, this state of being unable to deploy may be regardless of the member registered on the CRM marketing management system or the processing server (100) operating the CRM marketing management system, that is, the CRM marketing management system or the processing server (100) operating the CRM marketing management system according to one embodiment of the present invention may output an interface screen for personalized CRM marketing management for each member, and at this time, despite the interface screen being personalized for each member, the state of being unable to deploy is not personalized status information for each member, but, for example, as the row element to which each partitioned section area belongs on the interface screen expressed in a table form for CRM marketing management When considering funnel stages (or funnel marketing stages), this may imply that a promotion or campaign cannot be conducted as a column element to which the relevant segmented section area belongs; for example, in the sense that it is impossible to conduct a promotion or campaign to facilitate inflow targeting customers who have already progressed to the inflow stage of a relevant member or an app or website operated by that member within the funnel stage as a row element, for example, a promotion or campaign to facilitate inflow may be expressed as unconductable in a segmented section area belonging to the inflow stage as a row element (funnel stage), for example,In the sense that it is not possible to conduct promotions or campaigns to encourage adding items to the cart for customers who have already progressed to the cart stage as a row element (funnel stage), for example, in a segmented section area belonging to the cart stage as a row element (funnel stage), promotions or campaigns to encourage adding items to the cart may be expressed as being in a state where they cannot be conducted. For example, regarding the state where promotions or campaigns cannot be conducted as a column element belonging to the corresponding segmented section area under the funnel stage as a row element to which the corresponding segmented section area belongs, as illustrated in FIG. 3, for example, in a funnel stage or funnel marketing stage intended to induce the purchase of products already in the cart, promotions or campaigns such as inducing recommendations for other products (recommended product notifications) or notifications such as special exhibitions (special exhibition notifications) may be expressed as being in a state where they cannot be conducted.
[0039] As such, in one embodiment of the present invention, the "unable to unfold" state may display objective status information for each member rather than subjective status information for each member, for example, in an embodiment providing an interface screen for personalized CRM marketing management for each member, rather than displaying personalized status information for each member. This is done to indicate that a promotion or campaign of a column element cannot be unfolded according to the funnel stage of a row element to which the corresponding segmented section area belongs. In various embodiments of the present invention, the "unable to unfold" state may display personalized status information for each member or subjective status information for each member. For example, the "unable to unfold" state may be displayed as personalized status information or subjective status information for each member according to the policy of each member or the app or website operated or managed by each member, that is, according to a price discount policy or a point accumulation policy. For example, according to the price discount policy of the corresponding member, a promotion or campaign such as a price discount may be expressed as an "unable to unfold" state for the member sign-up induction stage as a row element in FIG. 3.
[0040] In one embodiment of the present invention, as a state of a partitioned section area partitioned according to the column element (promotion or campaign) and row element (funnel stage), a state of non-expansion may be displayed as common state information for each member, such as a state of non-expansion (e.g., a state of non-expansion according to the column element and row element to which the corresponding partitioned section area belongs) and a subjective state of non-expansion for each member (a state of non-expansion according to a personalized policy for each member). In one embodiment of the present invention, the state of each partitioned section area may display the loading or unloading state of content stored in a processing server (100) or a database (150) connected to the processing server (100) in conjunction with the corresponding partitioned section area. Thus, as a state for a plurality of partitioned section areas partitioned according to the row element and column element, the loading state and the unloading state of content are state information personalized for each member, and on the processing server (100) or the database (150) connected to the processing server (100) that provides a platform for CRM marketing management for each member, the corresponding partitioned section area and Subjective state information indicating the loading or unloading status of linked and stored content may be displayed. For example, the loading status of the content may mean a state in which content linked and stored with the corresponding partition section area exists, allowing for the development of a promotion or campaign as a column element to which the corresponding partition section area belongs, and the unloading status of the content may mean that promotion or campaign development as a column element to which the corresponding partition section area belongs is impossible due to the absence of content linked and stored with the corresponding partition section area, but can indicate that promotion or campaign development as a column element to which the corresponding partition section area belongs is possible as the content is loaded, that is, as the content linked to the corresponding partition section area is loaded from the corresponding member onto the processing server (100).
