Artificial intelligence-based customer manager digital clone system, method and medium

Abstract

The embodiments of the present disclosure disclose an artificial intelligence-based customer manager digital clone system, method and medium. It comprises a communication interface, a processor and a memory. By receiving a multi-modal service request of a customer, a dynamic customer portrait construction module is used to integrate data in real time to form a customer portrait; an intelligent decision engine generates a decision instruction based on the portrait, business rules and an AI model; a service automation interface automatically calls resources to perform service operations according to the instruction. The system adopts a distributed and event-driven architecture, and can integrate a memory network, continuous learning and a security and compliance module. The present application solves the problems of strong artificial dependence, response delay and data island in traditional customer service systems, realizes intelligent identification of customer demand, accurate decision and automatic service execution, and significantly improves service efficiency and customer experience.

Description

Technical Field

[0001] Applications generally involve the field of artificial intelligence technology, and in particular, an AI-based digital avatar system, method, and medium for customer managers. Background Technology

[0002] In customer service within industries such as telecommunications and finance, current technical solutions are primarily based on traditional manual processing and limited system support. After a customer submits a request through standardized channels such as a service hotline, online portal, or mobile application, the request is typically transferred to a human customer service representative or account manager. The handler needs to log into multiple independent business back-end systems, manually query scattered customer history information, product data, and business rules, and rely on their personal experience to subjectively assess the customer's needs and formulate a processing strategy. Once a solution is determined, staff still need to coordinate necessary resources across departments using traditional tools such as internal emails and instant messaging software. Ultimately, a human executes the specific operational instructions in the corresponding business system to complete the service loop.

[0003] The existing technical solution has several inherent flaws, leading to challenges in service efficiency, quality, and cost. Specifically, its shortcomings are mainly reflected in the following aspects: First, the level of automation and intelligence is low, and processing capacity is limited. Most of the automated functions integrated into existing systems rely on pre-defined, fixed rule bases or simply use relatively simple machine learning models such as decision trees and support vector machines. These technologies can only handle simple, highly standardized scenarios that are clearly defined in the rule base. When customer needs exceed the preset range, exhibit complexity, or have personalized characteristics, the system cannot effectively understand and process them, necessitating a process interruption and manual intervention. This leads to frequent "breakpoints" in the automated process, preventing the achievement of end-to-end intelligent services.

[0004] Second, there is insufficient understanding and support for dynamic interactive scenarios. While existing technologies can process formatted documents using optical character recognition and table extraction to build static knowledge bases, their focus is primarily on extracting and structuring information from fixed layouts. For dynamic interactive scenarios involving multi-turn dialogues and contextual relationships generated during service delivery, existing technologies lack deep semantic understanding, coherent contextual reasoning, and accurate intent and sentiment recognition capabilities. This makes it difficult for the system to accurately grasp the customer's true needs in real-world dialogue flows and provide intelligent responses with consistency and contextual relevance.

[0005] Third, the barriers to technological upgrades and professional deployment are high. Enterprises wishing to introduce more advanced deep learning models to enhance service intelligence for their specific business scenarios will face significant practical difficulties. This process typically involves complex model selection, extensive domain data cleaning and labeling, and demanding localized deployment, integration, debugging, and post-deployment maintenance. The high technical threshold, time costs, and resource investment severely restrict the deep application and rapid promotion of artificial intelligence technology in vertical business sectors.

[0006] In summary, existing customer service technology systems have significant shortcomings in terms of flexibility, intelligence, scenario adaptability, and implementation cost-effectiveness. Therefore, there is an urgent need for a new solution that can deeply integrate intelligent technologies, possess autonomous analysis and decision-making capabilities, and efficiently collaborate with existing enterprise infrastructure to systematically improve customer service experience and operational efficiency. Summary of the Invention

[0007] This disclosure addresses some of the shortcomings mentioned in the background art by providing an artificial intelligence-based digital avatar system, method, and medium for customer managers.

[0008] In a first aspect, embodiments of this disclosure provide an artificial intelligence-based digital avatar system for account managers, comprising: At least one client device is used to collect and send input data containing user multimodal interaction requests; The access gateway communicates with the client device to receive and standardize the input data, forming a standard request. A processing server, communicatively connected to the access gateway, the processing server comprising: The customer profile dynamic construction module is used to receive the standard request, construct and dynamically update the customer profile corresponding to the user based on multimodal interaction data and historical behavior data, and the customer profile includes behavioral characteristics, preferences and demand prediction information; The intelligent decision engine is connected to the customer profile dynamic construction module and is used to generate service strategies for the standard request based on the customer profile, preset business rules and artificial intelligence models. The service automation interface module is connected to the intelligent decision engine and is used to call internal functional modules or external system services based on the service strategy to execute corresponding service operations. The memory network and continuous learning module are connected to the customer profile dynamic construction module, the intelligent decision engine and the service automation interface module, respectively, to store customer interaction history data and to optimize and update the artificial intelligence model and business rules in the intelligent decision engine based on the data. In one embodiment of the first aspect, the customer profile dynamic construction module further includes: A multi-source data fusion unit is used to parse and identify the intent of the text, voice, and image information in the standard request; The real-time update and event-driven unit is used to calculate the incoming real-time behavioral data stream and extract immediate features; The intelligent analysis and prediction unit is used to integrate the real-time features, static archive data and structured knowledge provided by the memory enhancement and continuous learning module to output an updated customer profile.

[0009] In one embodiment of the first aspect, the intelligent decision engine employs a hybrid inference architecture, including: Rule units are used to execute predefined deterministic business rules; The model prediction unit is configured to integrate multiple trained machine learning models for performing neural network-based prediction and inference. The strategy unit is used to fuse the outputs of the rule unit and the model prediction unit to generate the service strategy.

[0010] In one embodiment of the first aspect, the service automation interface module adopts an event-driven architecture, including: The instruction parsing unit is configured to receive the service policy output from the intelligent decision engine module and generate an atomic sequence of service actions based on the service policy. The intelligent routing unit is used to convert the service actions into call instructions that conform to the protocols of different internal business systems; The task execution and status monitoring unit is used to monitor the execution status of service actions and feed the execution results back to the memory network and continuous learning module.

