Berkeley IT Department
As UC Berkeley launched its 2025 campus-wide AI initiative, a key challenge emerged: ensuring the tools would be trusted and adopted by staff. Yet, fewer than 10% of intended staff adopted the AI tool, River. To address this gap, I led a research effort to uncover staff needs and pain points. These insights informed a strategic UX report with actionable principles to guide the long-term integration of AI across campus operations.
Role
UX & Research Intern
Timeline
Jun 2025 - Aug 2025
Team
1 UX & Research Intern
1 Project Manager
1 User Experience Designer
My Role
I supported the project manager in conducting 21 user interviews and analyzing a campus-wide admin survey (800+ responses). I then led thematic synthesis through affinity mapping and translated key insights into a strategic direction.
Impact
The final user research report was presented to the Chief Technology Officer, helping shape the AI adoption roadmap at Berkeley IT. Checkout the Spring 2025 Information Technology Survey Analysis here ↗
When AI Lacks Trust, Clarity, and Seamless Integration
To improve administrative efficiency and modernize staff workflows, Berkeley piloted River, a custom AI assistant. Modeled after similar initiatives at peer institutions, River aimed to boost staff efficiency and reduce manual work.
However, less than 10% of the staff actually adopted the tool. This revealed a core problem: efficiency alone isn’t enough. For AI to succeed in this context, it must be trusted, understood, and seamlessly fit into existing workflows.
The Challenge
AI Adoption User Research Goals
The Berkeley AI Hub (Sponsor)
As the primary sponsor funding the project, the AI Hub's main concern was return on investment (ROI). They needed to justify the budget for River AI and required quantitative outcomes from me to prove its value and necessity for the university.
The AI/Automation Team
The AI/automation team was focused on implementing agentic workflows to automate repetitive tasks. A key goal behind this initiative was to see whether staff could realistically use Rivers existing agentic workflow to streamline their existing process.
The Berkeley IT Department
Berkeley IT champions a human-centered AI approach. Their priority was to ensure that River AI was not just technically proficient, but also usable, trustworthy, and valuable for the staff. They wanted to show why user research for AI integration from the start was essential.
Uncovering Core Pain Points Through 2 Key Research Methologies
After conducting 21+ interviews and a mixed-methods study with 800+ survey responses, I had hundreds of data points. To synthesize them, I used affinity mapping, a key step that turned scattered data into three core, evidence-based thematic insights that shaped the report.
A Need for Better Guidance on Effective AI Prompting
Staff were eager to use the tool, but many struggled to prompt it effectively. Questions like “What phrasing works best?” and “How are others using it?” were common, with 68% of survey respondents saying they needed clearer guidance and examples.
When AI Workflow Automation Isn’t Intuitive
Another key insight I gathered was that staff wanted to offload repetitive tasks to AI but faced barriers due to the tool’s complex, technical automation features. The interface prevented 52% of staff, who were interested in automation, from effectively using it.
Addressing Data and Privacy Concerns
Users were uncertain about how their data was handled due to unclear communication around privacy and security. This lack of transparency led staff to avoid using River AI for sensitive student or administrative tasks. Addressing this requires visible safeguards and clearer, more transparent communication to rebuild trust.
Beyond the Data: The 3 Personas That Define AI Use
Furthermore, our user research revealed a clear insight: 'staff' isn't a single user type. Attitudes toward AI varied widely, from skepticism and fear to curiosity and excitement. To better visualize this diverse audience, we translated our interview and survey data into three personas grounded in real findings.
Framing the Opportunity
How might we encourage adoption of AI tools among Berkeley staff by integrating these technologies into their existing workflows and reducing friction through trust, relevance, and usability?
Translating Insights into Early Concepts
Given these findings, I explored multiple design directions to address the core pain points identified such as navigation complexity, unclear hierarchy, and inefficiencies in agentic task flow. I did mid-fidelity ideation focused on translating research insights into tangible design solutions, establishing the foundation for future iteration.
Next Steps
Expand Research & Track Impact
I suggested that the IT team explore how specific departments, could integrate AI into their unique workflows, since each team has distinct use cases, needs, and levels of AI readiness. They are also planning to rerun the campus-wide survey 6–12 months post-launch to measure changes in sentiment, trust, and adoption.
Phase the Implementation Roadmap
Pending the CTO’s decision on whether we’ll build a custom model, I’ll collaborate with developers later in the semester to shape a product roadmap, starting with an MVP focused on trust and usability. Early priorities will likely include transparent data handling and privacy tools.