ElancoGPT

A generative AI chatbot powered by OpenAI's GPT-3.5 and Google's PaLM-2 large language models.

Background

Elanco, formally known as Elanco Animal Health Incorporated, is a global company that develops products and services to improve animal health and protein production. Established as a division of Eli Lilly and Company in 1954, it became an independent company in 2018 following its initial public offering. Elanco operates in the pharmaceutical sector, focusing on both companion animals (pets) and livestock. The company's portfolio includes a wide range of products, including vaccines, therapeutics, feed additives, and other health-related products designed to prevent disease, improve health, and increase the production efficiency of animals.

Problem Statement

A newly formed enterprise engineering team at Elanco is tasked with adopting and integrating generative AI tools to enhance enterprise support and decision-making processes. The team needs to design a conversational interface that can handle multi-modal inputs, offer customizable options, and deliver interpretable outputs. Leveraging advanced models like OpenAI’s GPT-3.5 and Google’s PaLM-2, the solution must be platform-agnostic, providing a consistent and accessible experience across Mac OS, iOS, Android, and Windows devices. The challenge lies in ensuring seamless integration, user adoption, and the effective use of these generative tools to meet enterprise needs. I was responsible for designing this innovative solution.

Outcome

I designed the experience for ElancoGPT, a responsive AI chatbot powered by OpenAI's GPT-3.5 and Google's PaLM-2 large language models for Elanco (formerly Bayer Animal Health).

Project Details

My Role

Senior UX Designer

Timeline

4d Design

10d Development

Team

Enterprise Engineering Team
Mary Bartlett
Charlie Hammerston-Budgen

Tools

Figma
TailwindCSS
TailwindUI
GitHub
Codepen

Discover

Stakeholder Interviews

In order to understand the user, I conducted several interviews from various functional groups including marketing, business, and enterprise engineers to determine their needs, expectations, and any problem areas that they encounter.

Competitive Analysis

I performed a competitive analysis of conversational and generative tools like Discord, Microsoft Teams, ChatGPT, and Midjourney to explore conversational interaction flows, unique features and solutions for a multi-modal experience (uploading documents, spreadsheets and images).

Design and Developer Collaboration

I met with the enterprise engineering team to review the initial user flows, reusable components, and wireframes.

Define

MVP

If we build a generative chatbot to achieve routine or repetitive business tasks then stakeholders can accelerate productivity and streamline their workflows.

User Needs

• Answers (Research Studies, Guides, Articles)
• Ideation (Brainstorming)
• Content Creation (Documents, Policies, Processes and Communication)
• Content Review (Literature, Journals, Presentations)
• Summarization (Executive Summaries, Key Messages/ Findings, Reports)
• Research (Qualitative Analysis, Observations, Deductive Reasoning, Survey Qs)
• Software/ Data Engineering (Code Interpreter)
• Translations (Multi-National Content Creation)

Primary & Secondary Users

1º      Marketing, Animal Sciences & Business Stakeholders

2º     Enterprise Engineering Team

Develop

Implementing responsive design using TailwindCSS breakpoints

The enterprise engineering team used TailwindCSS and TailwindUI for experimental projects. Because I worked on the component library, I decided to use the grid system provided by TailwindCSS to create a more consistent look and feel. 

Deliver

Key Interactions & Components

Having a conversational interface as the initial MVP allowed me to focus on key user interactions and UI components focusing on navigation, icons, user initiation, response, queries/ prompts and system notifications (alert, warning, confirmations).

Mobile

These images depict key user interactions such as user initiation, prompting, response, and common system alerts. 

Tablet

Web

First Iteration

Key Learnings

Non-technical stakeholders vs technical stakeholders have different needs respective of their functional groups and tasks. 

Being a "fly on the wall" at the early stages has its benefits in determining needs that might not be vocalized during the stakeholder interview.

Maintenance is a huge factor when making UI changes, modularity of components is important to achieve a unified look and feel without exhausting engineering resources. 

A picture is worth 1,000 words. A click-through protoype will speak volumes and aid in communicating design decisions from high level to small details.

Align with stakeholders early and often even before discovery begins. Having them be a part of the process at it's inception helps build rapport, trust and buy-in for future design decisions.

Raf Laus, Portfolio. https://rlaus.xyz. © 2024