[0041] In one embodiment of the present invention, three types of states, including an unexpandable state, a content unloading state, and a content loading state, are defined as states for a plurality of partitioned section areas partitioned according to the column element and row element, and can be expressed exclusively from each other. For example, each partitioned section area can be exclusively classified into one of the three types of states: an unexpandable state, a content unloading state, and a content loading state. In the case of so-called hard clustering, which does not allow overlapping classifications, the state of each partitioned section area can be exclusively expressed as one of the three types of states described above: an unexpandable state, a content unloading state, and a content loading state.In this case, although the aforementioned "unable to deploy" state and "content unloading" state are currently states where the deployment of a promotion or campaign is impossible, they need to be rendered with different display elements because they are states where deployment is impossible for different reasons—one being objectively impossible regardless of member actions, and the other being a state where the deployment of a promotion or campaign is possible (content loading state) depending on a member action (the action of loading content). For example, the aforementioned "unable to deploy" state and the other two states may be expressed with different background colors; specifically, when the "unable to deploy" state is rendered with a white background, the other two states may be rendered with a background color different from white (e.g., a background color such as yellow). The classification of the other two states may be rendered using an asterisk (*) within the corresponding segmented section area, but in different colors. For example, for the "content unloading" state intended to attract member action or attention, a red asterisk may be used within the corresponding segmented section area One side can be displayed, and for the loading status of the other content, a blue asterisk can be displayed on one side of the corresponding divided section area.
[0042] Referring to FIG. 3, on the interface screen for CRM marketing management provided to each member, a plurality of segmented section areas partitioned by column elements and row elements as previously described may be displayed, and along with each segmented section area, the state of each segmented section area may be displayed, and as previously described, different display elements may be introduced to render them so as to be differentiated from one another. In addition, in one embodiment of the present invention, on the interface screen for CRM marketing management, customers belonging to different funnel stages (or different funnel marketing stages) may be represented as a population as a row element, and a classification task assigned to customers belonging to each population as a column element may be displayed. For example, the number of customers belonging to different funnel stages (or different funnel marketing stages) may be represented as the number of entities in the population as a row element, and while displaying the number of customers belonging to each row element (funnel stage) in this manner, the number of customers assigned to each column element may be displayed as the number of entities in each column element belonging to each row element. For example, in one embodiment of the present invention, the processing As a classification task, the server (100) can assign the number of customers belonging to each row element to a different column element belonging to each row element, and as a result of this classification task, the number of customers belonging to each row element of the population can be displayed on the interface screen, and the number of customers assigned to a different column element among the number of customers in the population can be displayed on the different column element belonging to each row element, and as a result, the number of customers in the population displayed on each row element can be expressed as the total sum of the number of customers assigned to a different column element belonging to the corresponding row element.Accordingly, in a CRM marketing management system according to one embodiment of the present invention, for customers belonging to the same funnel stage, CRM marketing can be planned or managed to maximize marketing performance according to each customer's behavior by observing customer behavior online, including on-site behavior observed on an app or website operated or managed by a member or customer behavior observed on another app or website, or by observing customer behavior history collected over time regarding customer behavior, and different promotions or campaigns can be implemented for different customers currently in the same funnel stage so as to implement personalized promotions or campaigns for each customer. To this end, a classification task can be performed by classifying customers belonging to the same funnel stage as row elements into different column elements (promotions or campaigns) of said population (e.g., promotion / campaign optimization unit (115) for predicting an optimized promotion or campaign for each member in FIG. 1). In one embodiment of the present invention, performing a classification task to assign each customer belonging to the population to a different promotion or campaign by analyzing the online behavior or behavior history of each customer forming the population and displaying the result of such classification task on an interface screen may mean that the number of members belonging to the population displayed in the row element on the interface screen or the number of members belonging to the promotion or campaign displayed in the column element belonging to each row element may be displayed as zero (0).
[0043] Referring to FIGS. 7 to 9, in one embodiment of the present invention, in a classification task for matching or predicting an optimized promotion or campaign for each customer so that an optimized promotion or campaign can be developed for each customer by analyzing the online behavior or behavior history of each customer forming a population, the customer's behavior history on the online (e.g., the movement path of the customer's funnel stage) is input as sequential data over time, and an optimized promotion or campaign in the current state can be predicted by inputting the sequential data. To this end, a CRM marketing management system according to one embodiment of the present invention may include a sequence model for performing the classification task.
[0044] In one embodiment of the present invention, the sequence model can predict an optimized promotion or campaign at each time step at different time steps (discrete time steps) by using the customer's behavior history (the customer's current funnel stage as an input at the current time step) on-line as time-series data, and the sequence model can generate a hidden state or output of the current time step from the input of the previous time step and a hidden state or output of the current time step, and can be implemented as a sequence model such as a recurrent neural network (RNN) or a long short term memory (LSTM).
