A multi-channel cooperative target customer cultivation system and method

Abstract

The application provides a multi-channel cooperative target customer cultivation system and method, and relates to the technical field of computers. User data from multiple channels is received to identify the identity of a user and generate a unique marketing user identifier for the user. When a cultivation task is performed on the user, all channels associated with the marketing user identifier of the user are obtained, the highest priority single channel is selected according to a preset channel priority strategy, and the marketing content of this time is sent to the user only through the single channel. A cultivation workflow containing conditional branch nodes is constructed, the cultivation workflow is driven by user behavior events, and subsequent predefined branch actions are triggered based on user behavior events and conditional branch nodes. The application solves the problem of marketing resource waste and user experience fragmentation caused by data silos, realizes the transformation of marketing processes from extensive broadcasting to precise personalized reach, and improves marketing efficiency, customer experience and business conversion rate.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and in particular relates to a multi-channel collaborative target customer cultivation system and method. Background Technology

[0002] With the fragmentation of the internet ecosystem, enterprise customer data is scattered across multiple isolated systems and platforms, such as traditional CRM, WeChat official accounts, mini-programs, WeChat Work, and advertising platforms. The data architectures and user identification systems of these systems are incompatible, creating severe data silos. Against this backdrop, existing technologies face the following fundamental technical bottlenecks when conducting user marketing and nurturing:

[0003] 1. Rigidity and High Error Rate of Identity Fusion Technology: Existing Customer Data Platforms (CDPs) or identity recognition solutions mostly use rules based on fixed priorities for identity merging (e.g., prioritizing matching phone numbers, then email addresses). This one-size-fits-all static rule cannot adapt to the business characteristics of different industries (e.g., phone numbers have high weight in the financial industry, while email addresses have high weight in the e-commerce industry), lacking flexibility. More seriously, it lacks support for fuzzy matching and comprehensive judgment of multiple evidences. When faced with data noise (e.g., incorrectly entered phone numbers), missing identifiers (new anonymous visitors), or conflicts (the same phone number corresponds to multiple WeChat OpenIDs), it is prone to erroneous merging or merging failures, leading to distorted user profiles and creating underlying data vulnerabilities for precision marketing.

[0004] 2. Disorder and resource waste in multi-channel outreach: Due to the lack of a unified user view, marketing tools cannot identify whether users across different channels are actually the same person. Operations personnel must repeatedly configure push notifications for the same target audience across multiple independent systems such as email, SMS, and WeChat. This results in valuable marketing budgets being used to repeatedly send the same content to the same user, causing significant resource waste and inconveniencing users due to frequent message sending, severely damaging customer experience and brand image. Existing solutions only provide simple channel management and fail to achieve intelligent decision-making for an optimal path for each user.

[0005] 3. The disconnect between behavioral data and identity data: Although numerous user behavior analysis tools exist, the behavioral events they collect (such as page views, clicks, and downloads) are often isolated from core user identity markers (phone numbers, email addresses), especially behavioral data during anonymity periods. This makes it impossible to construct a continuous and complete view of the user interaction journey from anonymity to recognition, resulting in distorted marketing analysis attribution and a lack of data support for strategy optimization.

[0006] Therefore, the industry urgently needs a new generation of systems and methods that can fundamentally solve the problem of user identity consistency, achieve intelligent channel collaboration, and drive personalized marketing processes based on real-time feedback. Summary of the Invention

[0007] Therefore, the technical problem to be solved by the present invention is to provide a multi-channel collaborative target customer cultivation system and method, which realizes full-process automation from data governance to precise reach and intelligent cultivation.

[0008] In a first aspect, the present invention provides a multi-channel collaborative method for cultivating target customers, comprising:

[0009] S1. Receive user data from multiple channels, identify user identities, and generate unique marketing user identifiers for each user;

[0010] S2. When performing a nurturing task on a user, obtain all channels associated with the user's marketing user identifier, select the single channel with the highest priority according to the preset channel priority strategy, and send the marketing content to the user only through the single channel.

[0011] S3. Construct a nurturing workflow that includes conditional branch nodes. The nurturing workflow is driven by user behavior events and triggers subsequent predefined branch actions based on the user behavior events and the conditional branch nodes.

[0012] Further, step S1 includes:

[0013] Based on the user data, a multimodal identity association matrix is ​​constructed, which includes business scenarios, identifier types, association rules, baseline confidence levels, and custom weights.

[0014] The association rules include association combinations and association scenarios; the association combinations are the combination relationships of different identifier types; the association scenarios are the correspondence relationships between the business scenarios and the identifier types;

[0015] The baseline confidence level includes the identifier baseline confidence level and the associated combination baseline confidence level; the identifier baseline confidence level is a preset confidence level for a single identifier; the associated combination baseline confidence level is a preset confidence level for the associated combination.

[0016] The custom weights include identifier weights and associated combination weights; the identifier weights are preset weights for a single identifier; and the associated combination weights are preset weights for the associated combination.

[0017] Furthermore, the association combinations include unidirectional associations, bidirectional associations, and multidirectional associations;

[0018] The unidirectional association indicates a subordinate relationship between the identifier types, where the identifier type is a parent identifier and the identifier type is a child identifier; the bidirectional association indicates a bidirectional mutual authentication relationship between two identifier types with equal priority; the multidirectional association indicates a cross-identifier mutual authentication relationship between three or more identifier types that point to the same user with equal priority.

