700

%

Clinical researchers, full AI assistance in 6 weeks

700

%

Clinical researchers, full AI assistance in 6 weeks

670

%

Faster patient recruitment uisng embedded AI

670

%

Faster patient recruitment uisng embedded AI

2500

+

Life sciences products in scope

2500

+

Life sciences products in scope

Challenge

IQVIA’s product teams operated largely in silos with little coordination across products. Competitors were winning clients with more cohesive user experiences, while employees and clients hesitated to adopt Ada because they did not yet trust it.

strategy

I mapped leadership and customer definitions of product value, assessed each product team’s readiness for product-led change, co-authored an enterprise UX strategy with Jaime Levy, and built the operating model around it. In parallel, I created Ada’s trust-based AI adoption framework and experimentation tools.

Results

During my $4.4M UX pilot, I uncovered executive resistance driven by feature-based performance metrics. I helped shift leaders toward outcome-based incentives, unlocking 30+ product upgrades while Ada’s rollout across 15+ products accelerated patient recruitment by 67%.

Challenge

The trust gap

IQVIA’s AI group had built models that could simulate patient recruitment, predict trial delays, and flag statistical risks before they escalated. The problem was that researchers and clients still hesitated to use them.

The barrier wasn’t the technology. It was trust.

Researchers were being asked to rely on systems that could influence high-stakes clinical decisions while still remaining professionally accountable for the outcome.

So the CTO handed the problem to me.

Challenge

The trust gap

IQVIA’s AI group had built models that could simulate patient recruitment, predict trial delays, and flag statistical risks before they escalated. The problem was that researchers and clients still hesitated to use them.

The barrier wasn’t the technology. It was trust.

Researchers were being asked to rely on systems that could influence high-stakes clinical decisions while still remaining professionally accountable for the outcome.

So the CTO handed the problem to me.

Earning the CTO's trust

A year earlier, IQVIA had no design organization after the merger.

At the same time, competitors were marketing usability as a differentiator.

I met with the CTO and proposed a small pilot team: 6 designers, researchers, and PMs focused on proving that customer-centered development could help win and retain pharma clients across 5 products.

Earning the CTO's trust

A year earlier, IQVIA had no design organization after the merger.

At the same time, competitors were marketing usability as a differentiator.

I met with the CTO and proposed a small pilot team: 6 designers, researchers, and PMs focused on proving that customer-centered development could help win and retain pharma clients across 5 products.

Examples of IMS/Quintiles various styles in 2016

The pilot defined design’s role and value

The pilot contributed $4.4M in new contracts and exposed the operational challenges IQVIA would need to solve to become product-led.

That changed how leadership viewed design.

When the company needed someone to bridge the gap between sophisticated AI systems and the people expected to trust them, the CTO came to me.

The pilot defined design’s role and value

The pilot contributed $4.4M in new contracts and exposed the operational challenges IQVIA would need to solve to become product-led.

That changed how leadership viewed design.

When the company needed someone to bridge the gap between sophisticated AI systems and the people expected to trust them, the CTO came to me.

Strategy

Before people could trust Ada, they needed to watch it work.

Uncovering the real problem

I led research across trial sponsors, statisticians, and regulatory teams to understand why highly accurate AI models still weren't being adopted. The issue wasn't comprehension. Teams understood the recommendations. What they couldn't do was safely stand behind them when sponsors questioned the results. There was no clear path back to the evidence behind the model's conclusions.

Strategy

Before people could trust Ada, they needed to watch it work.

Uncovering the real problem

I led research across trial sponsors, statisticians, and regulatory teams to understand why highly accurate AI models still weren't being adopted. The issue wasn't comprehension. Teams understood the recommendations. What they couldn't do was safely stand behind them when sponsors questioned the results. There was no clear path back to the evidence behind the model's conclusions.

Experience design

Designing the trust model

I hypothesized that trust, not accuracy, was the real barrier.

Drawing on the trust research of Lee & See and Hoff & Bashir, I designed Ada as a phased trust framework that expanded AI autonomy only after users had verified its behavior over time.

Experience design

Designing the trust model

I hypothesized that trust, not accuracy, was the real barrier.

Drawing on the trust research of Lee & See and Hoff & Bashir, I designed Ada as a phased trust framework that expanded AI autonomy only after users had verified its behavior over time.