[0045] More specifically, a sequence model for predicting a promotion or campaign optimized for a customer at each time step, with the customer's funnel stage at each time step as input as the customer's behavior, may include a weight (Wh) for the hidden state (h) at the previous time step, a weight (Wx) for the input (x) at the current time step, and a weight (Wy) for the output (y), and may be represented as an expanded state of a memory cell as time steps advance, but the weight (Wh) for the hidden state (h) at each previous time step, the weight (Wx) for the input (x) at the current time step, and the weight (Wy) for the output (y) may be learned as constant values throughout the entire time step regardless of time steps advance, and may be maintained as constant values even in the inference step. For example, in one embodiment of the present invention, the hidden state (h) at any time step can be calculated by adding a bias (corresponding to an intercept) to a weighted sum with each weight (Wh, Wx) applied, and applying an activation function thereto. For example, in one embodiment of the present invention, a hyperbolic tangent can be applied as the activation function.
[0046] For example, the above sequence model may take the customer's behavior (funnel stage at the current time step) at each discrete time step as input, and may generate a hidden state or output of the current time step from the input at the current time step and the hidden state or output of the previous time step, for example, the hidden state or output of the current time step may include a prediction regarding a promotion or campaign optimized for the customer, and as previously described, may calculate prediction probabilities regarding different promotions or campaigns as column elements on an interface screen, and may determine the promotion or campaign for which the highest prediction probability is calculated among the prediction probabilities that each different promotion or campaign belongs to optimization as column elements as optimization, and may classify the corresponding member as a promotion or campaign predicted to be optimized according to the result of such a classification task. For example, in one embodiment of the present invention, the processing server (100) renders the interface screen by reflecting the prediction of a sequence model that performs a classification task, and if the promotion or campaign predicted as optimized with the highest probability in the classification task is in a stage where it cannot be deployed or in a content unloading state, the promotion or campaign with the highest prediction probability among the promotion or campaign with the next highest prediction probability or the promotion or campaign in a content loading state may be classified as optimized.
[0047] A CRM marketing management system according to one embodiment of the present invention may not conduct the same promotion or campaign targeting the same member, and given that the same promotion or campaign targeting the same member may be perceived as spam advertising that can induce negative sentiment in customers, the CRM marketing management system may have a policy set not to conduct the same promotion or campaign targeting the same member (e.g., the same member observed at the same funnel stage). Accordingly, regarding a promotion or campaign predicted as optimized by a sequence model, the CRM marketing management system may not reflect the prediction of such a sequence model, but instead reflect the promotion or campaign predicted in the next rank or the promotion or campaign that is the first in the history of promotion or campaign deployment as the one with the highest predicted probability. For example, in one embodiment of the present invention, a promotion or campaign predicted as optimized at each time step may be predicted based on the behavior of a member at each time step, and the promotion or campaign predicted at each time step (a promotion or campaign executed as predicted as optimized) may be input into the hidden state of the previous time step at the next time step, and a sequence model learned according to the policy of the processing server (100) not to deploy the same promotion or campaign for the same member (e.g., the same member observed at the same funnel step) may not recommend as optimized at the current time step for a promotion or campaign that was already deployed at the previous time step or predicted as optimized at the previous time step.For example, in one embodiment of the present invention, the hidden state at the previous time step (promotion or campaign predicted as optimized at the previous time step) is used as input to output the hidden state at the next time step (prediction regarding the optimized promotion or campaign), so even if the funnel step at each time step is used as input as the customer's behavior at each time step, the promotion or campaign predicted as optimized at the previous step can be avoided over time and a different promotion or campaign can be predicted.
[0048] The above sequence model can analyze the correlation between the customer's behavior history (funnel steps input sequentially at every hour step or time-series data) and the optimized promotion or campaign (target label) as time-series data (training data) input over time through learning, and the sequence model that learns the correlation between these data in the training stage can predict the optimized promotion or campaign from the unlearned time-series data (customer behavior) in the inference stage.
[0049] For example, in one embodiment of the present invention, the sequence model can predict a promotion or campaign that can derive the highest marketing performance from the history of customer behavior accumulated at each time step and the history of promotion or campaign launched for the customer. For example, if a promotion or campaign launched at a preceding time step elicited a customer action as a customer response at a subsequent time step, a high weight can be assigned to the corresponding promotion or campaign that elicited the customer response regarding the promotion or campaign that elicited the customer action.
