Case study 02 · Enterprise search · AI
Rebuilding enterprise search around what people actually type
Mined 250K+ real queries and interviewed employees to turn a fragmented search landscape into an AI-forward experience with sourced answers, smart filters and an expert finder — launched in 2026.
- Role
- Co-researcher & UX Architect
- Timeline
- 2025 – 2026
- Platforms
- Web · Enterprise intranet
- Partners
- Core UX, search platform owners, IT, Sales Engineering, content owners
Open a return request in the portal, attach the serial number, and ship within 30 days of approval[1]. EU orders route through the regional hub[2].
- 250K+
- queries analyzed across six months of logs
- 10 · 7 · 115
- repositories, search interfaces and data sources mapped
- 40%
- of queries already written in natural language
Context
Employees were hunting across 10 repositories and 7 different search interfaces. Everyone agreed search was frustrating — nobody could say exactly why, or what to fix first.
The question
“Where precisely does search break down, and what would a single, trustworthy search experience need to do?”
Approach
- 01
Let the data talk first
Analyzed six months of query logs (250K+ queries) with AI-assisted mining to cluster intents, dead ends and repeat searches.
- 02
Map the mess
Catalogued 10 repositories, 7 search interfaces and 115 underlying data sources to expose duplication, gaps and ownership.
- 03
Hear the why
Recruited a 41-person cohort across sales engineering, support and IT to connect query behavior to real motivations and workarounds.
- 04
Turn findings into a build plan
Translated insights into prioritized capabilities and an AI-forward direction the platform team could ship against.
Key insight
Four in ten queries were already full questions. People were asking — and a keyword engine was answering with a list of links.
Design decisions
Answers with receipts
Every AI answer links to highlighted source quotes so people can verify before they trust.
Fewer, better results
Prioritize relevance and de-duplication over adding more filters to a noisy list.
Find people, not just pages
Expert Finder surfaces who knows the answer when the document doesn't exist yet.
Outcome
The research shaped Search+, launched in 2026 with improved relevance, fewer duplicates, AI answers tied to source quotes, smart filters, previews, summaries, Ask AI and Expert Finder.