Phase 1

Observer

Watch Ada work. No delegation. Build pattern recognition before trust is asked for.

Phase 2

Peer

Collaborate with Ada. Review outputs. User retains the final clinical call.

Phase 3

Assistant

Delegate to Ada. Full AI assistance. Advancement triggered by comfort, not calendar.

Experience design

Scaling the trust framework

Progression from observation, to collaboration, to delegation was driven by behavioral signals like previews, dismissals, and direct acceptance patterns, creating a visible evidence trail of when users chose to review, reject, or rely on the system.

The result: I advanced 70%+ of clinical researchers to full AI assistance in 6 weeks

Experience design

Scaling the trust framework

Progression from observation, to collaboration, to delegation was driven by behavioral signals like previews, dismissals, and direct acceptance patterns, creating a visible evidence trail of when users chose to review, reject, or rely on the system.

The result: I advanced 70%+ of clinical researchers to full AI assistance in 6 weeks

Example plot of AI adoption by users from Observer (phase1), Peer (phase 2), to Assistant (phase 3) over 6 weeks.

scaling innovation & product adoption

The admin experience

The real bottleneck

Ada adoption slowed for a simple reason: changing anything was expensive. Every workflow required alignment across AI outputs, compliance requirements, sponsor expectations, and AI scientist availability. Even small adjustments could take weeks to validate, so teams stopped experimenting unless the need felt urgent.

The deeper problem was not the models themselves. Teams had no safe way to explore ideas before asking others to approve them. Without a sandbox environment, every proposed change carried production-level scrutiny and depended on specialists already committed to larger initiatives.

scaling innovation & product adoption

The admin experience

The real bottleneck

Ada adoption slowed for a simple reason: changing anything was expensive. Every workflow required alignment across AI outputs, compliance requirements, sponsor expectations, and AI scientist availability. Even small adjustments could take weeks to validate, so teams stopped experimenting unless the need felt urgent.

The deeper problem was not the models themselves. Teams had no safe way to explore ideas before asking others to approve them. Without a sandbox environment, every proposed change carried production-level scrutiny and depended on specialists already committed to larger initiatives.

scaling innovation & product adoption

The admin experience

The real bottleneck

Ada adoption slowed for a simple reason: changing anything was expensive. Every workflow required alignment across AI outputs, compliance requirements, sponsor expectations, and AI scientist availability. Even small adjustments could take weeks to validate, so teams stopped experimenting unless the need felt urgent.

The deeper problem was not the models themselves. Teams had no safe way to explore ideas before asking others to approve them. Without a sandbox environment, every proposed change carried production-level scrutiny and depended on specialists already committed to larger initiatives.

Example of flow diagram used to map tasks and evaluate UI concepts

Designing for safe autonomy

Working closely with the AI scientists, I designed the admin experience to make experimentation practical for the people accountable for the outcome.

The interface exposed parameter settings, algorithm behavior, compliance implications, and the operational impact of proposed changes directly inside the workflow. Product leaders could evaluate tradeoffs and test adjustments themselves instead of waiting on scientists for every iteration.

Designing for safe autonomy

Working closely with the AI scientists, I designed the admin experience to make experimentation practical for the people accountable for the outcome.

The interface exposed parameter settings, algorithm behavior, compliance implications, and the operational impact of proposed changes directly inside the workflow. Product leaders could evaluate tradeoffs and test adjustments themselves instead of waiting on scientists for every iteration.

Annotated key features and screenshot of the AI admin console.

Resuts

Through more than 10 structured experiments, we reduced barriers tied to trust, error recovery, feedback speed, and accountability visibility. The resulting tools cut ML adjustment time from 3 days to 30 minutes across 24 clinical products, enabling non-technical teams to shift from quarterly update cycles to weekly optimization.

Resuts

Through more than 10 structured experiments, we reduced barriers tied to trust, error recovery, feedback speed, and accountability visibility. The resulting tools cut ML adjustment time from 3 days to 30 minutes across 24 clinical products, enabling non-technical teams to shift from quarterly update cycles to weekly optimization.

The customer experience

AI-enabled research, just in time for COVID-19

Ada’s recommendations were never presented as answers alone.

The interface exposed the underlying data and reasoning behind each recommendation so clinical researchers could inspect the evidence before acting on it. Researchers could see how Ada reached a conclusion before deciding whether to rely on it.