[0019] For the unidirectional association, the baseline confidence level of the association combination is the baseline confidence level of the parent identifier; for the bidirectional association, the baseline confidence level of the association combination is the average of the baseline confidence levels of the involved identifiers.

[0020] For the multi-directional association, the baseline confidence level of the association combination is higher than the mean baseline confidence level of the identifiers involved.

[0021] Furthermore, the user identification process employs a weighted confidence model, including:

[0022] S11. Create a user profile for user data from each channel;

[0023] S12. Based on the multimodal identity association matrix, identify the matching identifier common to the two user profiles to be matched;

[0024] S13. Calculate the total association confidence score between the two user profiles based on the baseline confidence score and the custom weight;

[0025] S14. Compare the total confidence score of the association with a preset threshold to generate the marketing user identifier.

[0026] Further, in step S14, comparing the total association confidence score with a preset threshold includes:

[0027] The thresholds include a merging threshold and a conflict threshold; the merging threshold is used to determine the lowest weighted confidence score when two user profiles are merged; the conflict threshold is used to determine the highest weighted confidence score when two user profiles are from different users.

[0028] When the total confidence score of the association is greater than or equal to the merging threshold, it is determined that the two user profiles belong to the same user and identity merging is performed.

[0029] When the total confidence score of the association is less than or equal to the conflict threshold, the two user profiles are determined to belong to different users and remain independent.

[0030] When the total confidence score of the association is less than the merging threshold but greater than the conflict threshold, it shall be arbitrated manually.

[0031] Furthermore, when the total association confidence score is less than the merging threshold but greater than the conflict threshold, the following steps are performed before manual arbitration:

[0032] 1) Construct a probabilistic graph node from all matching identifiers and non-matching conflict identifiers of the two user profiles;

[0033] 2) Based on the association rules and baseline confidence levels defined in the multimodal identity association matrix, conditional probabilities are assigned to the dependencies in the probability graph;

[0034] 3) Using the belief propagation algorithm, calculate the final posterior probability that the two user profiles belong to the same user, given that some identifiers are known to match;

[0035] 4) Compare the posterior probability with a preset decision threshold. If the posterior probability exceeds the decision threshold, it is automatically merged; if it is below the decision threshold, it is automatically rejected. If no decision can be made, it is arbitrated manually.

[0036] Further, in step S13, calculating the total association confidence score includes:

[0037] When the associated combination is not present in the matching identifiers, the total confidence score of the association is the sum of the scores of all individual identifiers; the score of each individual identifier is the product of the identifier baseline confidence score and the identifier weight.

[0038] When the associated combination exists in the matching identifier, the total confidence score of the association is the sum of the scores of all associated combinations and the scores of all other single identifiers;

[0039] The association portfolio score is the product of the association portfolio baseline confidence level and the association portfolio weight.

[0040] Furthermore, the cultivation workflow in step S3 is implemented in the following way:

[0041] S31. Establish a visual nurturing workflow canvas to construct a multi-round nurturing process that includes a start node, action nodes, waiting nodes, and conditional branch nodes; wherein, the action nodes include sending marketing content or tagging users;

[0042] S32. After the cultivation workflow is started, the execution process is driven by the user behavior event; when the user behavior event that matches the condition set by the condition branch node in the cultivation workflow is captured, the action node corresponding to the subsequent condition branch node is executed.

[0043] Furthermore, the method also includes:

[0044] S4. Establish a user journey view, including:

[0045] S41. Continuously collect user behavior data from the multiple channels to obtain user behavior event streams;

[0046] S42. For each user behavior event in the user behavior event stream, determine the marketing user identifier through the multimodal identity association matrix;

[0047] S43. Using the timeline as the axis, aggregate and display the user behavior events that identify the marketing user to generate a user journey view.

[0048] Secondly, the present invention also provides a multi-channel collaborative target customer nurturing system, the system being used to implement the method described in the first aspect, comprising:

[0049] The unified identity fusion engine is used to receive user data from multiple channels, identify user identities, and generate unique marketing user identifiers for each user.

[0050] The collaborative outreach strategy executor is used to obtain all channels associated with the marketing user identifier of the user when performing nurturing tasks for the user, select the single channel with the highest priority according to the preset channel priority strategy, and send the marketing content to the user only through the single channel.

[0051] A dynamic nurturing workflow engine is used to construct nurturing workflows that include conditional branch nodes. The nurturing workflow is driven by user behavior events, and subsequent predefined branch actions are triggered based on the user behavior events and the conditional branch nodes.

[0052] Beneficial effects:

[0053] This invention achieves accurate and flexible matching of user identities across channels by setting custom weights and dynamic thresholds for customers; based on the generated unique marketing user identifier, the system constructs a collaborative outreach strategy for each user, intelligently selecting the best available channel for message delivery, fundamentally eliminating duplicate marketing.

[0054] Employing an event-driven dynamic nurturing workflow engine, it can automatically trigger the next round of personalized interaction based on real-time user behavior feedback, ultimately forming a closed-loop marketing optimization system based on a panoramic user journey view.