[0050] In a CRM marketing management system according to one embodiment of the present invention, an optimized predicted promotion or campaign can be executed for each customer at a fixed trigger interval, and a learning step or an inference step can be performed by making this trigger interval a discrete step with respect to a time step, for example, a history regarding the behavior of a member observed at each trigger interval can be accumulated, and a history of a promotion or campaign can be accumulated at each trigger interval.
[0051] In one embodiment of the present invention, the sequence model may perform backpropagation through time (BPTT) in the learning step, using time-series data as input as learning data, in a direction that reduces the error between the predicted patient state and the target label assigned to the learning data, that is, so that the prediction from the learning data tracks the target label. For example, backpropagation through time in the sequence model may measure the error at each time step and propagate it to the previous time step. For example, the error between the target label and the prediction at each time step may be calculated, and parameters (weights and / or biases) may be updated from these errors. For example, the sequence model may update the weights (weights and / or biases, Wh) for the hidden state at the previous time step, the weights (weights and / or biases, Wx) for the input (x) at the current time step, and the weights (weights and / or biases, Wy) for the output (y). A sequence model including weights updated from such learning data is temporally different By using the history of customer behavior on the online platform accumulated in discretized time steps and the history of promotions or campaigns conducted targeting those customers as sequential data or sequential inputs, it is possible to predict an optimized promotion or campaign as a final output.
[0052] In one embodiment of the present invention, the number of members displayed in each segmented section area on an interface screen for managing CRM marketing may display data regarding the execution status at the time when a promotion or campaign to which the corresponding segmented section area belongs is executed according to a pre-set trigger interval or a member's trigger input, or may display data regarding the execution status at the time when the corresponding interface screen is rendered according to a member's request. For example, in an interface screen for managing CRM marketing according to one embodiment of the present invention, overall statistical data including the number of members displayed in each segmented section area may be updated according to a pre-set trigger interval or a member's trigger input, or at the time when the interface screen is rendered, and accordingly, an optimized promotion or campaign for each member at each corresponding time may be predicted. The input data of the above sequence model may include data regarding input times, and an optimized promotion or campaign can be predicted from the sequence model by using input times for each time step along with the customer's behavior history online, where each prediction time point (pre-set trigger interval, member's trigger input time, interface screen rendering time, etc.) may not be accumulated at equal intervals separated from each other along the time axis. For example, data regarding input times may be included as training data for learning the sequence model and input data for inference.
[0053] In one embodiment of the present invention, individual data (α) regarding each customer on the interface screen may move to different promotions or campaigns along the row direction, before and after a trigger in which a promotion or campaign is executed. For example, individual data (α) regarding a customer for whom no customer response or action is taken even after the trigger may move to different promotions or campaigns along the row direction (α2 in FIG. 3). In contrast, individual data (α) regarding a customer for whom a customer response or action is taken after the trigger may move to the next funnel stage (funnel marketing stage) along the column direction (α1, α1' in FIG. 3). In this way, depending on whether there is a customer reaction or action after the trigger, for example, individual data (α) of a customer for whom a customer reaction or action has been taken can move to the next funnel stage (funnel marketing stage) according to the customer's action (column direction movement α1, α1' in FIG. 3), for example, the movement of individual data (α) toward the next funnel stage along the column direction can be implemented (column direction movement a1, a1' in FIG. 3), and individual data of a customer for whom an action has been taken can move downward on an interface screen in which the first stage to the last purchase stage is arranged from an upward position toward a downward position in a funnel model regarding a product purchase process (see FIG. 2).
[0054] In one embodiment of the present invention, even after a trigger where a promotion or campaign is executed, no customer response or action is taken, but the individual data (α) of such member does not remain in the partitioned section area prior to the trigger, but can move along the row direction toward another promotion or another campaign belonging to the same row element belonging to a different promotion or another campaign as a different column element (α2 in FIG. 3). A CRM marketing management system according to one embodiment of the present invention may not execute the same promotion or campaign targeting the same member (the same member observed in the same funnel stage), and since the same promotion or campaign targeting the same member may be perceived as spam advertising that can induce negative sentiment in customers, the CRM marketing management system may have a policy set not to execute the same promotion or campaign targeting the same member. Accordingly, the above CRM marketing management system may move the individual data of a member toward another promotion or campaign in accordance with a policy that does not deploy the same promotion or campaign to the same member even if no reaction or action was taken by the member after the trigger of a previous promotion or campaign, and as previously described, in a classification task to classify each member according to the prediction of optimization among different promotions or campaigns, the above CRM marketing management system may not reflect the prediction of such a sequence model, even if the highest prediction probability as optimization is calculated for a previously deployed promotion or campaign from the sequence model to perform the classification task, and may reflect as optimization the promotion or campaign predicted in the next rank or the promotion or campaign that was the first in the history of promotion or campaign deployment, which has the highest prediction probability.