The customer experience

AI-enabled research, just in time for COVID-19

Ada’s recommendations were never presented as answers alone.

The interface exposed the underlying data and reasoning behind each recommendation so clinical researchers could inspect the evidence before acting on it. Researchers could see how Ada reached a conclusion before deciding whether to rely on it.

Example of client-facing research dashboard with Ada (Phase 1) active.

Results

That distinction mattered in regulated clinical environments where researchers remained accountable for scientific, patient, and regulatory outcomes. Accuracy alone would not have been enough. Researchers needed to understand why the system reached a conclusion before they could confidently trust it.

In trials where the system was deployed, patient recruitment moved 67% faster.

Shortly after, when COVID-19 accelerated vaccine research worldwide, IQVIA’s AI-enabled research systems helped support nearly 1/3 of global vaccine research efforts.

Results

That distinction mattered in regulated clinical environments where researchers remained accountable for scientific, patient, and regulatory outcomes. Accuracy alone would not have been enough. Researchers needed to understand why the system reached a conclusion before they could confidently trust it.

In trials where the system was deployed, patient recruitment moved 67% faster.

Shortly after, when COVID-19 accelerated vaccine research worldwide, IQVIA’s AI-enabled research systems helped support nearly 1/3 of global vaccine research efforts.

Design Foundations

Scaling the organization behind the transformation

While Ada was rolling out, I was building the organization needed to support IQVIA’s remaining 250 products.

I co-authored the enterprise UX transformation strategy with Jaime Levy, creating the phased approval model that funded the expansion. The original six-person pilot team grew to 45 across research, design, content, and product management in 36 months, expanding coverage from 5 to 30+ products.

I built research operations from scratch, embedded designers directly into product teams, and introduced a six-pod operating model that reduced cross-team bottlenecks by 40% and tripled delivery capacity without proportional headcount growth.

I also founded a storytelling studio with Disney animators to help clients and executives understand complex product strategy.

Scaling Design Coverage at IQVIA

Products

Design owned

Design influenced

Designers

Specialists

Merger 2016

Year 1 Pilot 2017

Ada Year 1 2018

Org Complete 2020

Map of the IQVIA Design Org (2021)

Example "sizzle reel" the team created at the end of each year highlighting outstanding stories, events, and craft from that year.

key trade-off

The CTO wanted 150 designers. I told him to wait.

The choice

Challenge: My 1-year design pilot worked. The CTO wanted to scale it across 250 products in 160 languages immediately. The problem: the success came from a small SWAT team. We had proven design could work. We had not yet built the infrastructure to scale it responsibly.

The decision came down to two paths

Scale Fast

The CTO's plan: Hire 150 designers immediately and embed them across the org while building infrastructure in parallel. The debt would follow.

Scale Fast

The CTO's plan: Hire 150 designers immediately and embed them across the org while building infrastructure in parallel. The debt would follow.

Strategic Growth

What I chose. Spend one quarter building the foundation first. Then scale through targeted pods aligned to the UX strategy and product readiness.

Strategic Growth

What I chose. Spend one quarter building the foundation first. Then scale through targeted pods aligned to the UX strategy and product readiness.

Outcome

I slowed the rollout. Adding 150 designers before establishing shared systems and governance would've buried the organization in debt we couldn't unwind later.

It cost the CTO face internally. It cost me goodwill. By 2022, the org peaked at 75 designers instead of 150.

But the slower path held. Design became part of product strategy instead of a downstream service function. Over time, that decision earned me a voice in nearly every major initiative, and the CTO became one of our strongest advocates, presenting our work company-wide for three consecutive years.

Results

Select Highlights

What started as a year-one pilot to prove design's value at IQVIA evolved into a new operating model for how the world's largest CRO made AI decisions in clinical research.

Clinical Trials
470

%

Patient recruitment time in deployed trials.

Clinical Trials
470

%

Patient recruitment time in deployed trials.

AI Adoption
700

%

Users adopted AI-assisted workflows in 6 weeks

AI Adoption
700

%

Users adopted AI-assisted workflows in 6 weeks

Org. Development
00
450

UX practitioners by end of year 3

Org. Development
00
450

UX practitioners by end of year 3

Contract Impact

$

40

M+

Value of new contracts attributed to pilot

Contract Impact

$

40

M+

Value of new contracts attributed to pilot