[0055] It effectively transformed the marketing process from extensive broadcasting to precise and personalized outreach, improving marketing efficiency, customer experience, and business conversion rates. Attached Figure Description

[0056] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0057] Figure 1 This is a schematic diagram of multimodal identity association in Embodiment 1 of the present invention;

[0058] Figure 2 This is a flowchart of the identity intelligent fusion algorithm based on a configurable weighted confidence model in Embodiment 1 of the present invention;

[0059] Figure 3 This is a flowchart of the canvas-based automated cultivation process in Embodiment 1 of the present invention;

[0060] Figure 4 This is a visualized user journey diagram in Embodiment 1 of the present invention. Detailed Implementation

[0061] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. The principles and features of the present invention are described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other. The embodiments given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0062] This invention relates to the intersection of customer relationship management (CRM) and digital marketing technologies in the field of computer technology. Based on a distributed system architecture, it is used for intelligent fusion of user identities across platforms and multi-channel collaborative interaction to cultivate target customers. The technologies involved include data processing, real-time stream computing, and automated marketing decision-making.

[0063] To address the problems described in the background section, this invention provides a fully automated process from data governance to precise targeting and intelligent nurturing through unified identity fusion.

[0064] Example 1

[0065] This embodiment provides a multi-channel collaborative method for cultivating target customers, including the following steps:

[0066] 1. Implement unified identity integration;

[0067] Unified identity integration is the data cornerstone of this invention. Its core task is to break down data silos and create a 360-degree unified view of marketing users for enterprises.

[0068] 1.1 Multimodal Identity Association Matrix

[0069] like Figure 1As shown, the multimodal identity association matrix is ​​the core data structure of the unified identity fusion engine. Essentially, it is a system of identity association rules that supports multi-dimensionality, scalability, and scenario-based adaptation. It aims to break down the isolation of identity identifiers from different channels and of different types, providing a standardized data association foundation for cross-platform identity fusion. This matrix possesses three core characteristics: full identifier coverage, configurable association rules, and dynamic confidence level adaptation. The specific design is as follows:

[0070] 1.1.1 Comprehensive System for Multi-Type Identifiers

[0071] The matrix includes a built-in set of identity identifiers for all enterprise marketing scenarios. It not only covers core identifiers for mainstream online and offline touchpoints but also supports custom extensions, ensuring comprehensive capture of various user identities. Specific categories and descriptions are shown in the table below:

[0072]

[0073] 1.1.2 Scenario-based Association Rule Configuration System

[0074] The matrix provides predefined basic association rules for each type of identifier combination and allows enterprises to flexibly adjust them according to business scenarios, ensuring that the association logic is highly adapted to actual business needs. The rule system includes the following core dimensions:

[0075] (1) One-way association rule:

[0076] One-way association: Applicable to the subordinate relationship of "parent identifier - child identifier" (such as UnionID and WeChat Official Account OpenID, one UnionID can be associated with multiple OpenIDs, but the reverse is not true).

[0077] Two-way association: Applicable to mutual authentication relationships of identifiers with equal priority (such as mobile phone number and corporate email address, the two of the same user can be associated and matched with each other).

[0078] Multi-directional association: Suitable for scenarios involving cross-verification of multiple identifiers (such as mobile phone number + email address + device fingerprint ID, all of which point to the same user).

[0079] (2) Scene adaptation rules

[0080] For different business scenarios (such as financial compliance, e-commerce marketing, and SaaS trial scenarios), the matrix supports configuring differentiated association activation rules:

[0081] Financial scenarios: Only enable the high-confidence association combination of "mobile phone number (real-name verification) + corporate email (corporate authentication)" and disable anonymous identifier association.

[0082] E-commerce scenario: Enable multi-dimensional association of "mobile phone number + email + device fingerprint ID + mini program OpenID" to adapt to the identity tracking of users throughout the entire shopping process.

[0083] Anonymous nurturing scenario: Allows weak associations of "persistent cookie ID + temporary cookie ID + page browsing behavior tags" to support continuous tracking of anonymous user behavior.

[0084] (3) Data quality verification rules

[0085] Built-in identifier format validation and validity verification logic ensures the accuracy of associated data.

[0086] Format validation: Standardize the format validation of mobile phone numbers (regular expression matching international area code + number length), email addresses (domain validity validation), OpenID (WeChat interface format validation), etc., and filter invalid identifiers.

[0087] Validation: Real-time verification via API to check whether the mobile phone number is out of service, whether the email address can receive emails normally, and whether the OpenID has expired (e.g., the user has unsubscribed from the official account), avoiding association errors caused by invalid identifiers.

[0088] 1.1.3 Confidence Level Basic Configuration System

[0089] The matrix predefines a basic confidence range for each type of identifier and associated combination, serving as the initial basis for subsequent weighted confidence calculations, while also allowing enterprises to dynamically adjust this range based on data quality. Details are as follows:

[0090] Single identifier base confidence level: divided according to the uniqueness, real-name authentication and stability of the identifier (e.g., UnionID is unique across WeChat applications, so the base confidence level is set to 1.0; temporary CookieID is easily expired, so the base confidence level is set to 0.2).

[0091] Associated combination basic confidence: For multi-identifier combination scenarios, a pre-defined combination confidence rule is set (e.g., the basic confidence of the "phone number + email" combination is the average of the basic confidence of the two, and the basic confidence of the "UnionID + WeChat Official Account OpenID" combination is equal to the basic confidence of UnionID due to the subordinate relationship).

