Knowledge base

Customer-service AI: when should it help, and when should it hand over?

Customer-service AI is useful when it resolves recurring, source-backed questions quickly and routes uncertain, sensitive or decision-heavy cases to the right person.

Which questions can be automated?

Product information, opening hours, process status, required documents and policy-backed questions are practical starting points. Answers should use an approved knowledge base, not unchecked general model output.

When is human handover mandatory?

Handover when no adequate source exists, sources conflict, a personal or financial decision is required, a complaint is raised or the customer asks for a person. Pass a concise conversation summary and only the necessary collected data.

What makes operation safe?

Use permissions, logs, data minimisation, restricted topics, visible sources and a representative test set. The NIST Generative AI Profile emphasises governing, measuring and continuously managing risk.

What should be measured?

Track self-service resolution, handover rate, incorrect answers, first-response time, satisfaction and recurring questions missing from the knowledge base. Explore customer-service AI, service automation and enterprise RAG.

Source and further data: https://www.nist.gov/itl/ai-risk-management-framework/ai-risk-management-framework-resources