[0011] In one embodiment of the first aspect, the memory enhancement and continuous learning module includes: The memory encoding and vectorization unit is used to convert key information of customer service interaction history into vectors and store them in a vector database; The memory retrieval and context reconstruction unit is used to retrieve relevant memories from the vector database according to the current request context and input them into the intelligent decision engine to assist in decision-making; The incremental training unit is used to incrementally train and optimize the artificial intelligence model based on the feedback service result data.

[0012] In one embodiment of the first aspect, the processing server further includes a security and compliance assurance module, which interacts with the customer profile dynamic construction module, the intelligent decision engine, the service automation interface module, and the memory network and continuous learning module to provide security protection and compliance verification for the entire data lifecycle and business operation processes.

[0013] In one embodiment of the first aspect, the security and compliance assurance module is integrated into the system's data flow and business flow, and is used for: Encrypt and de-identify the input and output data; During data processing, a trusted execution environment is used for privacy-preserving computation; Real-time compliance verification of business operation processes is performed, and blockchain technology is used to store key operation logs as evidence.

[0014] In a second aspect, embodiments of this disclosure provide an AI-based digital avatar method for customer managers, applied to the AI-based digital avatar system for customer managers as described in the first aspect, comprising the following steps: The client device receives multimodal interaction requests from users, which are then standardized by the access gateway and sent to the processing server. The dynamic profile building module dynamically builds or updates the customer profile of the corresponding user based on the interaction request and associated data. The intelligent decision-making module generates a service strategy for the interaction request based on the customer profile, preset business rules, and artificial intelligence model. The automated service execution module automatically calls internal system resources and executes corresponding service operations according to the service policy. Through the memory enhancement and continuous learning modules, the interaction and execution results are stored as historical memory, and the artificial intelligence model and business rules are continuously learned and optimized based on the execution feedback.

[0015] In one embodiment of the second aspect, prior to performing the corresponding service operation, the method further includes: A service plan is generated based on the aforementioned customer management decision instructions; Perform automated compliance checks on the service plan; Specifically, performing the corresponding service operation involves deploying the service plan that has passed the compliance verification on the edge side. In a third aspect, embodiments of this disclosure provide a computer-readable storage medium having computer instructions stored thereon, which, when executed, perform the AI-based digital avatar method for customer managers as described in the second aspect.

[0016] This disclosure proposes an artificial intelligence-based digital avatar system, method, and medium for customer managers, which has the following beneficial effects: (1) Improved the intelligence and accuracy of customer service. This disclosure achieves a deep understanding and accurate prediction of customer needs by constructing dynamic customer profiles and adopting a hybrid intelligent decision-making mechanism. The system can analyze multimodal interaction information and combine real-time data with historical context to make inferences, thereby identifying complex and non-standardized service needs. The collaborative work of the rule engine and artificial intelligence model, while ensuring business compliance and explainability, gives the system the ability to handle unknown scenarios and make optimization decisions, significantly improving the pertinence and effectiveness of service strategies.

[0017] (2) End-to-end automation of service processes has been achieved, significantly improving operational efficiency. This disclosure breaks down the execution barriers between internal intelligent decision-making and external business systems through a service automation interface module. Based on event-driven and workflow orchestration technologies, the system can automatically decompose complex service strategies into atomic operation sequences that can be executed in heterogeneous systems, and complete scheduling and monitoring. This changes the traditional model of relying on manual serial operation of multiple systems, shortening business processing time from days to hours or even minutes, greatly reducing the proportion of manual intervention and operational errors, and freeing up human resources.

[0018] (3) Enhanced overall system reliability, scalability, and sustainable evolution capabilities. This disclosure adopts a cloud-native and microservice architecture, enabling each core functional module of the system to be deployed independently, scale elastically, and iterate rapidly. This architecture design ensures high availability of the system in high-concurrency scenarios and can be integrated with the enterprise's existing IT environment at a lower cost. More importantly, through the memory network and continuous learning module, the system can accumulate service experience and drive the algorithm model to continuously optimize itself, enabling the service capabilities of the entire digital avatar to continuously evolve with business development, reducing the technical threshold and cost of high one-time investment and difficult later upgrades in traditional vertical domain models.

[0019] (4) An embedded, proactive security and compliance assurance system has been constructed. This disclosure deeply integrates security and compliance controls into the entire process of data processing and business execution. By applying technologies such as privacy computing, automated compliance rule engines, and blockchain notarization, the security and privacy of customer information are ensured while fully exploring the value of data, and real-time compliance verification and tamper-proof auditing of business operations are achieved. This not only reduces the risk of data leakage and illegal operations for enterprises, but also lays a technical foundation for establishing reliable customer trust relationships.

[0020] (5) It creates new business agility and ecosystem value. This disclosure, through standardized capability encapsulation and open interfaces, enables the flexible invocation of the system's core intelligence (such as customer insights and strategy generation), thereby supporting the rapid construction of solutions or innovative service scenarios for different vertical industries. This helps enterprises shift from providing standardized products to providing personalized services and experiences based on deep understanding, explore new value growth points, and promote the construction of an ecosystem around core business capabilities.

[0021] In summary, the technical solution disclosed herein, through the synergy and linkage of various modules, has brought substantial improvements and enhancements in service intelligence, execution efficiency, system resilience, security compliance, and business innovation. Attached Figure Description

[0022] Figure 1 The present disclosure provides an architecture diagram of an AI-based digital avatar system for account managers. Figure 2 Here is a flowchart of the AI-based digital avatar method for account managers according to this disclosure; Figure 3 This is an architecture diagram of a traditional technical solution in the existing technology. Figure 4 This is a schematic diagram of the overall design of Embodiment 1; Figure 5 This is a schematic diagram of the technical architecture of this embodiment. Figure 6 This is a schematic diagram of the logical deployment in Embodiment 1; Figure 7 This is a schematic diagram of the conceptual model of Embodiment 1; Figure 8 This is a schematic diagram of the AI ​​architecture in this embodiment. Detailed Implementation

[0023] The present application / disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application / disclosure and are not intended to limit the scope of the present application / disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present application / disclosure are shown in the accompanying drawings, not the entire structure.

