An ai customer service side channel manual clarification method and system for maintaining terminal session
By introducing a side channel into the AI customer service system, the back-end clarifier provides business facts and feeds back the results, solving the problems of session interruption and information mismatch in the e-commerce intelligent customer service system, and improving user experience and efficiency.
Patent Information
- Application Number
- CN202611318306.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-06
AI Technical Summary
Existing e-commerce intelligent customer service systems suffer from problems such as session interruption, information mismatch, and auditing difficulties when handling complex rules, resulting in poor user experience and low efficiency.
By establishing a side channel between the AI customer service representative and the back-end clarifyer, the right to host the conversation and the right to respond to clarification are separated. The back-end clarifyer provides business facts and feeds the results back to the AI to continue hosting the conversation, thus maintaining the continuity and consistency of the end-user conversation.
It achieves uninterrupted user experience, more efficient manual processes, improved accuracy and auditability, reduces the risk of internal information leakage, and improves the efficiency and accuracy of sessions.
Smart Images
Figure CN123055831A_ABST
Abstract
Description
Technical Field
[0001] This invention mainly relates to the fields of artificial intelligence customer service and natural language processing technology, specifically to an AI customer service side-channel manual clarification method and system for maintaining terminal conversations, belonging to machine question answering and human-machine collaborative conversation control technology. Background Technology
[0002] E-commerce intelligent customer service needs to handle complex rules such as after-sales policies, promotional exceptions, inventory, and fulfillment. AI customer service agents based on large language models are machine question-and-answer systems. They can access business systems through tools. After semantic processing of user questions, if the knowledge base and tools cannot provide a definite conclusion, human intervention is required.
[0003] Existing products typically choose from the following modes: when the confidence level is below a threshold, a human agent is triggered to take over the conversation, i.e., the warm transfer mode; AI customer service provides clarification tools to ask questions to the current end user, i.e., the human-machine loop mode; an approval mode that requires human approval or rejection before high-risk operations; and a mode where agents communicate privately in instant messaging groups or ticket systems, and then manually edit responses for external use, which is disconnected from the AI toolchain.
[0004] The above-mentioned models have the following shortcomings in practical applications: In the whole-transfer mode, the manual time is long, users need to change the contact person, and the AI context is difficult to reuse; In the clarification mode for end users, users do not have information such as policy exceptions and warehousing rules, and they cannot get the correct answer when asking questions; In the approval-type human-machine loop mode, the semantics are mismatched, and the business scenario requires business facts or rulings, rather than simple agreement to execute; In the bypass communication mode, internal responses cannot be automatically entered into the model context, auditing is difficult, and inconsistencies in wording are easy to occur. Summary of the Invention
[0005] 1. The technical problem that the invention aims to solve: This invention provides an AI customer service side-channel manual clarification method and system for maintaining terminal conversations. It separates the front-end conversation hosting rights from the back-end clarification and response rights. Without interrupting the terminal user's conversation or transferring the front-end control, the back-end human provides clarification of business facts, and the clarification results are fed back into the tool results and the AI continues to host the conversation.
[0006] 2. Technical Solution: To achieve the above objectives, the technical solution provided by the present invention is as follows: A method for AI-powered customer service side-channel manual clarification to maintain terminal sessions includes the following steps: S1. During the front-end conversation with the end user, the AI customer service receives the end user's natural language query and performs semantic processing. When it is determined that a back-end business fact or policy decision is required, it calls the side-channel clarification tool to initiate a clarification request to a back-end clarifier who is different from the end user. S2. Switch the foreground session to a waiting clarification state, display a waiting prompt to the end user, and do not display the internal content of the clarification request to the end user; S3. After the person clarifying in the background submits a clarification response, the clarification response is encapsulated as a tool return result and written into the inference context of the AI customer service. S4. The AI customer service representative, based on the results returned by the tool, resumes reasoning, continues to host the front-end conversation, and outputs a response to the end user. The person clarifying in the background does not gain control of the front-end session by submitting the clarification response.
[0007] Furthermore, the front-end session is transmitted between the end user and the AI customer service representative through a front-end session channel, and the clarification request is transmitted between the AI customer service representative and the back-end clarifier through an internal clarification channel independent of the front-end session channel. The session identifier of the front-end session remains unchanged before and after the waiting clarification state.
[0008] Furthermore, the invocation parameters of the side-channel clarification tool include clarification topic, internal question, context summary, entity information, candidate options, priority, and timeout.
[0009] Furthermore, in step S2, during the waiting clarification state, new messages sent by the end user during the waiting period are buffered and processed together after the AI customer service resumes reasoning; and / or the AI customer service is prohibited from calling high-risk write-type tools during the waiting clarification state.
