Enabling non-technical users to visually train Watson on their data

My role

Lead Designer, collaborating with product management, engineering, and the innovation research lab in Zurich

Timeline

5 months

Outcome highlights

Honored with IBM’s Outstanding Technical Achievement Award, Minor (OTAA) for outstanding contributions involving exceptional technical skill and insight
The tool’s intuitive design and powerful capabilities are a major selling point for the Watson Discovery sales team
Influenced Watson Discovery API to rework some of its existing logistics to make way for a better experience for the user

Skill highlights

Influence - from the beginning, convinced product management and engineering to ground our work in user outcomes
Technical problem solving - I pride myself in making the effort to understand the underlying technology (machine learning and data ingestion in this case) to know how I can best simplify for the user and where I can push back on assumed constraints
Collaboration - involved product management and engineering throughout the design process, as well as got involved in the engineering process by contributing to the API design proposal

The challenge

Up until this point, Watson Discovery wasn’t able to parse unstructured data very well, which prevented users from getting very specific answers when searching their data.

The Watson Discovery product team discovered a technology developed in our research labs that would give us the opportunity to solve this problem. The research technology provided a way to visually train AI to understand documents.

So we set ahead to productize that research technology by adapting it to fit our specific use cases and fit into our technology stack.

Our overarching goal was to enable a non-technical Discovery user to get more relevant search results by teaching Watson about their content for no more than a few minutes. This new capability would be named Smart Document Understanding (SDU).

Design process

Define user scenarios

The first exercise I led the team in was defining use cases and scenarios for Smart Document Understanding; ensuring that our solution would be user-centered and not simply a port of the existing technology. This initial exercise became a touchstone of the design process, resulting in several defining statements that we returned to repeatedly to ground us in user goals.

Benchmark current UX

I also wanted to make sure that we established a baseline on the existing UX, so we performed a heuristic evaluation on the as-is state as well the part of the Watson Discovery flow that the new capability was going to fit into. We used the results of the evaluations to prioritize usability problems and gaps to focus on.

Create ideal state

I decided the next step was to create the ideal states for our outlined scenarios. I did this in the form of a collaborative story mapping sessions. Together, the squad drafted a vision of our user’s future experience to show how our ideas address their current needs.

Contribute to API design

Along the way, I participated in the Smart Document Understanding API proposal, evangelizing for the user. I brought to light an opportunity to rework a portion of the overall Watson Discovery API that was contributing to a poor experience when the user reprocessed their documents with their new changes. With the introduction of the new SDU feature, the pain point would become even more apparent. I articulated my case to the API board and influence a change that would simplify the user’s reprocessing journey.

Design wireframes

After the team was aligned on the end to end flows, I created wireframes to bring these flows to life. I focused on the layout and interaction details. I made sure to build out thorough wireframes for the various end-to-end scenarios, rather than concentrating on static screens.

Validate user experience

As engineering was building out the groundwork, we validated the user experience design with users. I collaborated with a design researcher to define the research goals, and the scenarios and tasks to test with. We conducted several moderated sessions with companies we had established as beta users for SDU. The research findings unearthed issues that we then triaged and addressed.

Design high fidelity UI

After we had the details established for the overall flow and experience, I was ready to nail down the UI visual design. I incorporated guidelines and UI elements from IBM’s design system to ensure users would have a cohesive experience across products.

Results

Smart Document Understanding was released as Beta 5 months after we kicked off the project and went GA 2 months after that.

We achieved our goal of allowing a non-technical user to train Watson on their business documents in just a few minutes. Specifically, for one massive banking client who needed to analyze billing statements, SDU does in 2 minutes what once took 10 days.