[0092] 1.2 Dynamic Real-time and Batch Association Mechanism:

[0093] The association process is continuous and dynamic, triggered by four mechanisms:

[0094] Real-time event-driven: When a user performs a key action (such as submitting a form on the official website, logging in on a mini-program, or clicking a menu on a public account), the event is captured in real time and an immediate attempt is made to associate and match it with existing marketing users.

[0095] Batch Tasks: Provides a management interface that allows administrators to configure batch tasks to periodically (e.g., daily at midnight) scan CRM leads, customers, and other objects, and perform large-scale identity merging and deduplication based on preset rules (e.g., phone number matching).

[0096] Data Change Monitoring: The engine monitors the change logs (such as CDC) of core data tables. When a record's key identifier (such as a mobile phone number) is updated, a re-association process is automatically triggered to break the old association and establish a new one, ensuring that the data is always up-to-date and accurate.

[0097] Manual intervention: Provides a management backend that allows operations staff to manually merge or split two or more user profiles based on business intuition and additional information, handling complex situations that the algorithm cannot resolve.

[0098] 1.3 Identity Intelligent Fusion Algorithm Based on Configurable Weighted Confidence Model

[0099] This algorithm is the core and innovation of this invention. It abandons the traditional, rigid fixed priority matching rules and introduces a new, flexibly configurable intelligent fusion mechanism based on a weighted confidence scoring model.

[0100] 1.3.1 Multi-dimensional Identifier Confidence Management Matrix

[0101] The algorithm's core advantage lies in the full-scenario configurability of its parameters. Enterprises can flexibly adjust parameters according to industry characteristics, data quality, and business needs to ensure algorithm adaptability. All parameters support visual configuration and real-time application through the management backend. The core parameters are as follows:

[0102] (1) Identifier weighted confidence parameter

[0103] Base Confidence Score: An inherent confidence level (a floating-point number in the range of 0-1) preset for each type of identifier, calibrated based on industry data and practical experience. Specific values ​​are shown in the table below:

[0104]

[0105] Customer-DefinedWeight: A weighting coefficient adjusted by the enterprise based on business scenarios (default value 1.0, adjustment range 0.5-2.0), supporting differentiated configuration by industry, business line, and customer group.

[0106] (2) Decision threshold parameter

[0107] MergeThreshold: The lowest weighted confidence score required to merge two user profiles (default 1.0, configurable range 0.8-1.2).

[0108] For high-precision scenarios (such as financial customer identity matching): the merging threshold can be raised to 1.1, allowing only high-confidence matching files to be merged, thus reducing the risk of erroneous merging.

[0109] In high-coverage scenarios (such as anonymous user cultivation): the merging threshold can be lowered to 0.8, allowing the merging of profiles with moderate confidence levels, thereby improving identity fusion coverage.

[0110] ConflictThreshold: The highest weighted confidence score that determines whether two user profiles are different users (default value 0.5, configurable range 0.3-0.7).

[0111] In scenarios with high data noise (such as online promotion and customer acquisition), the conflict threshold can be increased to 0.6 to reduce erroneous splitting caused by data noise.

[0112] For scenarios with high data quality (such as internal customer data of an enterprise): the conflict threshold can be lowered to 0.4 to ensure that the profiles of different users are strictly separated.

[0113] (3) Arbitration configuration parameters

[0114] Arbitration trigger condition: Manual arbitration is automatically triggered when the weighted confidence score is between the conflict threshold and the consolidation threshold.

[0115] Arbitration Priority: Supports configuring arbitration priority based on weighted confidence score range, customer value involved, and importance of related channels (e.g., suspected conflict cases of high-value customers have higher priority than ordinary customers).

[0116] Arbitration timeout handling: Set the arbitration timeout period (default 24 hours). Cases that are not processed after the timeout can be configured to be handled in three ways: automatically judged by the merger threshold, automatically judged by the conflict threshold, or kept in the pending status.

[0117] As a specific implementation method, when the total association confidence score is less than the merging threshold but greater than the conflict threshold, the following steps are performed before manual arbitration:

[0118] 1) Construct a probabilistic graph node from all matching identifiers and conflicting identifiers of the two user profiles;

[0119] 2) Based on the association rules and baseline confidence levels defined in the multimodal identity association matrix, conditional probabilities are assigned to the dependencies in the probability graph;

[0120] 3) Using the belief propagation algorithm, calculate the final posterior probability that two user profiles belong to the same user, given that some identifiers are known to match;

[0121] 4) Compare the posterior probability with the preset decision threshold. If it exceeds the threshold, the decision will be automatically merged; if it is lower than the threshold, the decision will be automatically rejected. If no decision can be made, the decision will be arbitrated manually.

[0122] The following example illustrates how to perform deep conflict analysis based on a probabilistic graphical model (PGM) to achieve intelligent automatic arbitration and reduce manual intervention when the total association confidence scores of two user data records to be matched are between a preset conflict threshold and a merging threshold.

[0123] a. Scene setting

[0124] Suppose there are two user data records to be matched:

[0125] Record A: From the official website form, containing {phone number: Aphone, device fingerprint: DEV001, email: AUser@xxx.com}.

[0126] Record B: From a WeChat mini-program, containing {WeChat OpenID:BUser, Device Fingerprint:DEV001, Nickname:Zhang San}.