[0010] Furthermore, step S2 also includes routing the clarification work order to the background workbench of the corresponding skill group based on the clarification topic, skill group, or background clarifier load of the clarification request, supporting sticky routing to dispatch clarification requests from the same end user or the same topic to the same background clarifier.
[0011] Furthermore, the clarification response is a structured response, including response type, selected options, remarks, and an indicator indicating whether AI is allowed to inform the user. The response type includes option selection, free text, and rejection response.
[0012] Furthermore, in step S4, the AI customer service rewrites the original content of the clarification response before outputting it to the end user, and the original content of the clarification response is not visible to the end user.
[0013] Furthermore, if the clarification request does not receive a clarification response within a preset timeout period, it enters a timeout state, and the AI customer service executes a downgrade strategy, including outputting a conservative response or prompting the end user to try again later; in the timeout state, it can optionally be upgraded to transfer the entire session to a human agent, transferring control of the front-end session to a real agent.
[0014] Furthermore, both the clarification request and the clarification response are recorded in the tool-level audit log, which is stored in association with the session identifier of the foreground session.
[0015] This invention also provides an AI-powered customer service side-channel manual clarification system for maintaining terminal sessions, comprising: The front-end conversation module is used to maintain the front-end conversation between the end user and the AI customer service, and to display a waiting prompt to the end user when waiting for clarification. The AI inference and tool orchestration module is used to invoke the side-channel clarification tool when it is determined that a back-end business fact or policy decision is required, and to resume inference and continue to host the front-end session after receiving the tool's return result. The side-channel clarification tool module is used to initiate a clarification request to a back-end clarifier who is different from the terminal user, and encapsulate the clarification response submitted by the back-end clarifier into the tool return result and write it into the inference context of the AI inference and tool orchestration module. The visibility isolation module is used to isolate the internal content of the clarification request and the original content of the clarification response from the foreground session, and not display them to the end user; The routing and queue module is used to route clarification work orders to the backend workbench; The person clarifying in the background submits the clarification response through the background workbench, but does not obtain session control of the front-end session module.
[0016] 3. Beneficial effects: Compared with the prior art, the technical solution provided by this invention has the following advantages: Uninterrupted experience: Users do not need to change their contact person, and the front-end session identifier remains stable with the host; Human intervention is more efficient: the back-end clarifier only answers one structured question, and the average time spent is less than the entire transfer. Improved accuracy: Policy or exception-related questions are decided by authorized human judges, and then AI organizes the wording; Auditable: Clarification requests and responses are logged at the tool level, facilitating quality control and accountability. Internal and external isolation: The internal question stem and original answer are not directly output to reduce the risk of improper disclosure.
[0017] It should be noted that the structures not described in this invention are not related to the design points and improvement directions of this invention, and are the same as or can be implemented using existing technologies, and will not be elaborated here. Attached Figure Description
[0018] Figure 1 This is a schematic diagram comparing the existing whole-segment switching mode with the side-channel clarification mode of the present invention; Figure 2 This is a schematic diagram of the session state machine of the present invention; Figure 3 This is a timing diagram for side-channel clarification in this invention; Figure 4 This is a system module structure diagram of the present invention; Figure 5 This is a flowchart illustrating the processing of an e-commerce embodiment of the present invention. Detailed Implementation
[0019] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate several embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.
[0020] This embodiment constructs a three-party system: end-user U, AI customer service A, and back-end clarifier H. AI customer service A is a machine question-answering agent based on a large language model, engaging in dialogue with end-user U through natural language processing. The front-end conversation channel transmits messages between end-user U and AI customer service A, while the internal clarification channel transmits clarification requests and responses between AI customer service A and back-end clarifier H; these two channels are separate. The conversation identifier for the front-end conversation remains constant throughout the process, unchanged before and after the clarification waiting state.
[0021] like Figure 1 As shown, the left side represents the existing seamless transition mode: after the user's conversation with the AI reaches an uncertain point, the system hands over control to a human agent. The user then primarily interacts with the human agent, the AI exits the moderation, and the human agent takes over the remaining conversation. The right side represents the side-channel clarification mode of this invention: the user maintains the same conversation with the AI throughout. The AI initiates side-channel verification with another human agent in the background. The user only sees a waiting state. After the human agent replies with the tool's result, the AI continues to moderate and respond to external requests. The human agent does not take over the front end due to the response.
[0022] like Figure 2 As shown, the session state machine of the present invention includes the following states: The ACTIVE_AI status indicates that the AI customer service is normally managing the front-end conversation. The WAITING_CLARIFICATION status indicates that the AI customer service has invoked the side-channel clarification tool, and the foreground session is in a waiting-for-clarification state. The CLARIFICATION_TIMEOUT status indicates that the clarification request has timed out. The HANDOFF state indicates that the entire process will be switched to manual control.
