Telco startup replaces static FAQs with LLM-powered contextual help

AI · 6 min read

Support logs showed users repeatedly consulting FAQs that were generic and poorly linked to in-app states. The product team decided to embed an LLM that would surface concise, context-aware guidance based on the current screen, user data, and recent actions rather than a static help center.

Design decisions focused on transparency and control: the assistant suggests steps and highlights where it is making an inference, offers one-click actions where safe (resend OTP, refresh SIM provisioning), and provides a “why this suggestion” toggle that explains the signal used. Rate limits and fallback to human support were incorporated to handle regulatory and safety needs.

Within eight weeks, call center volume for common issues dropped by 30% and average in-app help uptimes improved. Designers reported an unexpected benefit: the LLM surfaced edge-case friction patterns that product analytics hadn’t flagged, prompting several product fixes and copy rewrites.

The case emphasizes that embedding generative AI is as much a UX problem as a systems one—product teams must design affordances for self-service, clear fallback paths, and explainability so users trust contextual suggestions.