Billing, meter-reading and moving tasks required menu knowledge.
Everyday requests did not match service categories.
Actions varied by household, contract, date and enrollment.
The client needed in-app help for routine service requests.
Objective
Let customers reach services using everyday language.
Route each context to lookup, requests or human support.
My role
Co-defined scenarios and scope; directly built the ontology.
Configured and tested dialogs; coordinated service integration.
02 / SYSTEM DESIGN
How it connects.
Processing / data flowIteration / feedbackService / environment boundary
01
Intent is only the beginning.
The same request can require a different action depending on household, contract and application state. Connect recognized intent to context-sensitive service paths.
02
Test the whole conversation.
Build the ontology directly, then test consecutive turns, response cards and service integrations. Use operating logs to distinguish missing intent coverage from configuration errors.
03 / RESULTS & SCOPE
The outcome.
595,658Conversation records, including opening messages
349,087All conversation sessions
96%Utterance recognition
Records: May 2, 2024–April 23, 2026, including opening messages. Recognition is from a separate Q2 2024 report.
