Cynthia Yue

Demystifying Natural Language Processing

IBM's Embedded Business AI Framework

EBA — for short — is an AI framework used to build natural language processing assistants.

One example of a solution can be an assistant that enables a digital marketer working at an ad agency to quickly pull reports from multiple channels (i.e. Google ad network, Facebook ads, Twitter ads…) using simple dialogue such as: “What is the ROI from Q4’s digital advertising campaigns?” It can also provide recommendations as to where to optimize running campaigns by looking at past data.

In many cases, employees who rely on data to do their jobs spend disproportionate amounts of time on low level work. This can include conducting ad hoc analyses on large data sets where data is manually filtered, combined and cleaned. These individuals are found in every industry — finance, logistics and healthcare to name a few.

IBM EBA addresses these inefficiencies by allowing organizations to build solutions that make it possible to gather necessary data through simple dialogue, and methodically reach their goals through AI powered recommendations.

My Role
UX Designer

The Team
3 UX Designers
1 Visual Designer
1 Content Designer
2 Offering Managers

Deliverables
Research Findings
User Personas
Med-Fi Wireframes
Go to Market Strategy

Tools
Invision
Sketch
Whiteboards
Keynote

Project Summary

This was a part of my 6-week incubator project at IBM’s Austin Design Studios. During this process, our team of 7 received extensive coaching from seasoned designers and engineers. This also showcases my first project utilizing IBM’s Enterprise Design Thinking framework!

The Problems

While the idea of being able to provide our data reliant employees with quick access to their data sounded great, there were some huge issues with the platform.



Our Solution

Our Process

Domain Knowledge

back to table of contents

Stakeholder map

To start, we spoke with EBA engineers and product managers to understand the product space. To add clarity to our conversations with the product team, we mapped out key players' relationships to one another in a stakeholder map.




Work allocation

Through desk research we were also able to validate a strong need for this product. Employees who rely on data spend a disproportionate amount of time on menial tasks.






Key Terms

User Research

back to table of contents

Interviewed:

7

Software developers

2

Decision makers

2

End users

Key findings

Existing issues with EBA

Software developers expressed their frustration with the EBA platform:

"We had to work pretty closely with the [original EBA] development team… just to understand how we were expected to program with the library provided."


--


"We had to code a pattern… for every possible dimension and metric… we obviously weren’t going to do that for hundreds of metrics and dimensions manually."

We also learned from decision makers that our marketing pages were lacking:

"Well, I don’t know what it [EBA] does..."

User Personas & Flows

It became clear that there were 3 users we needed to account for in our design.

Empathy maps have us write out in detail what our users say, think, do and feel.

User 1: The "End User"

User 2: The "Decision Maker"

User 3: The Software Developer

Align on the Problem(s)

back to table of contents

Problem Statements and Project Hills

To align as a team and focus our efforts, we wrote problem statements and Hills. Hills are used at IBM to turn users’ needs into project goals, helping our team align around a common understanding of the intended outcomes to achieve.

Problem Statement and Hill 1

Problem Statement

It was difficult for decision makers to understand what EBA can do.



Hill

A decision maker can discover EBA, recognize how it will improve efficiency of his/her practitioners, and within minutes, want to integrate it into their ecosystem.

Problem Statement and Hill 2

Problem Statement

Most organizations don’t readily have NLP engineers at their disposal, and NLP is generally difficult for engineers to learn.



Hill

Any developer can discover and build natural language AI assistants for their business without possessing any prior knowledge of NLP.

Problem Statement and Hill 3

Problem Statement

Long ramp-up time as engineers are typically required o build solutions from scratch.



Hill

Any developer can learn to build and deploy an AI assistant in half the ramp-up time and without having to contact IBM support for assistance.

Design

back to table of contents

With our problem statements and hills clearly highlighted, it was finally time to start designing a solution.

"Big Ideas" -- an ideation exercise used at IBM.

Plotting our Big Ideas on a Prioritization Matrix.

"To-Be" flow

Final Solution

back to table of contents

Hill #1

A decision maker can discover EBA, recognize how it will improve efficiency of his/her practitioners, and within minutes, want to integrate it into their ecosystem.

Alan, a decision maker can now clearly see what EBA does and how it works. He can also see the main features that EBA supports including, the Marketplace, Developer Lab and Testing Lab. If Alan is still unsure, our new EBA page provides personalized demos for Alan to view and tinker with.

Hill #2

Any developer can discover and build natural language AI assistants for their business without possessing any prior knowledge of NLP.

Taylor, a software developer can now hit the ground running with an in-app NLP tutorial. This teaches NLP from a high level that’s sufficient enough for Taylor to start coding a solution within the EBA platform.

Hill #3

Any developer can learn to build and deploy an AI assistant in half the ramp-up time and without having to contact IBM support for assistance.

Now equipped with NLP skills, Taylor can start developing a solution for Tim, an end user. Rather than needing to start from scratch, she is able to build on up from templates provided for a variety of common use cases. Our end user, Tim, can now use this in-house solution to not only complete his menial tasks more accurately and efficiently. This leaves more time for Tim to harness EBA’s AI capabilities to methodically enhance his campaigns even further, and deliver stronger results for his clients.