[0127] After calculation using the basic weighted confidence model, the total confidence score for the association between the two is 0.75. The preset merging threshold is 1.0, and the conflict threshold is 0.6. Therefore, 0.6 < 0.75 < 1.0, and this case falls into a gray area, requiring manual arbitration according to the basic procedure. At this point, the enhanced process of this invention is initiated.

[0128] b. Probabilistic graphical model construction and analysis process

[0129] Step 1: Constructing Probabilistic Graph Nodes

[0130] All relevant identifiers and their relationships between two records are abstracted into probabilistic graph nodes and edges.

[0131] Observed variable nodes (known matching status):

[0132] Match(phone number, email): The value is False (record B has no email address, assuming the phone numbers are different).

[0133] Match (Device Fingerprint): Value is True (same as DEV001).

[0134] Match(WeChat OpenID, Nickname): Value is False (Record A has no such information).

[0135] Hidden variable nodes (to be solved):

[0136] SameUser: A Boolean variable indicating whether record A and record B belong to the same user. This is the core of our solution.

[0137] Conflicting Evidence Nodes:

[0138] Conflict (Phone Number): Record A has a phone number, record B does not have a phone number, but record B has a WeChat OpenID (usually bound to a phone number), which constitutes a potential conflict.

[0139] Conflict (Email): Record A has a corporate email address, while record B does not.

[0140] Step 2: Define conditional probabilities based on the multimodal identity association matrix

[0141] The predefined association strength (confidence) is queried from the matrix as the basis for conditional probability.

[0142] Defined in the matrix:

[0143] P(Device fingerprint matching|SameUser=True)=0.85 (When matching device fingerprints, the probability of it being the same user is relatively high).

[0144] P(Device fingerprint matching|SameUser=False)=0.15 (The probability of different users but the same device is low, such as sharing a device).

[0145] For strong identifier missing (such as phone number mismatch), define its likelihood function as: P(phone number mismatch|SameUser=True)=0.1 (the probability that the same user's phone number does not appear in another channel), P(phone number mismatch|SameUser=False)=0.9.

[0146] As a strong identifier, the conditional probability of WeChat OpenID appearing alone is also defined by a matrix.

[0147] Step 3: Calculate the posterior probability using the belief propagation algorithm.

[0148] Input the above nodes and conditional probability table into a Bayesian network (a probabilistic graphical model) and run the BeliefPropagation algorithm.

[0149] The algorithm will take the following into consideration:

[0150] Evidence supporting the same user: The same device fingerprint (DEV001) is strong supporting evidence.

[0151] Evidence / conflict against the same user: Core real-name identifiers such as mobile phone number and email address are not directly matched, constituting a conflict or lacking evidence.

[0152] Dependencies between identifiers: For example, WeChat OpenID is usually bound to a mobile phone number, so the existence of OpenID may indirectly affect the interpretation of a missing mobile phone number.

[0153] After iterative calculations, the posterior probability P(SameUser=True|All Observed Evidence) was finally calculated to be 0.82. This means that, after considering all matching and conflicting evidence, there is an 82% probability that record A and record B belong to the same user.

[0154] Step 4: Compare with the automated arbitration confidence threshold and make a decision

[0155] An automated arbitration confidence threshold of 0.80 is preset. This threshold is higher than the conflict threshold but lower than the consolidation threshold and is used for this round of advanced arbitration.

[0156] Comparison: Posterior probability (0.82) > Automated arbitration confidence threshold (0.80).

[0157] Decision: Automatically determine that record A and record B belong to the same user, and perform identity merging. This case does not require entering the manual arbitration queue.

[0158] c. Results and Outputs

[0159] Automatically merge all data (form information, WeChat interaction behavior, etc.) under record A and record B into the same marketing user identifier (e.g., UID_1001).

[0160] The audit log records: "Case ID: Case_123, Baseline Confidence Score: 0.75, PGM Posterior Probability: 0.82, Automated Arbitration Threshold: 0.80, Decision: Automatic Merging. Identifiers involved in the calculation: Device Fingerprint (Match), Mobile Number (Conflict)...".

[0161] This case was not added to the human arbitration queue. Only when the posterior probability is very close to the threshold (e.g., 0.79-0.81) or the algorithm fails to converge due to serious contradictions in the evidence will the case be sent to the human arbitration queue, along with a visualization analysis chart of the probabilistic graphical model (showing the node probabilities) as an auxiliary basis for human adjudication.

[0162] Through the above methods, this approach achieves the following:

[0163] By removing a large number of seemingly complex but computationally-driven cases from manual work, it is expected to reduce unnecessary manual arbitration by 50%-70%.

[0164] Probabilistic graphical models can handle dependencies and conflicts between pieces of evidence more delicately, and their decision results are more scientific and reliable than simple weighted total scores.

[0165] Even when submitted to manual arbitration, the attached model analysis results can help operations staff quickly understand the reasons for hesitation, improving the efficiency and accuracy of manual adjudication.

[0166] See the algorithm execution flow. Figure 2 .

[0167] 2. Message delivery based on integrated identity

[0168] This method is based on unified marketing user data to achieve precise and user-friendly marketing outreach.

