Designing AI in Healthcare Without Falling for the Hype
How I helped LeanTaaS separate AI excitement from real product value.
Project Snapshot
Impact
(different kind)
This project didn’t produce huge revenue metrics. Its impact was:
Avoided clinical and reputational risk
Shaped company AI direction
Built internal literacy around AI
Validated contained AI use cases
Prevented wasteful over-investment
KEY TAKEAWAY
Sometimes impact is what you ship. Sometimes it’s what you prevent.
“Ask iQueue pulls it so much faster than me clicking through all my locations… It pinpointed exactly what I needed. I didn’t expect it to be that helpful and it was amazing.”
Operations Leader at UC San Francisco
"Ask iQueue was pretty quick and accurate. If I want to know what was our monthly drop in the census, I could see that without having to go and navigate through metrics. I put the thumbs up on every question and answer that I received.
Operations Manager at Stanford
Customer
feedback
Organizational impact
Customer Success teams were initially weary of AI. I led sessions to explain:
What hallucinations are
How we mitigate risk
Where AI fits (and doesn't)
KEY TAKEAWAY
This built internal confidence and reduced fear around AI adoption.
Context
Company-wide push to deliver customer-facing AI
Constraints
No dedicated AI team
1 PM, 1 designer, 2 engineers
One quarter timeline
High hallucination risk
My Role
Product designer shaping AI direction, scoping, and validation
Core Challenge
Deliver AI value without damaging trust or credibility
When AI exploded, every software company rushed to prove they had AI.
LeanTaaS was no exception. Leadership wanted visible AI features fast — something customers could see, demo, and associate with innovation.
The challenge? We specialize in healthcare operations, where wrong answers don’t just look bad — they can undermine trust, ROI, and even patient safety.
KEY TAKEAWAY
This project wasn’t about shipping flashy AI. It was about deciding what AI should and shouldn’t do.
The hype vs reality
Leadership envisioned ChatGPT-level support: conversational, fast, and broadly capable.
But we lacked resources, infrastructure, and time. More importantly, we faced real risk.
If AI gave wrong answers, it could undermine our ROI claims, contradict operational guidance, and confuse customers about best practices.
KEY TAKEAWAY
In healthcare, trust is everything. We couldn't allow AI to undermine that.
Blindly shipping flashy AI could lead to:
Incorrect staffing guidance
Misinterpreted performance metrics
Hospitals questioning our credibility
Customer churn from perceived lack of ROI
Clinical staff fearing replacement or loss of human judgment
KEY TAKEAWAY
The risk wasn’t embarrassment — it was trust erosion. In healthcare, a wrong AI answer isn’t just incorrect — it can change clinical decisions.
My position: Start small, stay contained
Instead of broad AI, I proposed small, contained use cases with low hallucination risk. I created a mini-roadmap outlining:
What we could realistically build
How risk scaled with complexity
How to grow responsibly
We lacked long-term direction and I wanted a clear path forward.
AI for Navigation
AI-generated explanations for nurse recommendations
Automated staffing and assignment requests
We initially launched a modest feature: AI for navigation
Users could ask metric questions and AI would:
Navigate to the correct page
Apply filters
Surface the right view
KEY TAKEAWAY
It wasn’t flashy, it was practical. Some stakeholders initially doubted its value. So I proposed a limited pilot with trusted customers and clear expectations.
A moment of validation: Pilot feedback showed real utility
“Ask iQueue pulls it so much faster than me clicking through all my units. It pinpointed exactly what I needed. I didn’t expect it to be that helpful.”
“Ask iQueue was pretty quick and accurate. If I want to know what was our monthly drop in the census, I could see that without having to go and navigate through metrics. I put the thumbs up on every question and answer that I received.”
KEY TAKEAWAY
It also confirmed something important: AI doesn't need to be magical to be valuable. Reducing friction is real value.
Pushing toward higher value
Next, I proposed AI-generated explanations for nurse recommendations. Why this mattered:
Customers frequently questioned assignments
Our assignment algorithm had clear logic
Hallucination risk was low
Value was high
User testing was very positive. Unfortunately, shifting priorities pulled engineering away before launch.
A user could ask AI to explain nurse recommendation — something customers questioned daily.
AI provides reasons based on the prioritization score and built-in logic and automates manual tasks.
Result: Improved trust in our recommendations and automated manual workflows.
KEY TAKEAWAY
A good idea doesn’t always mean the right timing.
Shaping AI strategy company-wide
Over time, our approach evolved:
Instead of duplicating AI work across products, we moved toward:
Shared AI platform strategy
Contained use cases
Incremental learning
&
I worked with our CTO to align on:
Responsible scoping
Hallucination risk
Trust preservation
&
I led sessions across the company to build confidence and reduce fear of AI:
What are hallucinations
How we mitigate risk
Where AI fits (and doesn't)
What this project says about my approach
I don’t chase trends. I evaluate them and ask:
Is this valuable?
Is this safe?
Is this scalable?
Is this worth building?
KEY TAKEAWAY
Good design isn’t just about shipping more. It’s about protecting users and the business.
Reflection
AI in healthcare isn’t just a capability question. It’s a responsibility question. This project taught me:
Hype fades, trust remains
Small wins beat big risks
Responsible AI design is a product decision, not a feature