Pattern induction

Enable non-technical users to extract structured data from unstructured documents

ROLE

UX Designer

TIMELINE

2020

PROJECT SKILLS

AI, E2E Design

OVERVIEW

I designed, developed, and delivered a beta version of the pattern induction tool in 2.5 months that contributes to helping companies decrease 75%  of time spent reading through data sources.

USER PROBLEM

Business analysts within organizations often need to extract specific information from large volumes of unstructured documents, but existing approaches required technical expertise.

Challenges with existing processes include:

  • Information appears in different formats across documents

  • Business analysts know what they’re looking for, but not how to define rules

  • Regex and model training require engineering support

  • Manual review is slow and error-prone

This created a bottleneck where business analysts depended on data scientists to build custom extraction logic.

These project goals ultimately help increase our product’s value proposition while helping our enterprise clients reduce costs.

PROJECT GOALS

How might we design this feature in a way that non-technical users can easily train the system to extract patterns that pertain to a certain set of rules?

CONTEXT

Watson Discovery is an AI search product that helps enterprises find answers and trends within their unstructured data using natural language processing, machine learning, data mining.

SOLUTION

Pattern induction is an enrichment that uses natural language processing to recognize and extract patterns out of data 

THE PROCESS
SCOPING

Timeline was already tight and we spent a lot of time defining scope for MVP. For example, 

TRANSLATING USER NEEDS TO USER FLOW

Users needed a way to:

  • Extract custom, domain-specific fields

  • Work across messy, inconsistent documents

  • Avoid writing rules or training models

  • Iterate quickly with immediate feedback

ITERATIONS
→ GIVE USERS CONTEXT & A WAY FORWARD

Inspired by the Net Promoter Score and benchmarking study

→ FAILING FORWARD

Phases can’t be independent of each other

SEEKING VALIDATION

We sought feedback from both internal employees and existing customers. We got generally positive results with room for improvement.

Most qualitative feedback pointed to the fact that it was ambiguous as to what steps to take after identifying a few examples. This was good feedback for the next iteration of this tool.

“There is a lot going on on the screen. I’m not sure where on the screen I should be spending most of my time working”

LOOKING BACK
IF I HAD MORE TIME

If we had more time, below is the ideal experience I would have designed:

If we had more time, I’d also add a few changes to the design process. I would

  • Take a look at analytics to see where users are having trouble within the flow

  • Get a better understanding of the current user’s use case and pain points

  • Do more iterative design and usability testing

TAKEAWAYS & LESSONS LEARNED

  • Learned to work quickly under timeline constraints

  • Learned new technologies

  • Simplified complex workflows