IBM Watson Discovery:

Designing IBM’s AI Search Tool From Zero to One

Developer tools · AI · 0→1 · 3 min read

Client
IBM
Role
Interaction → Lead Product Designer
Timeframe
Aug 2016 – Apr 2019
Type
Developer tool, AI
  • 0→1launched Dec 2016
  • 3enhancements shipped
  • ESPN · H&R Block · Salesforceamong adopters

Disclaimer

The following case study is personal and does not necessarily represent IBM’s positions, strategies or opinions. I have omitted and obfuscated confidential information.

Summary

Overview

Watson Discovery takes messy data like documents and web pages, structures it, and uses AI to find what’s inside. It launched publicly in late 2016.

The problem

Tools for building with AI tend to be either too complex for most people or so simplified they lose their power. We needed an interface that kept the power and made it usable.

What I did

As lead designer from launch onward, I set up a continuous design loop with two other designers. Then I led the design of three enhancements that fixed post-launch usability issues: Insight Cards, Schema Explorer and Knowledge Graph.

Impact

Discovery reached general release and was adopted by enterprises including ESPN, H&R Block and Salesforce.

Who I brought togetherSquads of offering managers, developers, QA, researchers, designers and writers, senior Watson design leaders, and a research team in another city, all working from the same user needs.

Sticky notes mapping ingestion, enrichment and insights
Sticky notes mapping ingestion, enrichment and insights.

The Challenge

Intelligent systems are complex, so the tools for building them often are too.

Our goal was a tool developers of any background could use to build their own intelligent apps.

The Loop

Squads & workshops

Several squads were combining many microservices into one tool. Two other designers and I ran workshops so everyone could learn the technology and focus on user needs.

Rapid prototyping

Sketches became wireframes, then clickable prototypes.

Usability testing

Our researcher tested with real users every few weeks, and the whole team was invited to watch.

Together these became a continuous cycle of observe, reflect and make, based on IBM Design Thinking. It carried us to public release on December 15, 2016: users could upload data, enrich it with AI-powered APIs, normalize it to JSON and query it from their own apps.

Paper prototype of a data collection screen
Paper prototype of a data collection screen.

Enhancements

Insight Cards

After launch, users didn’t understand what the APIs did to their data. Insight Cards show key entities, concepts, keywords and sentiment as data is processed, and drew some of the strongest user feedback.

Schema Explorer

People couldn’t tell what their data looked like after processing, so querying was hard. Schema Explorer, born in a workshop with senior Watson design leaders, shows the structured fields and lets users try sample queries.

Knowledge Graph

Knowledge Graph links information across documents. It came from a research team in another city, so I ran remote workshops to agree on a user story and plan the integration.

Sketch of insight cards showing entities and sentiment
Sketch of insight cards showing entities and sentiment.
Schema Explorer.
Schema Explorer.

The key decision

Hide the complexity behind a simpler tool, or show what the APIs do to your data?

The options

We could hide the complexity behind a simpler tool, or keep the full power and help people understand what it was doing.

What we chose and why

Explain it. Insight Cards and Schema Explorer show what the APIs do to your data without taking any capability away.

The trade-off

More concepts for new users to learn, which is why we showed insights in real time as data was processed.

Impact

Watson Discovery went from a brand-new idea to a product powering apps at companies like ESPN, H&R Block and Salesforce.

3

major enhancements shipped after launch