[0169] 2.1 Visual Target Audience Builder: Provides a drag-and-drop, condition-filtered user interface. Operations personnel can flexibly combine conditions from a unified attribute library of marketing users (integrating attributes from various channels) and a behavioral event library (such as browsing product pages more than 3 times in the past 7 days), preview the number of audience packages and profiles in real time, and quickly create precise target audience segments.

[0170] 2.2 One-on-One Collaborative Outreach Strategy Executor: When performing nurturing tasks for a target audience, it doesn't simply send messages to all channels within the audience package. Instead, it executes the following logic for each marketing user:

[0171] Channel Availability Check: Check which valid outreach channels are associated with this username (e.g., if there is a mobile phone number, SMS can be sent; if there is an OpenID, template messages for official accounts can be sent).

[0172] Channel Priority Strategy: Based on the company's preset channel priority strategy (e.g., WeChat > ​​Official Account > SMS > Email), select the highest priority channel currently available as the sole channel for this outreach.

[0173] Message delivery: Send a message to the user only once via the selected best channel.

[0174] This mechanism intelligently enables operational planning to be carried out once, ensuring that each user receives a message only once through the optimal channel, perfectly solving the problems of resource waste and user frustration.

[0175] 3. Feedback-based automated nurturing workflow

[0176] This invention achieves intelligent and automated marketing processes, enabling dynamic adjustments to the nurturing path based on real-time user feedback. Figure 3 .

[0177] 3.1 Visual Workflow Canvas: Provides a graphical interface that allows operations personnel to build complex multi-round nurturing processes by dragging and dropping nodes (such as start, send email, wait, conditional branch, tag), without the need for developer intervention.

[0178] 3.2 Event-Driven Workflow Engine. Once a workflow is triggered and started, it does not execute according to a fixed schedule, but is driven by user behavior events. A classic new product trial and nurturing process can be configured as follows:

[0179] First round of outreach: Sending emails to users who have submitted the new product trial form.

[0180] Waiting and listening: Entering a 3-day listening period.

[0181] Conditional branches

[0182] Branch 1 (User clicks link in email): Automatically capture this action event and trigger subsequent actions:

[0183] Immediately tag this user with the "High Intent - New Product X" label;

[0184] After 24 hours, a link to send more in-depth technical documents or invite you to an online seminar will be automatically sent.

[0185] Branch Two (User did not click on the email): After the 3-day monitoring period ends, the following actions will be automatically triggered:

[0186] Tag this user "Cultivation - New Product X";

[0187] Five days later, a promotional message with different value points will be sent via SMS.

[0188] 3.3 Closed-Loop Feedback and Adaptive Optimization. All user behaviors (positive and negative feedback) throughout the workflow are recorded and correlated back to the marketing user. This data not only drives the operation of individual processes but also aggregates into a massive data pool to analyze the effectiveness of different nurturing strategies, providing data insights for subsequent optimization of marketing content, channel strategies, and process design, and possessing the ability for continuous self-optimization.

[0189] 4. Unify the user journey view and marketing dynamic attribution

[0190] This invention not only focuses on reaching and nurturing customers, but also aims to provide businesses with a complete panoramic view of customer interactions, which is the basis for precision marketing decisions.

[0191] 4.1 Multi-channel Behavior Collection and Normalization: User behavior events across various touchpoints are collected in real-time or near real-time using technologies such as API interfaces, SDK tracking, and log parsing. These behaviors include, but are not limited to: reading articles on official WeChat accounts, clicking menus, accessing mini-programs, browsing official websites, downloading white papers, online inquiries, registering for events, opening / clicking emails, and replying to SMS messages. The collected raw behavioral data is cleaned and normalized to form a standardized flow of behavioral events.

[0192] 4.2 Attribution of Marketing User Behavior: Each collected behavioral event carries one or more identity identifiers (such as CookieID, OpenID, or mobile phone number). The unified identity fusion engine matches and attributes these behavioral events to existing marketing users in real time. Upon successful matching, the behavioral event is recorded as a marketing activity on the corresponding user's timeline.

[0193] 4.3 Visualizing the User Journey and Insights: A complete user journey view is provided for each marketing user. See [link / details]. Figure 4 This view, presented in a timeline format, clearly shows a user's interaction history across various channels, from their initial anonymous visit to leaving a lead. Marketing and sales personnel can see the complete user interaction trajectory at a glance, such as: "The user first read an industry report on a WeChat official account, then visited the official website and downloaded the product white paper the next day, and received a marketing email a week later and clicked on a case study link." This cross-channel, continuous view greatly enhances the understanding of customer interests, intentions, and stages, providing crucial context for personalized communication and completely changing the previous fragmented and scattered nature of behavioral data.

[0194] This invention has the following features:

[0195] 1) An identity intelligent fusion method based on a configurable weighted confidence model

[0196] This invention pioneers a dynamically configurable identity fusion algorithm. The algorithm presets a baseline confidence level for each identifier (such as phone number, email address, and OpenID) and allows users to adjust custom weights according to business scenarios. It calculates the weighted total confidence score among the user profiles to be matched, and then intelligently decides whether to merge identities by comparing it with merging and conflict thresholds. Suspected conflict cases in the middle are automatically pushed to a human-machine collaborative arbitration queue.

[0197] It completely solves the industry problems of rigid fixed priority matching rules, inability to adapt to diverse business scenarios, inability to comprehensively utilize multi-factor evidence, and poor fault tolerance when data is incomplete.