[0024] Example 1 Figure 1 Based on the AI-based digital avatar system architecture diagram disclosed herein, the system adopts a cloud-native, microservice-based distributed architecture design, which can be generally divided into an access layer 10, a business capability layer 20, and a data support layer 30. The following will refer to... Figure 1 This document provides a detailed description of the AI-based digital avatar system for account managers disclosed herein.

[0025] Access layer 10 provides access services to a variety of clients through standardized and scalable application programming interfaces and protocol adaptation components. These clients include, but are not limited to, mobile applications, government and enterprise office systems, WeChat mini programs, and web portals.

[0026] The access layer 10 includes a user interaction interface 110, which is used to receive service requests from clients in a multimodal manner.

[0027] In one embodiment, such as Figure 4 As shown, when customers or account managers need to handle business or inquire through the digital avatar, they can initiate interaction through any of the aforementioned clients. The interaction method supports multimodal input; for example, users can directly use natural language to ask questions (such as "Open a 5G private network for my 50 new stores"), or upload text files containing requirement information, business forms, network topology diagrams, and other image materials. Upon receiving the raw input, the system's access layer first calls the integrated natural language processing model to automatically perform speech recognition and semantic understanding on the speech or text, extracting key intent entities and operational instructions. Simultaneously, through the integrated optical character recognition and image understanding model, structured information is extracted from the uploaded documents or images. Finally, the information from different modalities is fused into a structured, machine-understandable customer service request object and passed to the business capability layer for further processing. This process realizes the transformation from unstructured, vague human expression to structured, precise business instructions, forming the foundation for subsequent intelligent processing.

[0028] The business capability layer 20 includes a dynamic customer profile building module 210, an intelligent decision engine 220, a service automation interface module 230, a memory network and continuous learning module 240, and a security and compliance assurance module 250. These modules communicate and collaborate through a service mesh and an event-driven bus.

[0029] The dynamic customer profile building module 210 is used to build and update dynamic customer profiles in real time based on multimodal interaction data and historical behavior data.

[0030] In one embodiment, the customer profile dynamic construction module 210 is an intelligent core component of the mobile BOSS customer manager avatar. It aims to dynamically construct and continuously update customer profiles by integrating multi-source data in real time, enabling customer managers to accurately identify needs, predict behaviors, and formulate personalized service strategies, thereby improving customer management efficiency and satisfaction.

[0031] In one embodiment, the customer profile dynamic construction module 210 includes a multi-source data fusion unit, a dynamic tag system management unit, a real-time update and event-driven unit, an intelligent analysis and prediction unit, a feedback optimization closed-loop unit, and a personalized view and mobile adaptation unit.

[0032] The multi-source data fusion unit establishes connections with various internal and external data sources, enabling unified access and initial cleaning of structured business data, semi-structured log data, and unstructured interactive data. This data includes, but is not limited to: structured static data such as basic customer profiles, historical product order records, and detailed consumption records obtained from the enterprise's core business operation support system; behavioral data such as click streams, search records, and conversation text generated by customers in real-time through message middleware on front-end channels such as mobile applications and mini-programs; and anonymized external data, such as industry information or publicly available commercial information, introduced through secure interfaces under compliant authorization. The key is defining a comprehensive data model capable of aligning and associating data streams of different formats, frequencies, and levels of credibility, providing a consistent data foundation for subsequent processing.

[0033] The dynamic tagging system management unit, based on integrated data, constructs and manages a multi-dimensional, evolvable customer tagging system. This system includes not only factual tags describing basic customer attributes but also model tags and predictive tags generated through calculation. Machine learning algorithms are used to automatically mine customer behavior patterns, calculating and updating dynamic tags in real time, such as "high-value customer," "potential churn risk," "interest in a certain type of product," and "service sensitive." Tag weights and statuses are dynamically adjusted based on the input of customer behavior data streams and the progress of new analysis cycles.

[0034] The real-time update and event-driven unit uses a streaming computing framework to process real-time behavioral data streams, enabling rapid feature extraction and tag updates for immediate customer interactions to ensure the timeliness of customer profiles. The entire profile update process operates in an event-driven mode: a profile calculation task is triggered when a new customer interaction occurs or a batch processing cycle completes. The updated profile snapshot is published as a structured "Customer Profile Updated" event to the system's internal event bus, thereby notifying units such as the intelligent decision engine.

[0035] The intelligent analysis and prediction unit integrates multiple machine learning algorithms to analyze the fused data. Specifically, it uses clustering algorithms to dynamically segment customer groups; classification and regression algorithms to predict potential customer needs, service preferences, or risk levels; and natural language processing models to parse customer interaction text and identify sentiment tendencies and key intentions. These analytical results are transformed into structured feature vectors, which, together with basic labels, constitute a quantitative expression of the customer profile.

[0036] The feedback optimization closed-loop unit is used to establish an optimization channel from service execution results to the customer profile model. This includes collecting actual effect data (such as customer acceptance, changes in satisfaction, and achievement of business goals) generated by the service strategies executed by the digital avatar, and comparing this data with the original customer profile and prediction results that triggered the strategy. By analyzing the deviation between prediction and reality, it drives iterative tuning of the parameters of the relevant analysis and prediction models or initiates incremental learning, thereby enabling the profile construction and prediction capabilities to continuously improve as the system operates, forming a continuous learning closed loop.

[0037] The personalized view and mobile adaptation unit transforms the constructed structured dynamic profile data into personalized visualizations for different roles (such as account managers and system analysts) and different terminals (such as PC workstations and mobile apps). This ensures that key customer insights, alerts, and recommended summaries are synchronized in a suitable format to account managers' mobile work terminals in real time, providing data support for them in field or mobile scenarios.

[0038] In this embodiment, the dynamic customer profile building module transcends static data display dashboards. Through a closed loop of "real-time data perception, intelligent trend prediction, support for precise decision-making, and continuous optimization based on feedback," it drives a fundamental shift in customer management models from traditional methods relying on personal experience to a new paradigm driven by data and intelligence. Its ultimate goal is to liberate account managers from tedious information gathering and preliminary analysis, allowing them to focus more on high-value emotional connections and complex business negotiations, thereby profoundly reshaping the management of enterprise-level customer relationships.