[0023] The state transition is as follows: After the session begins, it enters the ACTIVE_AI state; When the AI customer service calls the side-channel clarification tool, it transitions from the ACTIVE_AI state to the WAITING_CLARIFICATION state. Once the person clarifying in the background submits a response and the tool feeds back the results, the WAITING_CLARIFICATION status returns to the ACTIVE_AI status. When the wait times out, the state transitions from WAITING_CLARIFICATION to CLARIFICATION_TIMEOUT, and after executing the degradation strategy, it returns to ACTIVE_AI state, or optionally upgrades to the entire process to be manually switched to HANDOFF state. When a user explicitly requests to switch to human assistance, the system transitions from the ACTIVE_AI state to the HANDOFF state. This path is not the main path of this invention.
[0024] Both the HANDOFF state and the session end state are final states.
[0025] like Figure 3 As shown, the side-channel clarification timing of the present invention is as follows: End user U sends a natural language query to the front-end conversation module, such as whether customized products can be returned without reason within seven days; The front-end conversation module forwards user-visible messages to AI customer service A; AI Customer Service A performs semantic processing and reasoning on user questions, but its judgment tools and knowledge base are insufficient, requiring backend adjudication. AI customer service representative A invokes the side-channel clarification tool request_backoffice_clarification; The visibility isolation module only displays a waiting prompt to the end user U, and does not show internal problems; AI customer service representative A creates a clarification ticket, which includes the internal question, a contextual summary, and candidate options. The clarification work order is dispatched to the skill group workbench via the routing and queue module; During this period, AI customer service representative A retains control of the front-end conversation, while back-end clarifier H does not take over the chat window; The person clarifying the situation, H, submitted a structured response. The system will encapsulate the response as a tool result and feed it back into the context of AI customer service A; AI customer service A resumes the reasoning and rewrites the external script to generate a response to the user; The foreground session module outputs a user-visible reply to the end user U.
[0026] like Figure 4 As shown, the system module structure of this invention includes: an end user, a front-end session module, a visibility isolation module, an AI inference and tool orchestration module, a side-channel clarification tool module, a state and persistence module, a routing and queue module, a back-end workbench, a back-end clarifier, and an audit log. The end user interacts with the AI inference and tool orchestration module through the front-end session module; the visibility isolation module connects the front-end session module and the AI inference and tool orchestration module to achieve internal and external message isolation; the AI inference and tool orchestration module connects to the side-channel clarification tool module and the audit log; the side-channel clarification tool module connects to the state and persistence module and the routing and queue module; the routing and queue module connects to the back-end workbench, through which the back-end clarifier receives and responds to clarification work orders.
[0027] like Figure 5 As shown, taking e-commerce after-sales policy consultation as an example, the processing flow in this embodiment is as follows: Users inquire about after-sales policies using natural language queries; AI customer service calls upon the knowledge base and order tools to make a judgment; if it can determine the answer, it will directly reply to the user. If it cannot be determined, then invoke the side-channel clarification tool; The user is shown a waiting prompt, and a work order is created and added to the queue internally. Back-end experts select options or fill in remarks on the workbench; The response was fed back in the form of tool results; AI-generated customer service replies to users after being rewritten; The session was still hosted by AI.
[0028] The side-channel clarification tool's invocation parameters include the clarification topic, internal question, context summary, entities, options, priority, and timeout. The clarification response is a structured response, including the response type (answer_type), selected option, note, and an option to allow AI to inform the user (allow_ai_inform_user). Response types include option selection, free text, and reject. An optional forced handoff flag (force_handoff) may also be included.
[0029] While awaiting clarification, the system can buffer new messages sent by end users during the waiting period and process them together after the AI customer service resumes reasoning; it can also prevent the AI customer service from calling high-risk write-type tools while awaiting clarification, to avoid performing irreversible operations during the period of unresolved clarification.
[0030] The routing and queuing module dispatches requests based on the clarification topic, skill group, or backend clarifier load. It supports timeout, reassignment, and sticky routing. Sticky routing can dispatch clarification requests from the same end user or on the same topic to the same backend clarifier, improving response consistency.
[0031] The visibility isolation module divides messages into user-visible channels and internal channels. Internal clarification questions and original backend responses are not displayed to end users by default. After receiving a clarification response, the AI customer service rewrites the original content into user-oriented wording before outputting it, ensuring that internal information is not improperly disclosed.
[0032] The audit log module records every clarification request and response. The logs are stored in association with the session identifier of the front-end session, which facilitates subsequent quality inspection and accountability.
