Smarter, structured knowledge powering One NZ’s next generation of internal AI copilots.
Problem
Broadband is one of One NZ’s most complex product areas, with multiple access technologies — including Fibre, ADSL, VDSL, Wireless, and HFC — each requiring different diagnostic flows and backend systems. The internal knowledge base, hosted in BrightSpot CMS, had become fragmented and difficult to navigate, with outdated or duplicate content scattered across hundreds of documents.
Support agents often struggled to locate the correct information quickly, especially during high-pressure interactions. A previous AI Concierge proof-of-concept surfaced concerns around data quality, prompting the team to prioritise a comprehensive review of broadband-related knowledge as the foundation for future AI-driven support.
Solution
We partnered with One NZ to deliver a proof-of-concept focused on cleaning, consolidating, and structuring the broadband knowledge base to make it AI-ready. Nearly 1,000 support documents were audited, de-duplicated, and restructured using a modular schema that supports retrieval via natural language queries.
Each document was tagged with structured metadata — including audience type, journey stage, access method, and device compatibility — and rewritten into a consistent format designed to align with how agents think and work. A rewrite interface using LLMs was also introduced to support future scalability.
In parallel, the same structuring logic was applied to call centre transcript data, surfacing discrepancies between documented knowledge and real-world practices. A benchmarking framework was delivered to enable evaluation of Copilot iterations going forward.
To support ongoing development of AI copilots, a benchmarking framework was also delivered. This tool uses Amazon Bedrock LLM as a judge to programmatically score responses across a standardised question bank. It gives internal teams a consistent, repeatable way to validate agent performance, track improvements, and iterate on new versions — enabling faster development cycles and clearer business confidence in Copilot outputs.
Results
One NZ now has a structured, AI-compatible knowledge corpus suitable for integration with Microsoft Copilot, Writer.com, and internal support tools. The refined knowledge base enables more accurate and faster retrieval, improves consistency in responses, and lays the foundation for intelligent triage and future AI enhancements.
The cleaned corpus, combined with structured transcript data and performance benchmarking tools, positions One NZ to roll out AI copilots with confidence, backed by high-quality, curated knowledge.
Technologies