[0039] The intelligent decision engine 220, connected to the customer profile dynamic construction module 210, is used to generate customer management strategies (such as marketing, service, and risk management) based on the service request, the dynamically constructed customer profile, the pre-set business rules, and the artificial intelligence model. This helps account managers make data-driven scientific decisions quickly in complex situations, reduce reliance on human experience, and improve the accuracy and agility of customer management.

[0040] In one embodiment, the intelligent decision engine 220 includes a decision-making unit, a rule unit, a model prediction unit, an association unit, a strategy unit, and a feedback optimization unit, wherein... The decision-making unit, serving as the engine's entry point and coordination center, is responsible for receiving external decision-triggered requests. This unit is configured to receive and integrate information from various sources, including but not limited to obtaining the latest dynamic customer profile data from the dynamic customer profile construction module 210; and retrieving relevant historical interaction records and knowledge fragments from the memory network and continuous learning module 240. By fusing this multi-dimensional information, a complete decision-making scenario object is constructed, encompassing the current state, historical context, and real-time intent. This object is then distributed to subsequent parallel processing units, and the outputs of each unit are coordinated and summarized.

[0041] The rules unit loads and maintains a digital business rules base, where rules are defined in a form that can be parsed and executed by a computer. This unit receives decision scenario objects from the decision-making unit and performs logical reasoning and verification based on the strategies, compliance requirements, and business constraints defined in the rules base. Its output includes a compliance judgment for the current decision scenario, a set of mandatory conditions that must be met, and a filtering result for obviously inapplicable strategies, thus defining clear decision boundaries for subsequent intelligent analysis.

[0042] The model prediction unit is configured to integrate multiple trained machine learning models for performing classification, regression, ranking, or prediction tasks. This unit receives a decision scenario object from the decision unit and constraints from the rule unit. Within the constraints, the unit invokes one or more related machine learning models for parallel computation. Each model operates independently, outputting quantified predictions, such as the expected response rate of a specific marketing campaign, a risk score for processing a service request, or a probability distribution of customer preferences for different product options.

[0043] The Association Unit manages a semantic relationship network, or knowledge graph, where nodes represent customers, products, services, market entities, etc., and edges represent various business relationships between entities. This unit receives key entities identified from decision-making scenario objects and performs traversal, querying, and reasoning operations on the knowledge graph. Its function is to discover implicit, indirect relationships between entities, such as identifying potential sales opportunities based on customer social relationships, or assessing the potential cascading effects of a strategy change on related services, thereby supplementing statistical model predictions.

[0044] The strategy unit receives constraints from the rule unit, quantitative prediction results from the model prediction unit, and correlation findings from the correlation unit. This unit is configured to perform a comprehensive trade-off and global optimization calculation on various prediction indicators (such as benefits, costs, risks, and satisfaction) while satisfying all hard constraints, based on a multi-objective optimization algorithm. Ultimately, the strategy unit generates one or more candidate strategy sequences ranked by overall utility. Each strategy includes specific action suggestions, an expected quantitative effect assessment, and an interpretable chain of generation justification.

[0045] The feedback optimization unit is responsible for the self-evolution of the intelligent decision engine 220. It is configured to listen for and record business result data (such as success rate, customer feedback, and actual benefits) returned by the service automation interface module 230. This unit periodically correlates the strategy execution results with the original decision-making scenario data used to generate the strategy, forming training samples. Using these new samples, the unit initiates model training or parameter tuning processes for specific machine learning models in the model prediction unit, achieving continuous iteration and improvement of the intelligent decision engine 220's performance.

[0046] In this embodiment, the decision-making unit initiates the decision-making process. This unit distributes the constructed decision scenario object in parallel to the rule unit, model prediction unit, and association unit via synchronous calls or asynchronous event notifications. Constraints generated by the rule unit are transmitted to the model prediction unit and strategy unit in real time to limit their computational space. The model prediction unit and association unit summarize their respective analysis results and submit them to the strategy unit for final fusion and optimization. The strategy sequence output by the strategy unit serves as the final product of the intelligent decision engine 220. Simultaneously, the execution results of the strategy flow to the learning unit through an independent feedback channel, forming a closed loop from decision-making to learning. Through the division and collaborative interaction of the above functional units, the intelligent decision engine 220 achieves a combination of deterministic and probabilistic reasoning, a fusion of data-driven and knowledge-driven approaches, and a linkage between decision-making and learning.

[0047] The service automation interface module 230 is configured to receive policy instructions from the intelligent decision engine 220 and automatically drive internal functional units or external business systems to execute specific service actions through standardized interfaces and protocols, thereby achieving an automated closed loop from policy to execution.

[0048] In this embodiment, the service automation interface module 230 is the core connector of the account manager avatar, designed to connect internal functional modules with external systems to achieve cross-platform data interaction and automated service processes. Its goal is to eliminate information silos through standardized interface protocols, ensuring that customer service actions (such as work order processing, marketing outreach, and resource scheduling) are automatically triggered and processed without manual intervention, thereby improving service response efficiency and consistency.

[0049] In one embodiment, the service automation interface module 230 includes an instruction parsing unit, a communication unit, an intelligent routing allocation unit, and a service feedback unit, wherein... The instruction parsing unit is configured to receive structured policy instructions output from the intelligent decision engine module, serving as the instruction input and parsing entry point for the service automation interface module 230. This unit parses the target action, execution parameters, constraints, and priority information contained in the instructions. Based on a pre-built process template library, it deconstructs abstract policy instructions into one or more specific, executable service task sequences, and determines the dependencies and execution order between tasks.

[0050] The communication unit is used for communication connections with external heterogeneous business systems. It maintains a connector library, where each connector encapsulates all the details required for interaction with a specific external system (such as a customer relationship management system, work order system, billing system, or network device controller), including but not limited to communication protocols, authentication methods, data format conversion rules, and exception handling logic. Based on the target system of the task, this unit calls and configures the corresponding connector to establish a secure and reliable data channel.

