Takeda Knowledge Extraction tool
Streamlining Pharmaceutical Literature Reviews Through Intelligent Design
Overview
My role: Product Designer
Timeline: 8 months (Discovery to MVP Launch)
Team: Cross-functional team of 12 (Product, Engineering, Data Science, Research Operations)
Focus Areas: Enterprise UX, AI/ML Interface Design, Workflow Optimization
The Problem
The Global Evidence & Outcomes (GEO) team at Takeda faced a critical efficiency challenge that was impacting pharmaceutical research timelines and resource allocation:
The Scale of the Challenge:
40+ literature reviews conducted annually
6+ months required for systematic literature reviews
$250K+ cost per systematic review
2-4 months for targeted reviews at $100K+ each
2-3 urgent ad hoc requests per month per lead
20+ hours of manual screening per urgent request
Core Problems Identified:
Time-to-insight bottleneck: Critical research decisions delayed by months
Resource drain: High-cost manual processes limiting team capacity
Risk of oversight: Tight deadlines leading to potential missed insights
Scalability limitations: Unable to handle increasing research demands
"We were spending more time collecting data than analyzing it. Our researchers were becoming data entry clerks instead of strategic thinkers." - GEO Team Lead
Findings after Research & Discovery Process
Users spent 70% of time on data extraction vs. analysis
Impact: High
Design Implications: Prioritize extraction automation
Each researcher had unique schema preferences
Impact: Medium
Design Implications: Build flexible, customizable schemas
Quality validation was the #1 concern with AI tools
Impact: High
Design Implications: Design transparent confidence indicators
The Solution: Intelligent Schema-Driven Extraction
1: Schema-First Approach
Rationale: Users needed control over data structure before AI processing
Impact: 40% reduction in post-processing time
2: Confidence-Based Review System
Rationale: Users required trust and verification mechanisms
Impact: 95% user acceptance rate for high-confidence extractions
3: Progressive Disclosure Interface
Rationale: Complex workflows needed to remain approachable
Impact: 60% reduction in onboarding time
Figma screens: Project reports, UI to create schema, Report data tables
Figma screens: Schema creation via AI prompt
AI schema prompting
Figma screens: Documenting report view
Figma screens: Filtering data rows by extracted data type
Sketching user workflow steps
User flow diagram
User journey flows
User journey flows
User journey flows
User journey flows