[0055] A CRM marketing management system according to one embodiment of the present invention can transmit a planned promotion or campaign through various message services including app push (push notification), Talk, and SMS according to a pre-set CRM marketing plan (e.g., according to aggregation on an interface screen rendered according to a member's request), for example, based on a trigger interval, or based on a member's trigger input, and the CRM marketing management system can process the content of the promotion or campaign converted to suit each type according to the type of message service through which the planned promotion or campaign is transmitted.
[0056] For example, the above CRM marketing management system can implement processing of the content of a promotion or campaign stored in conjunction with a segmented section area regarding a promotion or campaign classified by each member (content editing unit (120) of FIG. 1), taking into account the constraints of the overall message service, such as a fee system where images are not supported or separate marketing costs are incurred for transmissions involving images or where separate charges are added for transmissions involving images, or where the fee system exceeds the data capacity that can cause the fee system to rise rapidly even for content consisting of text without images, depending on the type of each message service (content editing unit (120) of FIG. 1), and regarding the content loaded from the member, it can be processed in an optimized form by reflecting the constraints of the above, for example, a summary of the content of the promotion or campaign, or deletion of images included in the content of the promotion or campaign, or creation of the corresponding image-to-text, taking into account the constraints of each individual message transmission.
[0057] Referring to FIGS. 10 through 12, in one embodiment of the present invention, a summary of the content of a promotion or campaign may be performed by extracting a context vector for the text forming the content of the promotion or campaign or by extracting a representation of the entire text forming the content of the promotion or campaign, and by generating a summary of the corresponding content of the promotion or campaign from the context vector or representation thus extracted. In one embodiment of the present invention, in the task of summarizing the content of a promotion or campaign, a context vector of the text forming the content of the promotion or campaign may be extracted from a sequence-to-sequence encoder including an encoder-decoder (see FIG. 10), or a representation of the entire text forming the content of the promotion or campaign may be extracted from a language model such as BERT (Bidirectional Encoder Representation from Transformer), and a summary may be generated from the extracted context vector or representation (see FIGS. 11 and 12). For example, a CRM marketing management system according to one embodiment of the present invention may include a Recurrent Neural Network (RNN) that generates a sequence of embedded input tokens for each word forming text from the content of a promotion or campaign, and sequentially inputs each input token forming the sequence according to a time step, and receives the hidden state (h) or output of the previous time step and the input of the current time step to generate the hidden state (h) or output of the current time step.For example, FIG. 10 illustrates an architecture that forms a sequence-to-sequence including an encoder and a decoder, wherein each encoder and decoder includes an RNN neural network to perform a translation task that outputs an English sentence from an input German sentence, wherein a context vector (v) can be extracted from the text of a promotion or campaign through the RNN neural network forming the encoder.
[0058] Referring to FIGS. 11 and 12, in one embodiment of the present invention, a task of summarizing the content of a promotion or campaign generates a sequence of input tokens embedded for words forming text from the content of the promotion or campaign, and may include a BERT (Bidirectional Encoder Representation from Transformer) neural network that receives each input token forming the sequence simultaneously, unlike the RNN neural network described above, and the first input token input to the BERT neural network may be a special token, and may include a [CLS] token corresponding to the representation of the entire content of the promotion or campaign, and the representation value (R) of the [CLS] token [CLS] ) may correspond to a context vector representing the entire content of a promotion or campaign. In summarization tasks involving the previously described RNN neural network or language models such as BERT neural networks, the context vector extracted from the content of the promotion or campaign or the representation of the entire content (total representation value, R) [CLS] A summary can be generated from ), for example, in one embodiment of the present invention, a summarized representation can be generated from a language model or embedding model that takes as input a context vector or a whole representation regarding the content of a promotion or campaign.
[0059] Referring to FIG. 12, the architecture of the encoder in the BERT model may include multi-head attention, which generates multiple attention matrices, concatenates multiple attention matrices, and multiplies them by a weight matrix (Wo) to produce a representation value of each word or token corresponding to the output.