[0051] The intelligent routing unit receives the task sequence generated by the instruction parsing unit. For tasks that can be executed in parallel, this unit performs resource evaluation and load balancing, distributing them to different execution threads or service nodes. For tasks that need to be executed sequentially or have dependencies, this unit controls their execution order and manages task status. Furthermore, the intelligent routing allocation unit can integrate simple decision-making models, such as priority dynamic adjustment algorithms based on real-time system load or task urgency, to achieve intelligent task routing.

[0052] The task execution and status monitoring unit is configured to invoke the connector prepared by the protocol adaptation unit to initiate specific business operation calls (such as creating work orders, issuing configurations, sending messages, and modifying tariffs) to the target system according to the task requirements. This unit synchronously monitors the execution process of each call, captures immediate responses, handles timeouts, and receives the final execution result (success, failure, and details). The status of all task executions is recorded in real time.

[0053] The service feedback unit collects the execution results of all tasks and summarizes and aggregates them according to the dimensions of the original policy instructions. It generates a complete policy execution report, including the overall execution status (full success, partial success, failure), details of the results of each sub-task, and business data generated during execution (such as generated work order numbers and configuration sequence numbers). The service feedback unit returns the complete execution result report to the upstream intelligent decision engine 220 or the memory network and continuous learning module 240 as feedback on the decision-making effect; on the other hand, it can send status notifications to relevant systems or personnel according to the configuration.

[0054] In this embodiment, the automated execution process begins with the instruction parsing unit. After parsing the policy instructions and serializing the tasks, this unit submits the task sequence to the intelligent routing unit. The intelligent routing unit requests and obtains the corresponding connector instance for each task from the communication unit. Subsequently, the intelligent routing unit drives the task execution and status monitoring unit to perform specific operations using the obtained connectors. The task execution and status monitoring unit reports the execution status of each task to the intelligent routing unit in real time for process control; simultaneously, it sends the final result to the service feedback unit. After all tasks are completed, the service feedback unit generates a summary report and distributes feedback. Throughout the process, each unit communicates and synchronizes its status asynchronously through an internal event bus and a shared state database, ensuring high throughput and reliability. Through the division and coordination of the above functional units, the service automation interface module 230 achieves the standardization, automated orchestration, and reliable execution of cross-system business processes, forming a key link connecting intelligent decision-making and physical world operations in the digital avatar system.

[0055] The memory network and continuous learning module 240 is communicatively connected to the customer profile dynamic construction module 210, the intelligent decision engine 220, and the service automation interface module 230. It is used to store historical customer interaction data and optimize and update the artificial intelligence model in the intelligent decision engine 220 based on the data.

[0056] In this embodiment, the memory network and continuous learning module 240 utilize brain-like memory mechanisms and adaptive learning technology to achieve long-term storage of customer interaction history, extraction of key knowledge, and dynamic optimization of the model. Its goal is to address the cognitive gap in customer interactions caused by the "forgetting effect" in traditional BOSS systems, ensuring that the customer manager's avatar can continuously improve service accuracy and scenario adaptability based on historical interactions, forming a virtuous cycle of "becoming smarter with use."

[0057] In one embodiment, the memory network and continuous learning module 240 includes a memory encoding and vectorization unit, a vectorization storage and index management unit, a memory retrieval and context reconstruction unit, a model incremental training and evolution unit, and a knowledge abstraction and pattern discovery unit, wherein... The memory encoding and vectorization unit transforms each complete customer service interaction into a semantic representation that can be efficiently stored and retrieved. Specifically, the memory encoding and vectorization unit employs an encoder model based on Transformer or MemGNN (Memory Augmented Graph Network) to model long interaction contexts containing multi-turn dialogues and operation sequences. This model effectively captures long-term dependencies and extracts semantic features such as core intent, sentiment, and key business entities from the dialogue. Finally, the memory encoding and vectorization unit outputs a high-dimensional memory vector representing the global semantics of that interaction.

[0058] The vectorized storage and indexing management unit is used to store and manage all memory vectors and their metadata using a vector database (such as Milvus, Pinecone, or other dedicated systems). This database constructs efficient indexes for high-dimensional vectors (e.g., based on IVF, HNSW algorithms), thereby supporting approximate nearest neighbor (ANN) searches on massive amounts of memory vectors and achieving millisecond-level semantic similarity retrieval. The vectorized storage and indexing management unit provides standardized data writing and query interfaces.

[0059] The memory retrieval and context reconstruction unit serves as the main functional interface for the system to invoke historical experience. When processing a new interaction, the memory retrieval and context reconstruction unit generates a query vector using the context information of the current interaction and initiates a retrieval request to the vectorized storage and index management unit to obtain the K most similar historical memory vectors. Subsequently, the memory retrieval and context reconstruction unit deduplicates, weights, and fuses these retrieval results, for example, by using an attention mechanism to filter the most relevant parts, reconstructing a structured "historical experience summary," which is then provided to the intelligent decision engine 220 as key decision input.

[0060] The incremental training unit (IBU) enables the system's continuous learning capability. It employs a continuous learning or lifelong learning framework (such as algorithms based on ACL and Lifelong-LSTM) to overcome the catastrophic forgetting problem in traditional machine learning. The IBU collects feedback data from service execution and pairs it with the original decision data to form incremental training samples. Using these samples, the IBU periodically performs incremental training or fine-tuning on upstream modules (such as the prediction model of the dynamic customer profile building module 210 and the recommendation model of the intelligent decision engine 220). To adapt to mobile or edge deployments, the IBU can apply techniques such as model quantization and network pruning to lightweight the updated model, reducing computational and storage overhead.

[0061] The Knowledge Abstraction and Pattern Discovery Unit extracts reusable high-level knowledge from massive amounts of concrete memory. Utilizing technologies such as graph neural networks, it analyzes implicit association patterns in the memory vector library, automatically discovering high-frequency successful service paths, typical customer demand evolution patterns, or potential risk combinations. This unit can abstract these patterns into new business rule hypotheses or customer segmentation strategies, structure them, and feed them back into the knowledge graph or rule base of the Intelligent Decision Engine 220, completing the transformation from tacit experience to explicit knowledge.

