Docbot
Transforming Access To Critical Procedures
Link
NDA
Timeline
2024 – 2025
My role
User Experience Lead
GenAI-powered chatbot designed to simplify access to procedural documents across company. Enterprise solution that now supports over 10K users globally.
Leading DocBot's design taught me that successful enterprise UX isn't just about making things pretty or easy to use – it's about understanding complex workflows, building trust through transparency, and creating solutions that adapt to individual context rather than forcing everyone into the same experience.
Most importantly, I learned that measuring UX success in business terms – time saved, productivity gained, compliance improved – creates a virtuous cycle where investing in user experience becomes obviously valuable to stakeholders, making it easier to secure resources for the next round of improvements.
The Problem We Faced
Picture this: You're an employee in Switzerland needing to find a specific procedure for handling materials. You know the process exists somewhere in the thousands of Standard Operating Procedures, but finding the right section feels impossible. You spend 15-20 minutes clicking through documents, searching keywords, and still aren't sure you found the most current version.
This was the daily reality for over 40,000 associates across Development and Operations. The existing system demanded deep procedural knowledge just to find basic information.
Core Problems:
- Manual search through thousands of documents required expert knowledge
- No personalization based on user training or location
- Limited multilingual support for global teams
- Inconsistent compliance due to outdated or wrong procedures
Understanding Our Users
I started by talking to the people who lived this problem every day. Through workshops and interviews with early adopters and Subject Matter Experts, a clearer picture emerged. The issue wasn't just about search – it was about context, personalization, and trust.
One operations manager told me: 'I know the information is there, but I need to know exactly what I'm looking for before I can find it. It's like asking someone for directions when you already know where you're going.'
Key Research Insights:
- Conversational need: People wanted to ask questions naturally, not construct search queries
- Context matters: Solutions needed to understand individual training, location, and role
- Trust is critical: In pharma, AI responses require transparent citations and confidence indicators
Design Strategy & Solutions
Armed with these insights, I began envisioning DocBot as more than just a search tool. It would be an intelligent conversation partner that could clarify ambiguous requests, suggest relevant filters, and provide personalized responses based on each user's unique context.
The breakthrough came when I realized we needed progressive disclosure – showing users just enough complexity to be powerful without overwhelming them. Instead of overwhelming users with every possible filter upfront, DocBot would listen to natural language queries like 'Is there a local process for handling materials in Switzerland?' and intelligently suggest relevant refinements.
Design Principles:
- Conversational Intelligence: DocBot asks clarifying questions for vague prompts
- Smart Filtering: Dynamic suggestions based on document types, scopes, and entities
- Personalization: Tailored responses using U4G curricula and site-specific documents
- Transparency: Citation panels and confidence scores build trust
Prototyping & Testing
I designed the interface around a simple conversation metaphor, but with powerful capabilities beneath the surface. The right sidebar would house contextual filters that appeared dynamically based on what users were asking about, while the main area would feel like chatting with a knowledgeable colleague.
The early adopter program became our design laboratory. Every week, we held open sessions where real users would try DocBot with their actual work scenarios. Watching someone successfully find a complex procedure in 2 minutes instead of 20 was incredibly rewarding, but the failures taught us even more.
Key Interface Elements:
- Left Sidebar: Dynamic filters and training groups with progressive disclosure
- Main Chat: Conversational interface with citation panels for transparency
- Filter Chips: Visual, selectable refinements that appear contextually
- Confidence Indicators: Trust-building elements for AI responses
Testing Approach:
- Weekly open sessions with early adopters
- Real-world scenario testing with actual work queries
- A/B testing of filter placement and visual hierarchy
- Cross-platform testing (Angular and React environments)
Results & Impact
Six months after launch, the numbers told a compelling story, but the qualitative feedback was even more powerful. DocBot had grown to serve 9,000+ users globally, with people saving an average of 15 minutes per search. More importantly, employees were finding procedures they didn't even know existed, improving both compliance and efficiency.
Quantitative Results:
- 9,000+ active users across global sites
- 75-85% accuracy improvement (from ~65% baseline)
- 15 minutes average time saved per query
- $XXM projected benefits over five years
- 500+ participants joined live sessions in just 2 weeks
The business impact exceeded all expectations. When DocBot won 3rd place in the Operations Innovation Awards, it felt like recognition not just for the technology, but for the human-centered design thinking that made it successful.
But perhaps the most meaningful measure of success was hearing from users themselves. One training specialist in Cambridge told me: 'DocBot doesn't just find information – it helps me understand how procedures connect to my specific role and training. It's like having a knowledgeable mentor available 24/7.'
Future Vision
DocBot's success opened doors to broader possibilities. We're now expanding to support seven local languages across seven sites, and exploring how to integrate DocBot into the tools people use every day rather than requiring them to visit a separate application.
Next Steps:
- Language expansion: Support for 7 local languages across 7 sites
- Agentic behavior: Enhanced handling of ambiguous queries
- Platform integration: DocBot as agent in Copilot Studio and MOBI
- Predictive assistance: Proactive SOP recommendations based on user workflow