THE ART & SCIENCE OF SEAMLESS EXPERIENCES

Building Trust in AI-Driven Healthcare Decisions

A shared mental model for analytics, ML, and AI products

Healthcare is hard enough when the stakes are just health. Add analytics, machine learning, and AI into the mix, and the challenge multiplies: patients, providers, and administrators all have to understand how a system reaches its conclusions before they'll trust it with a real decision. That trust doesn't happen by accident.

I set out to solve this at the portfolio level, not one product at a time. Rather than reinventing research for each new application, I developed a shared healthcare mental model — a common foundation that let teams design analytic experiences people could understand, trust, and act on across a wide range of products.

The Approach: A Shared Mental Model

A mental model captures how a person understands the way something works. My insight was that if we could uncover and validate the right healthcare mental model, we could reuse it across products — giving every team a shared, research-backed starting point instead of guessing anew each time.

Grounding it in research. I uncovered inconsistencies in how users understood healthcare costs and quality of care, then dug into healthcare educational materials to surface a general mental model connecting effectiveness, efficiency, and equity to cost and individual wellness. That model became the backbone of a set of applied design techniques.

Validating before designing. Because mental models can be incomplete or incorrect, I treated validation as non-negotiable. We confirmed the model before making design decisions on top of it — a discipline that kept teams from building confident products on shaky assumptions.

Designing for lineage. With a validated model in hand, we designed experiences that answered the trust questions directly, making data and model lineage visible rather than hidden. The goal wasn't just accurate output — it was output people could understand and stand behind.

At a Glance

Business problem: Users across a broad healthcare portfolio couldn't understand or trust analytic, ML, and AI recommendations — and inconsistent understanding was blocking adoption
My role: Created and championed a shared mental-model approach as a core, reusable design process across the product portfolio
The approach: Research-driven mental models focused on data lineage and model lineage — the trust questions behind every recommendation
Scalability: One validated mental model applied across multiple products and user types, from data scientists to care managers to patients
Recognition: Presented at HFES International (2020) as a repeatable method for UX in analytics and ML
Strategic impact: Turned explainability and healthcare literacy into a design advantage, building trust in model-driven recommendations across product families

The Challenge

A model can be brilliant and still fail. If the people meant to use it don't understand where its answers come from, they won't act on them — especially in healthcare, where the wrong decision carries real weight.

My teams kept running into the same wall. Users couldn't answer two basic but critical questions:

Data lineage: Where did this data come from? Can I trust it? How does it relate to my job?
Model lineage: How did the model arrive at this answer or recommendation? Can I trust the logic behind it?

These questions apply to analytic, machine learning, and AI models alike — and when they go unanswered, trust collapses. Worse, I found that users' mental models of core healthcare concepts were often incomplete or simply wrong, and that those misunderstandings varied across market segments.

The strategic problem was clear: designing product by product wouldn't scale, and it wouldn't fix the underlying trust gap. We needed a shared foundation the whole portfolio could build on.

Key Insights

Building this approach shaped how I now lead product and design in complex, high-trust domains.

Trust is a design problem, not just a data problem. The most accurate model in the world is useless if users can't see how it works. Explainability has to be designed in from the start.
Literacy drives adoption. When we helped users understand the concepts behind a recommendation, their trust in analytic, ML, and AI outputs rose with it.
A shared model creates leverage. One validated mental model, reused across products, reveals connections that would stay invisible if each team worked alone — and builds a consistent line of sight to end users across the whole portfolio.
Mental models must be earned, not assumed. Users' understanding is often wrong. Validating it first is what separates a durable design strategy from an expensive guess.

The Mental Model in Action

The real power showed up when one shared model served very different users. Take Jack, a 55-year-old patient facing a knee replacement and worried about the cost. The same mental model shaped his experience across three very different tools.

For the data scientist. These users generate precise cost estimates for every local orthopedic surgeon, knowing that price varies widely by surgical facility. Mental-model-inspired design gave them a clearer way to create a line of sight for patients like Jack — turning complex estimates into price options he could actually compare.

For the care manager. These users help patients navigate the healthcare system and their benefits. Jack had learned from his care manager to think carefully about choosing between the Emergency Room and Urgent Care, a decision with a big impact on his out-of-pocket costs. We transformed the care manager experience to make educating patients on the right setting of care easier and more consistent.

For the patient. Finally, the model shaped a benefits-literacy series built directly for patients like Jack — helping him understand how his choices affect both his costs and his outcomes, so he could act with confidence when he decided to address his knee pain.

One model. Three user types. A consistent, trustworthy experience across all of them.

Why It Mattered

This work reflects how I lead. I look for the systemic gap beneath the surface problem — here, a portfolio-wide trust deficit that no single product fix could solve — and I build a durable, scalable answer that outlasts any one project.

The mental-model approach became a core design process for healthcare analytic applications and earned recognition at HFES International in 2020. But its real value was organizational: it gave teams a shared language, a validated foundation, and a repeatable way to build products people could trust across the entire portfolio.

My conviction is simple: in healthcare, AI and analytics only create value when people understand and trust them. Accuracy isn't enough. My job is to make the reasoning behind a recommendation clear, connect it to how people actually think, and give teams the shared tools to keep doing it at scale.