[0062] As an example, after an interaction, the memory encoding and vectorization unit activates, using a Transformer encoder to generate memory vectors and storing them in a Milvus-based vectorized storage and indexing management unit. When processing new requests, the memory retrieval and context reconstruction unit performs an ANN retrieval from the vector library, sending the reconstructed context summary into the decision-making process. Simultaneously, the model incremental training unit uses a continuous learning framework to process feedback data, incrementally updating and lightweighting the relevant models. The knowledge abstraction and pattern discovery unit periodically analyzes global memory, mining new knowledge through graph neural networks and injecting it into the knowledge base. All units collaborate through service interfaces and event messages, transforming fragmented interactions into structured knowledge, allowing historical experience to drive future decisions. This enables enterprises not only to achieve precise and consistent customer service but also to build a sustainable competitive advantage in a rapidly changing market. The ultimate form of this capability is to give AI customer managers a human-like "customer intuition"—serving the right people at the right time, in the right way.

[0063] The security and compliance assurance module 250 interacts with the customer profile dynamic construction module 210, the intelligent decision engine 220, the service automation interface module 230, and the memory network and continuous learning module 240 to provide security protection and compliance verification for the entire data lifecycle and business operation processes.

[0064] In this embodiment, the security and compliance assurance module 250 acts as a "protective shield" and "legal advisor" for the mobile BOSS account manager, ensuring the security of customer data throughout its entire lifecycle and that business operations comply with global privacy regulations (such as GDPR, CCPA, and PIPL) and industry regulatory requirements. Its core objective is to prevent data leakage, misuse, and compliance risks through a dual mechanism of technical protection and process control, while simultaneously building a digital foundation for customer prosperity and corporate compliance.

[0065] It is configured to implement proactive security protection and embedded compliance verification throughout the entire lifecycle of customer data (collection, transmission, storage, processing, and destruction) and all business operation processes, aiming to build a trustworthy, reliable system operating environment that complies with laws, regulations, and industry regulatory requirements.

[0066] In one embodiment, the security and compliance assurance module 250 includes a data security and privacy computing unit, a granular access and permission control unit, a compliance automated verification and policy enforcement unit, an audit tracing and tamper-proof evidence storage unit, and an emergency response and disaster recovery management unit, wherein... The secure and privacy computing unit ensures confidentiality and privacy during data processing. It implements end-to-end encryption based on the TLS / SSL protocol at the data transmission layer. Sensitive information is encrypted and stored at the data storage layer. At the data computation layer, the secure and privacy computing unit integrates privacy-enhancing computing technologies, specifically including: using a hardware-based trusted execution environment (such as Intel SGX or ARM TrustZone) to perform encrypted computations on sensitive data in memory, ensuring that the data is not leaked during use; employing a federated learning framework in scenarios requiring joint modeling with external data, enabling parties to jointly train the model without exchanging original data; and applying homomorphic encryption algorithms for specific computational tasks, allowing direct computation on encrypted data.

[0067] The granular access and permission control unit is responsible for implementing identity authentication, authorization, and behavior monitoring. It employs a multi-dimensional permission model combining role-based access control and attribute-based access control. All access requests to the system and data must be verified by the granular access and permission control unit. This unit integrates AI-driven threat detection capabilities, such as deploying user and entity behavior analysis models to establish a baseline of normal user behavior through machine learning, and detecting and issuing real-time alerts for abnormal operations deviating from the baseline (e.g., high-frequency access outside of working hours, unauthorized attempts). Simultaneously, it utilizes natural language processing models to scan internal communications or customer service conversations in real time to identify potential phishing tactics, social engineering attacks, or unauthorized internal discussions.

[0068] The Compliance Automated Verification and Policy Execution Unit is the core component that translates legal regulations and regulatory requirements into executable code. Internally, this unit maintains a structured digital compliance rule library. This library maps specific requirements from regulations such as the GDPR (General Data Protection Regulation), CCPA (California Consumer Privacy Act), and PIPL (Personal Information Protection Act) and industry regulatory provisions into verification logic and policies that can be automatically executed by the system. At critical nodes in business processes (such as collecting personal information, conducting marketing pushes, and cross-border data transfer), the Compliance Automated Verification and Policy Execution Unit is automatically triggered to conduct real-time, penetrating compliance reviews of the current operation. For example, it automatically verifies the customer's "informed consent" status, assesses whether the data minimization principle is met, or checks whether the data export has a legal basis, and automatically approves, blocks, or triggers manual review processes based on the verification results.

[0069] The audit traceability and tamper-proof evidence storage unit is responsible for recording critical business operation logs and providing non-repudiable proof. It generates structured logs with timestamps, operator identities, and operation details for all critical operations initiated by digital avatars or involving sensitive data (such as contract signing, package changes, and sensitive information queries). To ensure the integrity and credibility of the logs, this unit synchronously uploads the cryptographic hash values ​​of the operation logs to an evidence storage platform built on blockchain technology (e.g., using a Hyperledger Fabric consortium blockchain architecture). Leveraging the distributed consensus and tamper-proof characteristics of blockchain, it ensures that any subsequent modifications to the logs can be detected, thereby providing legally valid electronic evidence for regulatory audits and dispute resolution.

[0070] The Emergency Response and Disaster Recovery Management Unit is used to develop and manage emergency response plans and business continuity plans for security incidents. This unit monitors security alerts and system health status from other units in real time. Once a pre-defined major security incident (such as data breach or large-scale service interruption) is detected or a disaster recovery threshold is triggered, the unit automatically or semi-automatically activates the corresponding emergency plan, including but not limited to: automatically isolating damaged nodes, switching traffic to the disaster recovery center, initiating data recovery processes, and simultaneously notifying relevant security operations personnel.

[0071] In this embodiment, the entire security and compliance assurance is achieved through data flow and control flow. When an external request or internal task triggers an operation involving sensitive data, the granular access and permission control unit first performs identity and permission verification and continuously monitors the behavior. Subsequently, the compliance automation verification and policy enforcement unit intervenes, conducting real-time compliance review based on a digital rule base. After the review is passed, the actual operation is executed by the data security and privacy computing unit in a protected environment (such as a TEE) or through a privacy computing framework. Simultaneously, the audit tracing and tamper-proof evidence storage unit records the entire audit log of the operation and anchors its hash value to the blockchain. If any verification fails or an anomaly is detected, the control flow will jump to the emergency response and disaster recovery management unit to initiate the corresponding handling process. All units share security events and risk intelligence, forming a complete security and compliance closed loop from prevention, detection, control to response and auditing.