[0060] Multi-head attention = concatenate(Z1,Z2,..,Zs)Wo
[0061] In the above BERT model, the attention matrix can be generated from self-attention as follows. That is, in self-attention, to generate the query matrix (Q), key matrix (K), and value matrix (V), W Q , W K , W V Three weight matrices are generated, and these weight matrices (W Q , W K , W V The query matrix (Q), key matrix (K), and value matrix (V) can be generated by multiplying the input matrix by ). Also, the inner product (QK) of the query matrix (query vector) and the key matrix (key vector) T The similarity value is calculated by ) and the calculated inner product (QK T ) is the dimension (d) of the key matrix (key vector). K The score matrix can be divided by the square root of ), normalized by applying a softmax function to the score matrix, and finally, the attention matrix (Z) can be produced by multiplying the value matrix by the normalized softmax function.
[0062]
[0063] In the above BERT model, the architecture of the encoder may include two sub-layers, which may include the previously described multi-head attention and feed-forward neural network (FFNN), and may include multiple encoder blocks in which inputs and outputs are sequentially connected, each containing the multi-head attention and feed-forward network, respectively. For example, the feed-forward network (FFNN) may be composed of a dense layer and an activation function. In addition, each encoder may include components of Add and norm for normalization (Add & norm), which connect the input and output values of the multi-head attention and the input and output values of the feed-forward network (layer normalization, residual connection).
[0064] A CRM marketing management system according to one embodiment of the present invention may delete images from the content of a promotion or campaign stored in conjunction with each partitioned section area according to the type of message service to transmit a promotion or campaign from a pre-planned CRM marketing trigger, or implement image-to-text generation. For example, in one embodiment of the present invention, among App Push (Push Notification), Talk, and SMS as message services to transmit the promotion or campaign, App Push (Push Notification) may not allow the transmission of images, whereas Talk or SMS may allow the transmission of images on the condition of transmission costs.
[0065] Referring to FIG. 13, in one embodiment of the present invention, captioning text to describe a corresponding image can be generated from an image-to-text task. For example, in the image-to-text task, captioning text can be generated from a multi-modal AI model that has learned a multi-modal embedding space of text-image or from a multi-modal embedding. The captioning text generated from the image-to-text task according to one embodiment of the present invention may be transmitted through a designated message service (e.g., an app push or push notification where image transmission is not allowed) together with the text of the content of a promotion or campaign, or may be transmitted through a designated message service (e.g., an app push or push notification where image transmission is not allowed) as text that is semantically connected to each other according to a similarity analysis (similarity based on Euclidean distance or similarity based on cosine similarity) with the text of the content of the promotion or campaign.
[0066] Referring to FIG. 13, in one embodiment of the present invention, a model trained to connect image and text-text pairs, i.e., image-text pairs, may be applied, and as described below, in one embodiment of the present invention, a multimodal AI model that has learned a multimodal embedding space of text-image may be applied to a spatial arrangement setting model or a background image generation model, etc.
[0067] FIG. 13 illustrates a contrastive pre-training (contrastive language image pre-training, CLIP) that takes a plurality of paired images and text corresponding to captioning describing the plurality of images as input, and trains text encoders and image encoders that match each other by maximizing cosine similarity between pairs that match between text embeddings output from each text encoder and image embeddings output from each image encoder, and minimizing cosine similarity between pairs that do not match each other.
[0068] For example, in one embodiment of the present invention, the multimodal AI model can convert an image and text into embeddings, respectively, and then predict the text embedding closest to the image embedding or, conversely, predict the text embedding closest to the text embedding. For example, in one embodiment of the present invention, CLIP (contrastive language image pre-training) can be applied as a multimodal AI model trained to connect image-text pairs to each other. In one embodiment of the present invention, the multimodal AI model can use web-based raw data, that is, large-scale raw data without human annotation from humans, as a training data set. For example, it can be trained using training data in which a web-based image and captioning text describing the image are image-text pairs, and can be trained using contrastive learning.
