A customer inquiry is complex when it consists of multiple parts, lacks information, requires system data, or results in a decision. An AI chatbot solves the first three cases if it can ask follow-up questions, conduct dialogues, and access systems. For decisions requiring discretion, in cases of dissatisfaction, or when dealing with knowledge not found in any source, it hands over to a human—along with the previous conversation history. This article breaks down complex inquiries into their core components and outlines rules for handover.
At a glance
Complexity typically stems from six causes: multiple sub-questions, missing details, system data, multi-step processes, discretion, and emotion.
A well-designed chatbot solves the first four on its own, using follow-up questions, guided dialogues, and integrations.
Discretion and emotion belong to humans.
If knowledge is missing, the chatbot acknowledges this and hands over.
An effective handover includes the conversation history and captured details so nobody has to start from scratch.
What makes an inquiry complex
An example from technical service: “I need a spare part for the dosing unit of our system—how long will delivery take, and is this still covered by warranty?” This single sentence contains three questions, one missing detail, and two system queries. That is precisely what makes it complex.

Multiple sub-questions. The customer asks about the part, delivery time, and warranty in a single sentence. A chatbot must recognize these distinct elements and address each one individually.
Missing information. Which system, which model series, which year of manufacture? Without these details, an accurate answer is impossible. The chatbot asks targeted follow-up questions.
System data. Delivery times and warranty status reside in the ERP or CRM. Without an integration, the chatbot cannot know them.
Multi-step processes. A complaint requires an order number, an error description, and contact details—collected and validated in a specific order. This requires a guided dialogue.
Discretion. Goodwill gestures, special terms, or exceptions to the rule: this involves a decision that the business intentionally entrusts to humans.
Emotion. An upset customer calling for the third time about the same issue needs someone who listens and takes action.
What an AI chatbot solves independently
Recognizing and answering sub-questions individually
A chatbot that only detects the single most likely intent will answer just one out of three questions. A better approach is an intent recognition system that selects and combines information paths: a knowledge-based answer for warranty conditions, a system query for delivery times, and a product selection flow with follow-up questions for the spare part.
Asking targeted follow-up questions
When information is missing, a follow-up question is the right response. The chatbot asks for the model series or serial number and proceeds seamlessly—without requiring the customer to repeat their question. This also applies to ambiguous phrasing, such as when a customer describes a symptom rather than naming a spare part.
Accessing backend systems
With integrations into ERP, CRM, or ticketing systems, the chatbot retrieves delivery times, order statuses, or warranty information during the live conversation. Without this connectivity, routing the request is the only option. Why this is the actual bottleneck for many LLM chatbots is explained in the article Why LLM-based chatbots cannot resolve service cases without integration.
Guiding multi-step processes
Structured editorial dialogues are ideal for processes with fixed steps: they collect the required details in the correct sequence, validate them, and respond to invalid entries with clarification prompts. Generative responses are used where open formulation is appropriate, while structured editorial steps are deployed where precision and compliance are required.
When the chatbot should hand over
Four rules cover most scenarios.
Knowledge is missing. If the answer cannot be found in an approved source, the chatbot states this transparently and hands over. A plausible yet fabricated answer is costlier in customer service than any escalation.
Discretion is required. Goodwill decisions, discounts, and policy exceptions must be decided by a human.
The customer is dissatisfied or asks for a human. The chatbot hands over immediately, without further automated attempts.
The process stalls. If a detail remains unclear after multiple clarification attempts, an escalation is faster than another loop.
For an overview of topics a chatbot should avoid answering from the outset, see the article What a chatbot should not answer.
What an effective handover looks like
A handover is only as good as the context it provides. The human agent taking over must see the conversation history, previously captured data such as system type, order number, or error description, and the reason for the transfer. This allows the conversation to resume seamlessly without repeating questions.
Routing is equally crucial: spare parts inquiries belong in technical support, invoice questions in accounting, and English-language inquiries with an English-speaking agent. Once resolved, the conversation can return to the chatbot—for example, to issue a confirmation or initiate the next step.
How Mercury.ai handles complex inquiries
At Mercury.ai, an orchestration of models identifies the user's intent and selects the optimal path: a knowledge-based answer with clarification, a product selection flow, or a structured editorial dialogue. If information is missing or facts are ambiguous, the chatbot asks targeted questions or routes into a structured dialogue. Via integrations, it reads from and writes to CRM, ERP, and ticketing systems during the live interaction. Hallucinations are significantly minimized because answers stem strictly from approved sources and are validated prior to output.
When human intervention is needed, the chatbot transfers the session within the Agent Desk complete with the full conversation history and pre-qualified data, routed by department, language, or agent capacity. Following the live chat, the conversation can return to a designated point within the chatbot. Learn more about the underlying architecture on the Mercury Intelligence page. To discover which recurring tasks can be fully automated, read our article on 15 use cases in customer service. The guide Automating customer service with AI details the five-step roadmap.
Conclusion
Complex inquiries are typically compound inquiries. Multiple sub-questions, missing information, system data, and multi-step processes can be resolved by a chatbot capable of asking follow-up questions, guiding dialogues, and connecting to backend systems. Discretion, emotion, and missing knowledge belong in human hands—and an effective handover is part of an optimal solution. Organizations that clearly separate the two deploy chatbots that customers trust, even with the tough questions.
Want to find out which complex inquiries your chatbot can handle? Talk to us.
About the Author
Dr. Maximilian Panzner
Chief Technology Officer @Mercury.ai
Dr. Maximilian Panzner is CTO and co-founder of Mercury.ai. He earned his doctorate in computer science at the CITEC institute of Bielefeld University, where he conducted research on multimodal machine learning and intelligent interaction systems. For over 20 years, he has been working on artificial intelligence, human-machine interaction, and conversational AI platforms for enterprise use.