[0198] The following technical effects are achieved:

[0199] Extreme flexibility: It transforms identity integration strategies from a one-size-fits-all approach to a customizable one, perfectly adapting to the business characteristics of different industries such as finance, retail, and SaaS.

[0200] Comprehensive intelligent decision-making: Decisions are based on the weighted sum of multiple pieces of evidence, rather than a single identifier's veto, making them more scientific and reasonable, and significantly improving matching accuracy.

[0201] Precise quantification and auditability: Transforming the qualitative matching process into quantitative calculations ensures that every decision has clear and auditable data support, meeting data compliance requirements.

[0202] Efficient human-machine collaboration: By defining the gray area of ​​decision-making through a dual threshold mechanism, introducing human arbitration and providing detailed decision-making basis, it takes into account both the efficiency of automation and the accuracy of human processing.

[0203] 2) A non-intrusive collaborative outreach mechanism and an event-driven dynamic nurturing mechanism.

[0204] Based on a unified marketing user identifier, the process doesn't simply involve multi-channel broadcasting when executing tasks. Instead, it intelligently performs channel availability checks and channel optimization strategies for each user, ensuring they receive the message only once through the best available channel. Furthermore, the nurturing workflow doesn't execute according to a fixed schedule; instead, it's driven by real-time user behavior events (such as clicks and views), enabling dynamic and personalized multi-round interactions.

[0205] It solves the problems of wasted marketing resources, impaired customer experience, and rigid traditional marketing automation processes that cannot be adjusted in real time based on feedback caused by identity fragmentation.

[0206] The following technical effects are achieved:

[0207] Zero waste and zero hassle: By eliminating repeated outreach across multiple channels, the mechanism greatly saves marketing costs and improves customer experience and brand favorability.

[0208] Dynamic personalized nurturing: This has transformed the marketing process from a one-size-fits-all approach to a personalized approach for each user, automatically triggering the most appropriate next step based on real-time user feedback, greatly improving nurturing efficiency and conversion rates.

[0209] Increased operational efficiency: It enables one-time planning and automatic execution of personalized outreach to each individual, freeing operations staff from tedious, repetitive multi-channel operations.

[0210] 3) Closed-loop marketing insights and optimization based on a panoramic user journey

[0211] This invention normalizes user behavior events collected from various channels using technologies such as SDKs and APIs, and leverages a unified identity fusion engine to accurately and in real-time attribute them to unique marketing users. This generates a cross-channel, cross-time visualized user journey timeline view for each customer. This panoramic view is not only a tool for gaining insights into customers, but also provides a data foundation for optimizing marketing strategies, forming a closed loop of outreach-feedback-attribution-insight-optimization.

[0212] It solves the problems of fragmented and scattered user behavior data, inability to be associated with a unified identity, and difficulty in forming a complete customer perception and conducting effective marketing attribution analysis.

[0213] The following technical effects are achieved:

[0214] 360-degree customer insights: Provides a complete context for understanding customer interests, intentions, and stages, giving sales and marketing teams an unprecedentedly clear perspective.

[0215] Data-driven closed-loop optimization: All reach effects and user behaviors can be measured and analyzed, providing solid data insights to support the continuous optimization of content strategies, channel strategies, and nurturing processes, making marketing campaigns increasingly precise.

[0216] Breaking down data silos: Ultimately, the entire chain from underlying data to top-level applications was connected, building a user-centric precision marketing system.

[0217] Example 2

[0218] This embodiment is a multi-channel collaborative target customer nurturing system. The system is used to implement the method described in Embodiment 1, including:

[0219] The unified identity fusion engine is used to receive user data from multiple channels, identify user identities, and generate unique marketing user identifiers for each user.

[0220] The collaborative outreach strategy executor is used to obtain all channels associated with the marketing user identifier of the user when performing nurturing tasks for the user, select the single channel with the highest priority according to the preset channel priority strategy, and send the marketing content to the user only through the single channel.

[0221] A dynamic nurturing workflow engine is used to construct nurturing workflows that include conditional branch nodes. The nurturing workflow is driven by user behavior events, and subsequent predefined branch actions are triggered based on the user behavior events and the conditional branch nodes.

[0222] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

[0223] The applicant confirms that the data collection and use in the technical solution described in this application comply with relevant laws and regulations. This application aims to protect the data processing method itself, which can be implemented legally and compliantly. The applicant promises that it will obtain user authorization in accordance with the law in practical applications.

Claims

1. A multi-channel collaborative method for cultivating target customers, characterized in that, include: S1. Receive user data from multiple channels, and construct a multimodal identity association matrix based on the user data; By using a multimodal identity association matrix, user identities are identified, and a unique marketing user identifier is generated for each user. The multimodal identity association matrix includes business scenarios, identifier types, association rules, baseline confidence levels, and custom weights; The association rules include association combinations and association scenarios; the association combinations are the combination relationships of different identifier types; the association scenarios are the correspondence relationships between the business scenarios and the identifier types; The baseline confidence level includes the identifier baseline confidence level and the associated combination baseline confidence level; the identifier baseline confidence level is a pre-set confidence level for a single identifier; The baseline confidence level of the associated combination is the preset confidence level of the associated combination; The custom weights include identifier weights and associated combination weights; the identifier weights are preset weights for a single identifier. The associated combination weight is a preset weight for the associated combination; S2. When performing a nurturing task on a user, obtain all channels associated with the user's unique marketing user identifier, select the single channel with the highest priority according to the preset channel priority strategy, and send the marketing content to the user only through the single channel. S3. Construct a nurturing workflow that includes conditional branch nodes. The nurturing workflow is driven by user behavior events and triggers subsequent predefined branch actions based on the user behavior events and the conditional branch nodes.