[0069] The above multimodal AI model may include an image encoder for producing a visual representation of an input image and a text encoder for producing a language representation of an input text, and input image-text pairs forming n pairs to each image encoder and text encoder to generate mini-batches of images (n image training data) and mini-batches of text (n text training data), then define text-image pairs that form pairs with each other as positive pairs and text-image pairs that do not form pairs as negative pairs, so that the cosine similarity for n positive pairs is maximized, and n 2 - The cosine similarity for n negative pairs can be learned to be minimized, and an image encoder and a text encoder can be trained together from such contrastive learning. For example, in one embodiment of the present invention, the multimodal AI model can learn the multimodal image-text embedding space, and the image encoder and the text encoder can be trained to map the respective image embeddings and text embeddings to each other, for example, from an image encoder that embeds an input image into an image space and a text encoder that embeds an input text into a text space.
[0070] Meanwhile, a CRM marketing management system according to another aspect of the present invention is,
[0071] A processing server that transmits a web document toward a member's terminal, wherein each funnel stage and different promotion or campaign types unfolding in each funnel stage from the initial stage to the final stage of the product purchase process are mutually exclusively selected from among row elements and column elements, and a plurality of divided section areas partitioned in a matrix form according to said row elements and column elements are rendered.
[0072] The processing server above, for each of the partitioned section areas,
[0073] Exclusively displaying one of three distinct states under the funnel stage as the row or column element to which the corresponding segmented section area belongs, wherein the deployment impossible state where the deployment of a promotion or campaign is impossible as the column or row element to which the corresponding segmented section area belongs, the content unloading state where the deployment of a promotion or campaign is impossible as the column or row element to which the corresponding segmented section area belongs due to the absence of content stored in association with the corresponding segmented section area, and the content loading state where the deployment of a promotion or campaign is possible as the column or row element to which the corresponding segmented section area belongs due to the content stored in association with the corresponding segmented section area.
[0074] The processing server described above can display the number of customers assigned to different partitioned section areas partitioned according to the column elements or row elements, based on the result of a classification task for assigning customers belonging to the population to different types of promotions or campaigns as column elements or row elements, using customers belonging to each funnel stage as row elements or column elements as a population.
[0075] For example, a CRM marketing management system according to a first embodiment of the present invention includes a processing server that transmits a web document toward a member's terminal to render an interface screen in which a plurality of divided section areas are expressed in a matrix form according to row elements and column elements, wherein
[0076] The processing server above, for each of the partitioned section areas,
[0077] Exclusively display one of three distinct states under the funnel stage as the row element to which the corresponding split section area belongs, wherein the column element to which the corresponding split section area belongs is unable to deploy a promotion or campaign, wherein the column element to which the corresponding split section area belongs is unable to deploy a promotion or campaign, wherein the column element to which the corresponding split section area belongs is unable to deploy a promotion or campaign due to the absence of content stored in association with the corresponding split section area, and wherein the column element to which the corresponding split section area belongs is capable of deploying a promotion or campaign due to the presence of content stored in association with the corresponding split section area.
[0078] The processing server described above can display the number of customers assigned to different partitioned section areas partitioned according to the column elements, based on the result of a classification task to assign customers belonging to the population to different promotion or campaign types as column elements, using customers belonging to each funnel stage as a population as a row element.
[0079] For example, a CRM marketing management system according to a second embodiment of the present invention includes a processing server that transmits a web document toward a member's terminal to render an interface screen in which a plurality of divided section areas are expressed in a matrix form according to row elements and column elements, wherein
[0080] The processing server above, for each of the partitioned section areas,
[0081] Exclusively display one of three distinct states under the funnel stage as the column element to which the corresponding segmented section area belongs, wherein the row element to which the corresponding segmented section area belongs is unable to deploy a promotion or campaign, wherein the row element to which the corresponding segmented section area belongs is unable to deploy a promotion or campaign, wherein the row element to which the corresponding segmented section area belongs is unable to deploy a promotion or campaign due to the absence of content stored in association with the corresponding segmented section area, and wherein the row element to which the corresponding segmented section area belongs is capable of deploying a promotion or campaign due to the presence of content stored in association with the corresponding segmented section area.
[0082] The processing server described above can display the number of customers assigned to different partitioned section areas partitioned according to the row elements, based on the result of a classification task to assign customers belonging to the population to different promotion or campaign types as row elements, with customers belonging to each funnel stage as the column element as the population.