[0033] In summary, this invention achieves uninterrupted manual clarification of terminal sessions through three core mechanisms: separating the front-end conversation channel from the internal clarification channel, ensuring that the back-end clarifier does not gain control of the front-end, and allowing the AI to continue hosting the clarification response after the tool results are fed back. This overcomes the shortcomings of existing models such as full-segment transfer, asking the user questions, and approval-type human-machine loops.
[0034] Extended solutions: This invention can be implemented using two methods: synchronous blocking wait and asynchronous pending_id wake-up. During the waiting period, the system can buffer consecutive messages sent by users or restrict high-risk write operations; If the timeout occurs or the user forces a switch to human intervention, the process can be downgraded to a complete switch to human intervention as a subordinate implementation. This invention can also be applied to non-e-commerce scenarios, such as when a front-end user is having a conversation with an AI assistant, the side channel confirms the rule interpretation with the back-end rule administrator before the AI continues to respond.
[0035] The above-described embodiments are merely illustrative of certain implementations of the present invention, and are described in a relatively specific and detailed manner. However, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements are all within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.
Claims
1. A method for artificial clarification via AI-powered customer service side-channel while maintaining terminal conversation, characterized in that, Includes the following steps: S1. When the AI customer service determines that a back-end business fact or strategy decision is required during a front-end conversation with the end user, it calls the side-channel clarification tool to initiate a clarification request to a back-end clarifier who is different from the end user. S2. Switch the foreground session to a waiting clarification state, display a waiting prompt to the end user, and do not display the internal content of the clarification request to the end user; S3. After the person clarifying in the background submits a clarification response, the clarification response is encapsulated as a tool return result and written into the inference context of the AI customer service. S4. The AI customer service representative, based on the results returned by the tool, resumes reasoning, continues to host the front-end conversation, and outputs a response to the end user. The person clarifying in the background does not gain control of the front-end session by submitting the clarification response.
2. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: The front-end session is transmitted between the end user and the AI customer service through the front-end session channel. The clarification request is transmitted between the AI customer service and the back-end clarifier through an internal clarification channel independent of the front-end session channel. The session identifier of the front-end session remains unchanged before and after the waiting clarification state.
3. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: The parameters for calling the side-channel clarification tool include the clarification topic, internal question, context summary, entity information, candidate options, priority, and timeout.
4. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: In step S2, during the waiting clarification state, new messages sent by the terminal user during the waiting period are buffered and processed together after the AI customer service resumes reasoning; and / or the AI customer service is prohibited from calling high-risk write-type tools during the waiting clarification state.
5. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: Step S2 further includes routing the clarification work order to the corresponding skill group's backend workbench based on the clarification topic, skill group, or backend clarifier load of the clarification request, supporting sticky routing to dispatch clarification requests from the same end user or on the same topic to the same backend clarifier.
6. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: The clarification response is a structured response, including response type, selected options, remarks, and an indicator indicating whether AI is allowed to inform the user. The response type includes option selection, free text, and rejection response.
7. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: In step S4, the AI customer service rewrites the original content of the clarification response and then outputs it to the end user. The original content of the clarification response is not visible to the end user.
8. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: If the clarification request does not receive a clarification response within a preset timeout period, it enters a timeout state. The AI customer service executes a downgrade strategy, including outputting a conservative response or prompting the end user to try again later. In the timeout state, it can optionally be upgraded to transfer the entire session to a human agent, transferring control of the front-end session to a real agent.
9. The AI-powered customer service side-channel manual clarification method for maintaining terminal sessions according to claim 1, characterized in that: Both the clarification request and the clarification response are recorded in the tool-level audit log, which is stored in association with the session identifier of the foreground session.
10. An AI-powered customer service side-channel manual clarification system for maintaining terminal sessions, characterized in that, include: The front-end conversation module is used to maintain the front-end conversation between the end user and the AI customer service, and to display a waiting prompt to the end user when waiting for clarification. The AI inference and tool orchestration module is used to invoke the side-channel clarification tool when it is determined that a back-end business fact or policy decision is required, and to resume inference and continue to host the front-end session after receiving the tool's return result. The side-channel clarification tool module is used to initiate a clarification request to a back-end clarifier who is different from the terminal user, and encapsulate the clarification response submitted by the back-end clarifier into the tool return result and write it into the inference context of the AI inference and tool orchestration module. The visibility isolation module is used to isolate the internal content of the clarification request and the original content of the clarification response from the foreground session, and not display them to the end user; The routing and queue module is used to route clarification work orders to the backend workbench; The person clarifying in the background submits the clarification response through the background workbench, but does not obtain session control of the front-end session module.