Making AI-generated insights something people actually act on.
I led the UX for an AI-generated insights dashboard for a global healthcare brand, working closely with engineering and data science to turn complex industry data into a clear, decision-ready interface for executives and operations teams.
An AI system was generating valuable healthcare insights, but the raw output wasn't usable by the people who needed it. Executives and operations teams were still relying on manual reporting cycles because they couldn't trust or interpret what the AI was surfacing.
Senior UI/UX Designer leading end-to-end UX for the dashboard. I worked closely with engineering, data science, and client stakeholders to figure out which insights to surface, how to structure them, and how to make them feel trustworthy and actionable.
A real-time dashboard that replaced a multi-day manual reporting cycle. Executives and operations leads shifted from weekly PDF reports to daily dashboard use, with time-to-insight dropping from days to minutes.
The challenge
The AI was doing its job. The problem was that nobody could use what it was producing.
A global healthcare brand had invested in an AI system that analyzed complex industry data and surfaced insights across their operations. But the output landed as raw data exports and dense reports that required a data analyst to interpret before anyone could act on them. Executives were still waiting on weekly summary PDFs. Operations teams were running parallel spreadsheets to double-check what the AI was saying. The technology was there. The trust and usability weren't.
My job was to lead the UX for a dashboard that would change that. That meant getting close to the people who'd use it every day, understanding what they actually needed to see, and designing an interface that made AI-generated insights feel clear, trustworthy, and ready to act on. Dense healthcare data doesn't have to be hard to understand. That's exactly what this project set out to prove.
How I approached it
The design work for a tool like this happens long before anyone opens a chart library. It starts with understanding what people actually need to know.
Understand the actual users
Interviews and workshops with executives, ops leads, and analysts. Not to learn what data existed, but to understand what decisions they were trying to make, what they were unsure of, and where they were losing time waiting on information.
Establish information hierarchy
Mapped AI-generated outputs to a visual hierarchy based on how often each insight was acted on. What surfaces at a glance, what lives one click away, what belongs in drill-down detail. This is where most of the real design work happened.
Design for AI transparency
Healthcare executives needed to trust what the AI was telling them before they'd act on it. I designed confidence indicators, data source labels, and plain-language explanations of AI outputs so users could understand not just what the insight was, but how it was derived.
Iterate with real users on real data
Live walk-throughs with the same executives and ops leads who shaped the brief, using actual client data. We refined until the dashboard answered their first three questions without scrolling, and until they trusted what it was showing them.
Decisions that mattered
Making AI outputs feel trustworthy, not just present
The biggest barrier to adoption wasn't the interface. It was trust. Healthcare executives weren't going to act on an insight they didn't understand. I designed plain-language explanations for every AI-generated output, confidence indicators, and clear data source labels. Showing the "why" behind each insight was as important as showing the insight itself.
One primary question per view
The temptation in any data-heavy product is to surface everything you have. We held firm on focus. Each view was designed to answer one primary question clearly, with supporting detail one click away. That discipline made every screen faster to read and easier to act on.
Designing empty and low-confidence states as carefully as loaded ones
In a healthcare context, a dashboard that goes silent without explanation is genuinely dangerous. Empty states, low-confidence AI outputs, and "data still loading" views all got first-class design treatment, because those are the exact moments when a user's trust either holds or breaks.
Filters built around user mental models, not the data schema
The default filter sets matched how executives and ops teams actually think about their work, by region, by product line, by time period, not how the underlying database was organized. That one IA decision had a bigger impact on adoption than any visual design choice we made.
What changed
Specific impact data is shared in the full case study Public-facing highlights:
Time-to-insight dropped from a multi-day manual reporting cycle to real-time dashboard access. Executives stopped waiting on weekly PDFs and started checking the dashboard as part of their daily routine.
Operations teams stopped running parallel spreadsheets to validate what the AI was surfacing. The dashboard became the source of truth, which was the clearest signal the trust problem had been solved.
Post-launch interviews with the client team showed self-reported time-to-decision dropped meaningfully, and users described feeling more confident acting on insights compared to the prior reporting process.
The visual system and design patterns were built to hold up as new AI outputs, new data sources, and new user roles got added. The client team has continued expanding the dashboard without needing to redesign the foundation.
What I'd carry forward
If users don't understand where an insight came from or how confident the AI is, they won't act on it, no matter how accurate it is. Explaining the reasoning behind each output in plain language mattered as much as any chart or layout decision I made.
Once you figure out the two or three things a user actually needs to see right now, the visual work becomes pretty straightforward. Getting that clarity is where the real design work happens, before you open a single tool.
A dashboard nobody opens is a design failure. It doesn't matter how impressive the AI behind it is. The measure of success was whether real people were checking it every day to make real decisions. That's the bar I designed toward.
Working with complex data? Let's talk.
I love talking about design and hard problems. Drop me a line or connect on LinkedIn. I'd genuinely love to hear from you.