[0083] Although the present invention has been described with reference to embodiments illustrated in the attached drawings, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Explanation of the symbols
[0084] 10: Member's terminal 20: Customer's terminal 100: Processing server 110: Interface screen rendering section 115: Promotion / Campaign Optimization Department 120: Content Editorial Department 125: Promotion / Campaign Dispatch Department 150: Database
Claims
Claim 1 A processing server that transmits a web document toward a member's terminal to render an interface screen in which a plurality of segmented section areas are expressed in a matrix form according to the row elements and column elements, wherein each funnel stage from the initial stage to the final purchase stage regarding the product purchase process is represented as a row element and different promotion or campaign types unfolding in each funnel stage are represented as column elements; wherein, for each segmented section area, the processing server exclusively displays one of three different states: a state where the promotion or campaign cannot be unfolded as a column element to which the corresponding segmented section area belongs, under the funnel stage as a row element to which the corresponding segmented section area belongs, a state where the promotion or campaign cannot be unfolded as a column element to which the corresponding segmented section area belongs, a state where the promotion or campaign cannot be unfolded as a column element to which the corresponding segmented section area belongs due to the absence of content stored in association with the corresponding segmented section area, and a state where the promotion or campaign can be unfolded as a column element to which the corresponding segmented section area belongs, according to content stored in association with the corresponding segmented section area; and wherein the processing server uses customers belonging to each funnel stage as a population as a row element, and the customers belonging to the population are column A CRM (customer relationship management) marketing management system characterized by displaying the number of customers assigned to different partitioned section areas partitioned according to column elements, based on the result of a classification task for assigning different promotion or campaign types as elements. Claim 2 A CRM marketing management system according to claim 1, wherein the processing server represents three different states regarding each of the divided section areas using different display elements so as to be identifiable from one another, and the indication regarding the unexpandable state and the indication regarding the content unloading state and the content loading state render the divided section areas with different background colors, and the content unloading state and the content loading state render different colors for different special characters under the same background color. Claim 3 A CRM marketing management system according to claim 1, wherein the processing server predicts an optimized promotion or campaign based on the online behavior history of each customer forming the population and the history of promotions or campaigns conducted targeting the corresponding customers in the classification task. Claim 4 A CRM marketing management system according to claim 1, wherein the processing server comprises a sequence model for predicting an optimized promotion or campaign from the sequential data, wherein the customer's online behavior history is treated as sequential data having a sequential relationship over time. Claim 5 A CRM marketing management system according to claim 4, wherein the processing server determines that the highest prediction probability among the next-ranked predicted promotion or campaign or the first promotion or campaign in the promotion or campaign history is the optimal one, despite the prediction of a sequence model inferred as the optimal for a promotion or campaign previously deployed for the customer, and reflects this on the interface screen. Claim 6 A CRM marketing management system according to claim 1, wherein the processing server executes a promotion or campaign as a column element to which each divided section area belongs on the interface screen in accordance with the trigger input of a member who has checked the interface screen. Claim 7 A CRM marketing management system according to claim 1, wherein the processing server implements a summary of content, deletion of images forming content, or image-to-text generation for images forming content from editing content stored in conjunction with each divided section area on the interface screen, depending on the type of message service to which the promotion or campaign to be executed is transmitted. Claim 8 A CRM marketing management system according to claim 1, wherein the processing server renders a modified interface screen such that, around a trigger point in which a promotion or campaign is executed, individual data regarding customers for whom a response or action was taken after the trigger point moves along the column direction to the next funnel stage, and individual data regarding customers for whom no response or action was taken even after the trigger point moves along the row direction to different promotions or campaigns. Claim 9 A processing server that transmits a web document toward a member's terminal to render an interface screen in which a plurality of divided section areas are expressed in a matrix form according to row elements and column elements, wherein, for each of the funnel stages from the initial stage to the final purchase stage regarding the product purchase process, and different promotion or campaign types unfolding in each funnel stage are mutually exclusively selected from row elements and column elements, and row elements and column elements and row elements, the processing server, for each of the divided section areas, [represents] a state in which the promotion or campaign cannot be unfolded as a column element or row element to which the corresponding divided section area belongs under the funnel stage as a row element or column element to which the corresponding divided section area belongs, a content unloading state in which the promotion or campaign cannot be unfolded as a column element or row element to which the corresponding divided section area belongs due to the absence of content stored in association with the corresponding divided section area, and a content loading state in which the promotion or campaign can be unfolded as a column element or row element to which the corresponding divided section area belongs due to content stored in association with the corresponding divided section area. A CRM (customer relationship management) marketing management system characterized by exclusively displaying, wherein the processing server displays the number of customers assigned to different partitioned section areas partitioned according to the column element or row element, based on the result of a classification task for assigning customers belonging to the population to different promotion or campaign types as column element or row element, using customers belonging to each funnel stage as a population as a row element or column element.