2. The method according to claim 1, characterized in that, The association combinations include unidirectional associations, bidirectional associations, and multidirectional associations; The unidirectional association indicates a subordinate relationship between the identifier types, where the identifier type is a parent identifier and the identifier type is a child identifier; the bidirectional association indicates a bidirectional mutual authentication relationship between two identifier types with equal priority; the multidirectional association indicates a cross-identifier mutual authentication relationship between three or more identifier types that point to the same user with equal priority. For the unidirectional association, the baseline confidence level of the association combination is the baseline confidence level of the parent identifier; for the bidirectional association, the baseline confidence level of the association combination is the average of the baseline confidence levels of the involved identifiers. For the multi-directional association, the baseline confidence level of the association combination is higher than the mean baseline confidence level of the identifiers involved.

3. The method according to claim 1, characterized in that, The user identification process employs a weighted confidence model, including: S11. Create a user profile for user data from each channel; S12. Based on the multimodal identity association matrix, identify the matching identifier common to the two user profiles to be matched; S13. Calculate the total association confidence score between the two user profiles based on the baseline confidence score and the custom weight; S14. Compare the total confidence score of the association with a preset threshold to generate the unique marketing user identifier.

4. The method according to claim 3, characterized in that, In step S14, comparing the total association confidence score with a preset threshold includes: The thresholds include a merging threshold and a conflict threshold; the merging threshold is used to determine the lowest weighted confidence score when two user profiles are merged; the conflict threshold is used to determine the highest weighted confidence score when two user profiles are from different users. When the total confidence score of the association is greater than or equal to the merging threshold, it is determined that the two user profiles belong to the same user and identity merging is performed. When the total confidence score of the association is less than or equal to the conflict threshold, the two user profiles are determined to belong to different users and remain independent. When the total confidence score of the association is less than the merging threshold but greater than the conflict threshold, it shall be arbitrated manually.

5. The method according to claim 4, characterized in that, When the total confidence score of the association is less than the merging threshold and greater than the conflict threshold, the following steps shall be performed before manual arbitration: 1) Construct a probabilistic graph node from all matching identifiers and non-matching conflict identifiers of the two user profiles; 2) Based on the association rules and baseline confidence levels defined in the multimodal identity association matrix, conditional probabilities are assigned to the dependencies in the probability graph; 3) Using the belief propagation algorithm, calculate the final posterior probability that the two user profiles belong to the same user, given that some identifiers are known to match; 4) Compare the posterior probability with a preset decision threshold. If the posterior probability exceeds the decision threshold, it is automatically merged; if it is below the decision threshold, it is automatically rejected. If no decision can be made, it is arbitrated manually.

6. The method according to claim 3, characterized in that, In step S13, calculating the total confidence score of the association includes: When the associated combination is not present in the matching identifiers, the total confidence score of the association is the sum of the scores of all individual identifiers; the score of each individual identifier is the product of the identifier baseline confidence score and the identifier weight. When the associated combination exists in the matching identifier, the total confidence score of the association is the sum of the scores of all associated combinations and the scores of all other single identifiers; The association portfolio score is the product of the association portfolio baseline confidence level and the association portfolio weight.

7. The method according to claim 1, characterized in that, The cultivation workflow in step S3 is implemented in the following way: S31. Establish a visual nurturing workflow canvas to construct a multi-round nurturing process that includes a start node, action nodes, waiting nodes, and conditional branch nodes; wherein, the action nodes include sending marketing content or tagging users; S32. After the cultivation workflow is started, the execution process is driven by the user behavior event; when the user behavior event that matches the condition set by the condition branch node in the cultivation workflow is captured, the action node corresponding to the subsequent condition branch node is executed.

8. The method according to claim 1, characterized in that, The method further includes: S4. Establish a user journey view, including: S41. Continuously collect user behavior data from the multiple channels to obtain user behavior event streams; S42. For each user behavior event in the user behavior event stream, determine the unique marketing user identifier through the multimodal identity association matrix; S43. Using the timeline as the axis, aggregate and display the user behavior events that identify the unique marketing user to generate a user journey view.

9. A multi-channel collaborative target customer nurturing system, characterized in that, The system is used to implement the method as described in any one of claims 1-8, comprising: The unified identity fusion engine is used to receive user data from multiple channels, identify user identities, and generate unique marketing user identifiers for each user. The collaborative outreach strategy executor is used to obtain all channels associated with the user's unique marketing user identifier when performing nurturing tasks for the user, select the single channel with the highest priority according to the preset channel priority strategy, and send the marketing content to the user only through the single channel. A dynamic nurturing workflow engine is used to construct nurturing workflows that include conditional branch nodes. The nurturing workflow is driven by user behavior events, and subsequent predefined branch actions are triggered based on the user behavior events and the conditional branch nodes.

Images