AI, Semantic Search & Knowledge Panels

Case Study
PROJECT MISSION: FIXING AN AI KNOWLEDGE ENHANCED BASE

Create an action oriented search experience to help users answer a question or take immediate action on information.

Client

Working with the executives and engineers at this company, we conducted discovery research and redesigned the search experience for their document management platform. The new design integrates AI into search to improve document findability and uses semantic search, facets, metadata, and knowledge graph techniques to support faster insights across complex workflows for their Fortune 500 clients.

What We Did

Our work centered around surfacing actionable knowledge and insights with zero-clicks.  

  • Discovery
  • Customer Research
  • Ideation Workshops
  • Metadata and Content Analysis
  • Knowledge Panel Education
  • UX/UI Design
  • Added AI Insights to Search

Outcomes

We evaluated their data model and introduced advanced AI search concepts, including knowledge panels and graph-based navigation. Based on direct user research with Fortune 500 clients, we designed innovative search experiences that improved actionability, clarity, connectivity, and relevance. The final solutions reflect how to get the value from AI-enhanced search, marking a significant departure from the legacy design.

The Challenge

Our client's SaaS platform automatically organizes documents for projects. The platform ingests assets and data from various sources and leverages machine learning to categorize and label information so that it is connected semantically at scale. 

Their AI worked great, but the search experience wasn't intuitive. Users reported the complex tabular search interface and filters as challenging to navigate. Inconsistent metadata compounded the issue, making it difficult for users to find the appropriate content.

Since this platform integrates data from various systems, each with distinct metadata categories, labels, and completeness levels, this resulted in a complex and cumbersome search experience.

As we set out to re-design the application, we aimed to:

  • Reduce in-app mental workload
  • Improve the intuitiveness and ease of use of search
  • Bring the value of their knowledge graph forward
  • Improve time-to-knolwedge

 

Our Approach

We followed our Search UX Framework to improve the platform.

Our framework involves:

  • Analyzing content types, data, and metadata
  • Assessing and refining taxonomies
  • Identifying and designing search result types
  • Defining actionable goals per search result type
  • Redesigning facets
  • Prototyping specific use cases
  • User testing with primary personas
  • Iterating designs based on user feedback

We believe in co-designing alongside key stakeholders and users to effectively create experiences that serve the needs of both audiences. Throughout our process, we sought feedback from and shared insights with our co-designers as we iterated. This approach allowed us to deliver rapid, vetted, effective solutions.

Developing Empathy

“(The platform) is not exactly giving all the search results I believe are in the data pool…I end up perusing through the documents manually to find what I need.”

Image shows a blurred screenshot of categorized user research.

Predictive UX deeply values evidence-based design so whenever possible, we prioritize capturing detailed feedback from users and internal stakeholders at every stage of the project. This approach ensures that we gather insights early and often, allowing us to make informed decisions that enhance the overall design and functionality of our solutions. By actively involving end-users and key team members in the feedback process, we aim to create products that truly meet their needs and expectations. Leading to more effective and user-centric outcomes.

Participant Interviews

Predictive UX conducted 5 remote moderated interviews with existing users. Their job titles ranged from VP of Operations to Project Manager to Facilities Manager. Some of the insights from participants were unique to their job title, while other insights overlapped.

Participants Desired:

  • File organization and standardization of file names so that data could be recognized and interpreted without having to click into a file.
  • Visible metadata within search so that search results are contextualized and discoverability is increased.
  • A clean layout with visual hierarchy and well spaced design elements.
  • Data status with confirmation and feedback loops around the project and its associated content.

 

Client Education

We began our work with a series of demos and workshops focused on education and knowledge sharing between our firm and the client's team. It was important to establish a shared understanding of the current state of their product while also examining concrete definitions of search, content types, metadata, and data structures.

The image shows two purple cards with questions about content types and shared metadata: How many unique content types are there? Which ones are used by everyone? and Is there a set of metadata that persists across all entities?
Image shows six blue sticky notes defining search types: Exact search, Site content search, Recommendation search, Unified search, Documents search, and Photo search.

Client

The US Chamber of Commerce T3 Innovation Network whose mission is to accelerate the use of digital tools to make the job market fairer and more inclusive.

What We Did

Led design and requirements refinement, collaborated with technical partners to align on decentralized storage, wallet integration, and credential rendering strategies.

Outcomes

A working POC with wallet-attached storage and a custom UI for linking verifiable credentials supported by a decentralized data model and resume rendering framework.

Delivery Time

12 weeks

 

 

 

Data Goals

Graph Traversal

A central feature of client's app is their graph traversal functionality. We leveraged Generative AI to showcase answers and related content and data relevant to a query, such as background information on specific products. While this concept can be implemented in various ways, for this design, we integrated it within the search results for contextual relevance.

Surfacing Metadata

We redesigned the search interface to make metadata more visible and actionable at every step. Previously buried in dense tables, metadata like file type, source, and status is now surfaced directly in results, filters, and previews, making it easier for users to scan, sort, and understand content at a glance. This shift empowers users to find the right documents faster and with greater confidence.

Surfacing Metadata

We redesigned the search interface to make metadata more visible and actionable at every step. Previously buried in dense tables, metadata like file type, source, and status is now surfaced directly in results, filters, and previews, making it easier for users to scan, sort, and understand content at a glance. This shift empowers users to find the right documents faster and with greater confidence.

Rich Visual Search Results

We introduced thumbnails to create a rich, visual preview experience for files like PDFs and drawings, allowing users to quickly recognize the content. Each result also includes the title, file type, source system, file size, and version, with location details more prominently displayed. Semantic highlights from within the document help users assess relevance at a glance, while suggested questions and related concepts, such as “People also ask” to provide deeper context and support faster decision-making.

Not sure what’s right for your company?

Book a call to discuss your needs.

 

Previous Project
AI in Healthcare
Next Project
Named Entity Recognition

PUXLogo_white

Predictive UX is for people-first organizations designing the future of AI + human interactions. Predictive UX named on Inc 5000 as one of the fastest growing companies in the US.

Predictive UX News and Insights.

Email:

Copyright © 2025, Predictive UX  | Privacy Policy