Watson Discovery

Beginning in March 2019, I had the privilege to join the Watson Discovery design team following my graduation from IBM’s interdisciplinary Patterns bootcamp.

Watson Discovery is an AI-powered application that uses text analytics and natural language processing to allow users to create robust enterprise search engines. Companies upload their massive bodies of data and then utilize Watson Discovery to analyze and find answers to complex domain-specific questions.

My Role & Objective

I took on a UX Design role in this project. I coordinated with our product owners and development to democratize the current user experience and transform it into a product that could allow non-technical Data and Business Analysts to build custom AI-powered document search applications.

The Challenge

Beginning in late 2018 the Watson Discovery design team visited clients in Japan in an effort to understand the needs of business users. Designers sat side-by-side with the employees to see their current workflows firsthand.

We were primarily concerned with how to democratize and simplify the current user experience and transform it into a product that could allow non-technical Data and Business Analysts to build custom AI-powered document search applications.

Identifying Issues

Through multiple rounds of workshops, UX sketching, and user journey mapping with sticky notes, we identified fundamental issues within the current version of Watson Discovery. User pain-points included long, inconsistent times (anywhere from multiple hours to several days) for existing users to begin testing their search. In addition, the build interface looked markedly different from the end-user search interface, making it harder for first time users to see the direct impact and value of their work.

Distilling Complexity

To remedy these problems and enable these non-technical users to more quickly build a robust internal search engine for their company, we decided on a "What You See Is What You Get" (WYSIWYG) interface methodology. This involved translating powerful, complex IBM technologies like natural language processing, understanding, and visual-recognition into easily understood, tangible features like field identification, relevancy training, and domain-dictionary creation – all of which will be explained in the following user journey.

The Scenario

From the beginning of 2020, the CDC has been on top of everything COVID-19 related. Tanya is part of a large team of analysts and researchers responsible for finding, collecting, storing, and analyzing the vast quantity of news, scientific journal publications, infection rates, and hospital capacities among many other things.

Lucky for Tanya, Watson Discovery is here to help quickly consolidate her many disparate data sources. She creates a new project to begin.

Tanya is presented with three project types upon entering the product. She selects ‘Document Retrieval’ since the CDC has tasked her with building a search engine to enable their scientists and analysts to evaluate and search through a constantly increasing mass of document and website data.

Here, Tanya is able to upload or connect her data from a selection of different places. Everything from secure private databases to continually refreshed website crawls is available to her. She uploads a selection of recently published scientific articles.

After uploading her data she is then prompted to use Field Identification to highlight key elements within a small selection of her documents. Tanya tags a few headers, paragraphs, and images. In doing this, Watson Discovery can accurately replicate the highlighted patterns across Tanya's entire corpus of data after only a few pages of her labeling.

Once her data is appropriately labeled Tanya can now navigate to the Project Preview. Here she has a visual way to build, test, and improve the CDC's AI-powered search. She wants to ensure that her search engine returns the best answers to her end-users so she runs a test query to see how it performs.

Once she enters her first query, Tanya sees some tools she can use to improve the quality of her project. She likes the idea of providing her end-users with ways to more easily filter their search. She decides to click on “Create facets” in the Improvement Panel on the right.

After creating a few facets for her teammates, Tanya navigates to "Dictionaries" is interested to see what terms and concepts Watson Discovery has extracted from her data corpus. She sees existing terms that she is currently using on the left and new recommended ones on the right.

After customising her dictionary, she sees that there is still more she can do to improve the quality of her search engine. She wants to improve the relevancy of her search results so she clicks on "Relevancy training" in the panel.

Tanya's Outcomes

After getting a taste for how to test her search engine and improve it via various improvement tools, Tanya is confident enough to send out a test link to her teammates. As time goes on, Tanya and her team at the CDC will have a robust AI-powered search engine that can transform and adapt to fit a constantly growing body of COVID-19 related data.

Overall Impacts

Our team managed to take complex tools that were initially designed for developers, data scientists, AI engineers, and translate them to a product that could enable non-technical business users to apply their domain knowledge and improve collaboration by building alongside their technical counterparts.

Our team's work led to improved time and cost savings and increased revenue through reduced employee workloads. In collaboration with our developers, we empowered customer employees to use cutting-edge natural language processing to spend less time searching for information and more time to act on valuable insights.