Walmart - Conversational AI Builder (Converse)

Converse is a conversational AI platform built for voice commerce. It allows developers to build and deploy any conversational application and serve customers on clients such as Google Assistant, Siri, Cortana, etc.

Role

Lead UX/UI Designer

Output

Overall Flow, Prototypes, Figma

Duration

18 months

Hypothesis

Enable non-technical users to create and build voice conversation flows without any coding experience in a matter of days instead of months.

Problem

Non-technical users within Walmart would require engineers to help facilitate setting up a conversation flow and depending on complexity of the task, it could take months to deploy.

Conversation flows today are manually built using JSON code, requiring knowledge of code to set up a flow.

Non-technical users within Walmart would require engineers to help facilitate setting up a conversation flow and depending on complexity of the task, it could take months to deploy.

Conversation flows today are manually built using JSON code, requiring knowledge of code to set up a flow.

Process

Working with product and engineering, our process was simple taking the approach of Define, Discover, Design, and Deploy to develop our product.

The process was far from simple or linear, as each phase required multiple iterations and adjustments. During the discovery phase, we found ourselves reassessing and redefining the design challenge, which allowed us to gain a deeper understanding of the core issue.

Working with product and engineering, our process was simple taking the approach of Define, Discover, Design, and Deploy to develop our product.

Solution

My goal was to concept and design a visual user interface to allow non-technical users (Business, product, UX) to create new and edit conversation flows without having any coding experience.

This experience should be simple, intuitive and easy to use allowing for quick deployment across the Walmart ecosystem. With this in mind, I began tinkering with the idea of authoring the JSON code visually with real time interactive preview to interact and test before publishing.

I began going through iterations of what the interface could potentially look like, gathering feedback from my partners along the way.

My goal was to concept and design a visual user interface to allow non-technical users (Business, product, UX) to create new and edit conversation flows without having any coding experience.

I began going through iterations of what the interface could potentially look like, gathering feedback from my partners along the way.

My initial approach was to visualize the JSON code. Understanding that each intent contained a root, task and response. These elements were placed in a node cards with “connectors” to see relationship to each other.

Node card design presented a challenge as to fitting all the pertinent data necessary for a user to either make connections or choose an intent. Each iteration was tested with users to gain insights as to creating the most optimal design.

With user testing on node card design complete, the designs were finalized and implemented into builder. What we noticed with the node cards space is an issue so the need to collapse them was necessary for a better user experience.

The final design implementation achieved the following high level use cases:

Visualize an existing flow including linkage to others
Ability to visualize the relationship between nodes and edges in the conversation.

Edit a conversation flow
Ability to update the configuration between and within nodes using the visual editor.

Create new conversation flow for an intent
Ability to create a conversation flow for a new intent using existing library of nodes and responses.

The final design implementation achieved the following high level use cases:

Visualize an existing flow including linkage to others
Ability to visualize the relationship between nodes and edges in the conversation.

Edit a conversation flow
Ability to update the configuration between and within nodes using the visual editor.

Create new conversation flow for an intent
Ability to create a conversation flow for a new intent using existing library of nodes and responses.

Key Learnings

Designing a multi modal experience for voice is a challenge.
A level of conversational knowledge will be needed to use the tool. Engineering still will need to play a role in facilitating deployment.
Refinements to the UI can be drastically improved with user research and testing. Getting the tool in front of stakeholders for validation and usability.

As of today the current user base consists of 112 Teams across the Walmart ecosystem with 671 users.

Note: Walmart acquired Botmock 2 years ago. Converse will integrate many of Botmock’s features in the next update.

Designing a multi modal experience for voice is a challenge.
A level of conversational knowledge will be needed to use the tool. Engineering still will need to play a role in facilitating deployment.
Refinements to the UI can be drastically improved with user research and testing. Getting the tool in front of stakeholders for validation and usability.

As of today the current user base consists of 112 Teams across the Walmart ecosystem with 671 users.

Note: Walmart acquired Botmock 2 years ago. Converse will integrate many of Botmock’s features in the next update.

Don Bambico

Don Bambico

Don Bambico