[0072] The data support layer 30 includes various data storage and processing facilities such as real-time data streams, business databases, knowledge graphs, and vector databases, providing unified and efficient data services for upper-layer business capabilities.

[0073] This disclosure proposes an AI-based digital avatar system for account managers, such as Figures 5 to 8As shown, this is a distributed architecture, employing a two-tier, three-layer structure (access layer - business layer - data layer), supporting horizontal scaling to handle massive requests. Data Model: Based on the "customer-user-account" three-user model, it expands upon it by adding digital avatar entity mapping. Real-time Processing: Behavioral data stream processing is implemented through Kafka, with a response latency of <200ms. Security Mechanism: Integrated digital certificate authentication and signature verification servers ensure business data is tamper-proof.

[0074] Example 2 Figure 2 To apply the AI-based digital avatar method flowchart for account managers disclosed herein to the AI-based digital avatar system for account managers described above, the following will refer to... Figure 2 This paper provides a detailed description of the AI-based digital avatar method for customer managers disclosed herein.

[0075] In step 101, customer demand information is received through multimodal interaction, and customer profiles are dynamically constructed and updated based on historical customer data and real-time interaction data.

[0076] In one embodiment, a dynamic customer profile construction process is initiated in response to customer service requests from multiple channels. First, multi-source heterogeneous data is accessed and integrated, including static customer attributes and historical records obtained from the enterprise's core business systems, as well as real-time customer interaction behavior stream data collected through message middleware. The data undergoes batch and stream processing: long-term statistical characteristics of customers are periodically updated through batch computing tasks; real-time behavioral data is processed instantly through a streaming computing engine to extract session-level dynamic features. Subsequently, the processed feature data is input in parallel into multiple machine learning models for intelligent analysis, including dynamic customer grouping using clustering models, prediction of customer needs or risks using classification or regression models, and parsing interactive text using natural language processing models to identify intent and sentiment. By synthesizing all model outputs, a structured dynamic customer profile is generated and updated, representing the customer's real-time status, preferences, and potential needs. The output of this step is continuously updated customer profile data and triggered customer status change events.

[0077] In step 102, customer management strategies are generated through hybrid reasoning by combining customer profiles, preset business rules, and AI models.

[0078] In one embodiment, a customer status change event or a new service request from step S101 is received, triggering an intelligent decision-making process. First, a decision context is constructed: the dynamic customer profile and current request content are integrated, and the memory network is queried to obtain relevant historical interaction memories, forming complete decision scenario information. Next, hybrid reasoning is performed: the decision scenario information is input into a pre-set business rule engine, performing compliance and policy verification based on deterministic logic to generate hard constraints; within the scope of the constraints, a pre-trained artificial intelligence prediction model is invoked for quantitative evaluation, and the knowledge graph is activated for relational reasoning to uncover deeper business connections. Finally, strategy fusion and optimization are performed: multi-objective optimization calculations are performed on the output of the rule engine, the prediction results of the artificial intelligence model, and the reasoning findings of the knowledge graph, balancing customer experience, business benefits, and operating costs, generating one or more priority service strategy sequences with effectiveness scores and explanations.

[0079] In step 103, the customer management strategy is automatically verified for compliance. After the compliance verification is passed, the corresponding service operations are automatically invoked and executed through a standardized interface.

[0080] In one embodiment, in response to the final service policy generated in step S102, an automated service delivery process is initiated. This step is implemented through an event-driven workflow engine: First, a predefined automated business process template is matched and instantiated according to the policy type. Then, the workflow engine decomposes complex service tasks into multiple atomic operations according to the template logic, and schedules corresponding dedicated executors sequentially or in parallel. Each executor encapsulates the protocol and logic for interacting with specific external business systems (such as resource management systems, order systems, and configuration systems), and is responsible for assembling request messages, calling target system interfaces, and parsing response results. During execution, the status of each atomic operation is monitored in real time, and the execution status is updated synchronously. Finally, when all atomic operations of the entire business process are successfully completed, a service delivery completion result event and structured feedback data are generated. Throughout this process, integrated security and compliance components perform real-time verification and audit logging of critical operations.

[0081] In step 104, the interaction process and results are stored in the memory network, and the AI ​​model and business rules are continuously learned and optimized based on the execution feedback.

[0082] In one embodiment, after the service interaction is completed, an experience accumulation and model evolution process is initiated. First, interaction memory is solidified: a memory-encoding neural network is used to semantically encode the complete context of this interaction (including request, decision, execution process, and result), generating a high-dimensional memory vector, which is stored in a vector database along with metadata to form long-term memory. Second, feedback-driven model optimization is implemented: the service execution result feedback data generated in step S103 is collected and associated with the original decision features generated in step S102 to form new training samples. Continuous learning technology is employed, using the new samples to incrementally train and fine-tune the customer profile analysis model in step S101 and / or the AI ​​prediction model in step S102 to adapt to changes in user behavior and the evolution of the business environment, thereby improving the accuracy of future decisions. Furthermore, the memory is analyzed periodically to extract high-frequency patterns and common knowledge to enrich and optimize the knowledge graph and business rules.

[0083] In one embodiment, encrypted transmission and encrypted storage are implemented at each stage of data flow; in the data processing stage, privacy-enhancing computing technology is used for sensitive computing tasks; at key nodes of business operations, a digital compliance rule base is automatically invoked for real-time verification to ensure compliance with relevant laws and regulations; and tamper-proof audit logs are generated for all key operations and solidified through trusted evidence storage technology to meet security audit and regulatory requirements.

[0084] In this embodiment, through the iterative and synergistic effects of steps S101 to S105 above, the method of the present invention achieves real-time perception, intelligent decision-making, automated execution and continuous evolution of customer needs, and constructs a complete intelligent service closed loop with self-optimization capabilities.

