Knowledge Cards

Intuitive & Dynamic Insight Panels for Enterprise Search

Overview

In the age of generative AI and enterprise search, users expect answers — not just links. Workplace AI aimed to deliver that value by surfacing knowledge cards: compact, contextual snapshots of key information tied to entities like people, projects, assets, and organisations. These cards would enable knowledge workers to quickly understand an entity without opening multiple search results — essentially helping them operate faster, with less cognitive load.

Knowledge cards are ubiquitous in public search products like Google and Bing, where they reduce friction and support quick understanding. But translating that UX into a corporate knowledge context — where data is complex, heterogeneous, and highly domain-specific — posed foundational questions about user needs, information structure, and trust.

My Role: UX research and UX/UI design lead
Team & Stakeholders: CEO, CTO, Senior Engineers, Cross-functional Product Team

Duration: 6 sprints

Product: Workplace AI (SaaS / enterprise search platform)

Workstream: Feature validation, ideation, and prototype evaluation

Context & Strategic Hypothesis

Enterprise knowledge work is messy: users routinely switch between disparate systems, repeated searches, and fragmented documentation just to get context on a person, project, or topic.

Hypothesis:
A knowledge card feature within the Workplace AI search experience would:

Deliver immediate informational context through intelligent summaries.
Increase user engagement by surfacing relevant entities tied to user intent.
Improve information discovery, connecting users to deeper insights and relationships

Goal

Our goal was not to ship polished UI first — it was to prove or disprove this hypothesis with evidence grounded in user feedback and early design exploration.

Research & Discovery

Rather than jumping into design, we devoted nearly half of our project timeline to discovery — a deliberate choice shaped by the ambiguity of user needs in enterprise search contexts.

Heuristic Evaluation

I started with a heuristic evaluation of how major search providers implement knowledge panels. I dissected:

Information hierarchy and card structures
Data categorisation (people, organisations, topics)
Interaction patterns that support quick comprehension

This gave us a baseline for structural thinking and informed what types of information users might expect to see on a knowledge card.

There were three main data categories we were looking to cover; organisational, personal and topical. Here I looked at each search engine's application for these data sets, what types of information were shown, when they were shown and how they were presented.

User Interviews

We conducted one-to-one interviews with 8 users across 6 departments at our PoC customer (Aiimi). These were structured around real information tasks — e.g., “find everything about client X, including recent engagements, contacts, and relevant documents.”

Key insights included:

Users did want quick summaries, but only when they were trustworthy and actionable
Users wanted entity relationships, not just facts — e.g., projects related to people or organisations
Different user roles valued different context types — team members versus executives versus analysts

This phase reframed our initial thinking: Knowledge cards are only useful if they reflect the right context for specific user tasks.

Design Exploration & Prototyping

Armed with research, we moved into proof-of-concept prototypes rather than a production build.

Turning Insights into Structure

I created an initial information architecture for cards based on the categories that emerged in interviews and heuristic analysis. This included:

Entity definitions
Key relationships (linked people or projects)
Contextual metadata and summaries

Instead of low-fidelity wireframes, we chose to create high-fidelity interactive prototypes with real production data where possible. This helped our users and stakeholders evaluate the experience, not just static visuals.

Rapid Iteration

We iterated the UI designs and prototypes with:

Frequent internal reviews (3× weekly)
Broader “Next Generation Group” demos (every 2 weeks)
User validations with task-based usability sessions

Responsive & Behaviour Design

Users needed to trust the knowledge being surfaced, and that trust is shaped by both design and interaction.

We tested responsive behaviours that adapted the card for smaller screens — transforming panels into horizontally scrollable columns, for example — ensuring the feature didn’t break the search experience on tablets or narrower layouts like mobile.

User Testing & Learnings

Usability testing revealed that users appreciated:

Glanceable contextual insights
Strong visual separation between entity types
The ability to quickly jump into related items

But they also highlighted areas needing deeper exploration:

Component behaviour and interaction nuances — something we learned only through usability sessions
Balancing detail vs. clarity for different user roles

Product Outcomes

What we proved:

Knowledge cards do surface meaningful context for enterprise search workflows.
Users engage more deeply when they can see relationships between entities.
Early design work provided confidence for stakeholders about future investment.

While the feature ultimately didn’t ship in that cycle, the research and design artefacts now inform future roadmap decisions — reducing risk for engineers and product leaders by providing evidence rather than assumptions.

Reflection & Lessons Learned

Research-Led Design Drives Clarity

Spending time on user interviews and heuristic research saved weeks of misguided design assumptions. Early qualitative evidence focused our hypothesis and shaped meaningful design decisions.

Prototypes are Communication Tools

High-fidelity prototypes acted as shared artifacts across design, engineering, and executive stakeholders — accelerating alignment and reducing ambiguity.

Context Matters More Than Content

Users cared less about data volume and more about relevance. Designing context-pivoted views (e.g., task-specific information) is far more impactful than generic card templates.

AI & Knowledge Work Demand Trust

In knowledge-centric AI products, perceived trustworthiness determines adoption. Transparency in data origin and concise context hierarchy are core design levers.

Next Steps

If this project continues, future work could include:

Personalisation layers that adapt card content based on user roles or past queries.
AI-driven summaries that synthesise entity history and relationships automatically.
Behavioural analytics to measure task success time with and without knowledge cards.

Continue Exploring...

AI Chatbot: Aiimigo

SaaS Product Feature Design

Using an AI chatbot and configured agentic models to efficiently deliver customer subject access requests.

Classification: AI Studio

SaaS Product Feature Design

The one-stop-shop for configuring and managing document classifications in Aiimi's Workplace AI.