Case Study 1 — AI Cancer Drug Discovery · Pfizer · 2024
HealthTech AI Research 2024

Accelerating Cancer Drug Discovery

Pfizer's oncology R&D scientists were drowning in biomedical data. I designed an AI interface that turns complex research queries into instant, visualised insights — fundamentally changing how drug discovery is conducted.

AI Interface Design Data Visualisation User Research Interaction Design Prototyping
Client
Pfizer (R&D)
My role
Lead Product Designer
Domain
AI · HealthTech · Data Viz
Platform
Web Application
Research Assistant — multi-device overview Research Assistant — mobile screen
80%
Faster research insights
64%
Time saved per query
92%
User satisfaction
Research throughput

The Problem Worth Solving

Pfizer's oncology researchers were drowning in data they couldn't query intelligently. Cross-referencing 15+ databases manually, spending 40% of their time on retrieval rather than analysis, and waiting days for insights that should take minutes. The opportunity was clear: eliminate the gap between question and discovery.

Project Goal

Design an AI assistant interface that lets researchers ask complex questions in natural language and receive structured, visualised insights — dramatically compressing the time from question to discovery.

From Question to Insight in Seconds

The core interaction model: researchers type a question in plain English. The AI parses intent, queries the relevant datasets, and returns a visualised insight panel — with source citations, confidence indicators, and suggested follow-up queries.

Research Assistant — desktop overview screen
The home screen — researchers select a focus area and enter a query in natural language.
Research Assistant — natural language query returning a grouped bar chart for Gefitinib resistance scores
A researcher queries resistance scores for Gefitinib across alteration types — the AI returns a structured, interactive chart in seconds.

AI transparency was not a nice-to-have for expert users — it was the product. Every AI response shows its sources, confidence levels, and the datasets it queried. Without interpretable reasoning, the tool would have been rejected regardless of its accuracy.

What the Research Revealed

"I don't want to learn a new query language. I want to ask the data a question the way I'd ask a colleague."

— Senior Research Scientist, Pfizer Oncology

I conducted 24 in-depth interviews with oncology researchers, data scientists, and clinical trial managers across three continents — plus shadowing sessions and a 2-week structured user test with 35 team members. Four themes consistently surfaced:

01
Discovery Timeline Bottleneck
3–6 months consumed in literature review alone — before a single hypothesis is tested.
Impact Delay in life-saving cancer treatments reaching patients.
02
Information Fragmentation Crisis
Scientists navigating 15+ disconnected databases daily with no unified query layer.
Impact Critical compound-target relationships are routinely missed.
03
Research Duplication Issue
Teams unknowingly duplicating efforts across siloed labs and institutions.
Impact Wasted resources that could fund entirely new research directions.
04
Visualisation Gap
90% of research presentations require manual chart and diagram creation from scratch.
Impact Automated generation of research visualisations remains entirely absent.

From Concepts to Decisions

Three interaction decisions that shaped the core experience.

01

Search interface vs structured query builder

Researchers wanted to ask questions using domain language rather than learn another system. Natural language removed the friction of query construction entirely.

Explored
Filters, query-builder, conversational input
Selected
Natural-language input
02

Answer first vs evidence first

Expert users needed to interrogate the reasoning before trusting the result. Surfacing sources, confidence, and queried datasets made the AI's logic visible and auditable.

Explored
Clean AI-generated answer / answer with expandable evidence
Selected
Sources + confidence + queried datasets
03

Static answer vs exploratory result

Researchers rarely stop after the first answer — one finding creates the next question. An exploratory model with interactive visualisations and suggested follow-ups matched how scientists actually think.

Explored
Generated summary / interactive visualisation + suggested follow-ups
Selected
Exploratory model

Results That Changed How Pfizer Approaches Research

After a three-month pilot with 40 researchers, the results were significant. The research team responded positively to the introduction of the AI, expressing enthusiasm for its integration into their daily work and reporting measurable gains in productivity within the first month.

80% faster insights
Time to retrieve and synthesise research dropped from days to hours in controlled task testing.
64% time saving
Per research query — freeing scientists to focus on interpretation rather than retrieval.
92% user satisfaction
Post-pilot survey. The natural language interface was consistently cited as the standout feature.
3× research throughput
Teams could explore 3× as many drug-target hypotheses per week compared to the legacy workflow.

"The rapid generation of these visualizations is truly remarkable. Previously, creating such insightful representations would have consumed two to three weeks of dedicated effort within my research workflow. However, with the assistance of this resource, I am now able to obtain these visualizations in under a minute, representing a significant acceleration of my research process."

Dr. Mika Chan*
Senior Computational Biologist

"I am highly impressed with the capabilities of Research Assistant. Its potential is evident, and I envision it being particularly valuable for my team, especially those members whose expertise lies in data analysis rather than software development. The intuitive nature of the platform could empower them to conduct more sophisticated analyses without requiring advanced programming skills."

Dr. Michael Rodriguez*
Director, Computational Biology

"The speed with which RA generates document summaries is truly impressive. This capability significantly accelerates my document creation process, enabling me to produce materials in mere minutes that previously required several days of manual effort. This represents a substantial improvement in efficiency, particularly given the monthly volume of documents I generate."

Dr. Lizzie Watson*
Research Team Lead

Reflections

User Diversity

Expert UX isn't necessarily complex UX.

Designing for scientists with radically different technical fluency taught me that domain sophistication doesn't mean users want sophisticated interfaces.

AI Trust

Explainability is part of the interaction model.

Building trust in AI-generated research outputs showed me that transparency isn't something added after the answer — it's woven into how the answer is given.

Data Communication

Visualisation is part of the reasoning.

Turning complex datasets into understandable research outputs showed that the right representation helps researchers interpret relationships, not just consume an answer.

Next case study
Designing a Premium Delivery Service Experience