[0085] This disclosure provides an AI-based digital avatar system, method, and medium for customer managers. By building customer manager technology through data-driven decision-making, automated process refactoring, and intelligent interaction innovation, it achieves a full-link upgrade from customer experience to organizational effectiveness. It integrates intelligent decision-making, resilient architecture, and ecosystem fusion: Intelligent Enhancement: Improves the accuracy of customer need identification through multimodal interaction and hybrid decision-making. Architectural Resilience: Achieves 99.99% availability based on cloud-native design, completing business switching within 30 seconds in the event of a single AZ failure. Ecosystem Value: The open capability platform supports the implementation of 200+ ecosystem innovation solutions annually. This disclosure not only restructures the traditional customer manager / customer service model but also becomes a core engine for operators' digital transformation, expected to reduce customer operating costs by 40% while increasing ARPU by 15-20%. In the future, through integration with cloud computing, digital twins, brain-computer interfaces, and other technologies, it will further expand service boundaries.

[0086] In a second aspect, embodiments of this disclosure provide an electronic device including a memory and a processor, wherein the memory stores a program that runs on the processor, and the processor, when running the program, performs the steps of the AI-based digital avatar method for customer managers as described in the first aspect. In a third aspect, embodiments of this disclosure provide a computer-readable storage medium having computer instructions stored thereon, which, when executed, perform the steps of the AI-based digital avatar method for customer managers as described in the first aspect.

Claims

1. A digital avatar system for account managers based on artificial intelligence, characterized in that, include: At least one client device is used to collect and send input data containing user multimodal interaction requests; The access gateway communicates with the client device to receive and standardize the input data, forming a standard request. A processing server, communicatively connected to the access gateway, the processing server comprising: The customer profile dynamic construction module is used to receive the standard request, construct and dynamically update the customer profile corresponding to the user based on multimodal interaction data and historical behavior data, and the customer profile includes behavioral characteristics, preferences and demand prediction information; The intelligent decision engine is connected to the customer profile dynamic construction module and is used to generate service strategies for the standard request based on the customer profile, preset business rules and artificial intelligence models. The service automation interface module is connected to the intelligent decision engine and is used to call internal functional modules or external system services based on the service strategy to execute corresponding service operations. The memory network and continuous learning module are connected to the customer profile dynamic construction module, the intelligent decision engine, and the service automation interface module, respectively. They are used to store historical customer interaction data and optimize and update the artificial intelligence model and business rules in the intelligent decision engine based on the data.

2. The AI-based digital avatar system for customer managers according to claim 1, characterized in that, The dynamic customer profile building module further includes: A multi-source data fusion unit is used to parse and identify the intent of the text, voice, and image information in the standard request; The real-time update and event-driven unit is used to calculate the incoming real-time behavioral data stream and extract immediate features; The intelligent analysis and prediction unit is used to integrate the real-time features, static archive data and structured knowledge provided by the memory enhancement and continuous learning module to output an updated customer profile.

3. The AI-based digital avatar system for customer managers according to claim 1, characterized in that, The intelligent decision engine adopts a hybrid reasoning architecture, including: Rule units are used to execute predefined deterministic business rules; The model prediction unit is configured to integrate multiple trained machine learning models for performing neural network-based prediction and inference. The strategy unit is used to fuse the outputs of the rule unit and the model prediction unit to generate the service strategy.

4. The AI-based digital avatar system for customer managers according to claim 1, characterized in that, The service automation interface module adopts an event-driven architecture and includes: The instruction parsing unit is configured to receive the service policy output from the intelligent decision engine module and generate an atomic sequence of service actions based on the service policy. The intelligent routing unit is used to convert the service actions into call instructions that conform to the protocols of different internal business systems; The task execution and status monitoring unit is used to monitor the execution status of service actions and feed the execution results back to the memory network and continuous learning module.

5. The AI-based digital avatar system for customer managers according to claim 1, characterized in that, The memory enhancement and continuous learning module includes: The memory encoding and vectorization unit is used to convert key information of customer service interaction history into vectors and store them in a vector database; The memory retrieval and context reconstruction unit is used to retrieve relevant memories from the vector database according to the current request context and input them into the intelligent decision engine to assist in decision-making; The incremental training unit is used to incrementally train and optimize the artificial intelligence model based on the feedback service result data.

6. The AI-based digital avatar system for customer managers according to claim 1, characterized in that, The processing server also includes a security and compliance assurance module, which interacts with the customer profile dynamic construction module, the intelligent decision engine, the service automation interface module, and the memory network and continuous learning module to provide security protection and compliance verification for the entire data lifecycle and business operation processes.

7. The AI-based digital avatar system for customer managers according to claim 6, characterized in that, The security and compliance assurance module is integrated into the system's data flow and business flow, and is used for: Encrypt and de-identify the input and output data; During data processing, a trusted execution environment is used for privacy-preserving computation; Real-time compliance verification of business operation processes is performed, and blockchain technology is used to store key operation logs as evidence.

8. A method for creating a digital avatar for a customer manager based on artificial intelligence, applied to the artificial intelligence-based digital avatar system for customer managers as described in claims 1-7, characterized in that, Includes the following steps: The client device receives multimodal interaction requests from users, which are then standardized by the access gateway and sent to the processing server. The dynamic profile building module dynamically builds or updates the customer profile of the corresponding user based on the interaction request and associated data. The intelligent decision-making module generates a service strategy for the interaction request based on the customer profile, preset business rules, and artificial intelligence model. The automated service execution module automatically calls internal system resources and executes corresponding service operations according to the service policy. Through the memory enhancement and continuous learning modules, the interaction and execution results are stored as historical memory, and the artificial intelligence model and business rules are continuously learned and optimized based on the execution feedback.

9. The method for creating a digital avatar for a customer manager based on artificial intelligence according to claim 8, characterized in that, Before performing the corresponding service operation, the following is also included: A service plan is generated based on the aforementioned customer management decision instructions; Perform automated compliance checks on the service plan; Specifically, performing the corresponding service operation involves executing the service plan that has passed the compliance verification.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, The computer instructions, when executed, perform an AI-based digital avatar method for customer managers as described in any one of claims 8-9.

Images