SageMaker Canvas
A no-code Amazon Web Service that creates machine learning data models for use in predictive analytics.
Sr. Interaction Designer | Alpha Design Sprint | 6 Weeks
Problem
Based on a PRFAQ and using their “Working Backwards” design process, Amazon planned to release an alpha product demo of SageMaker Canvas for testing by data analysts who work for a global manufacturer. A style guide and high-fidelity interactions were needed within six weeks.
Solution
Useful for both technical and non-technical professionals, SageMaker Canvas enables customers to automate the machine learning process without code. Via automated processes, the user can import data sources and activate workflows to run predictions on ML models.
Process
• Learn
• Discover
• Define
• Design
• Discover
• Define
• Design
Deliverables
• User Flows
• Wireframes
• Interactions
• UI Design
• Wireframes
• Interactions
• UI Design
Team
• Design Director
• Visual Designer
• Technical PM
• FE Manager
• Visual Designer
• Technical PM
• FE Manager
Tools
• Chime
• Figma
• Slack
• Trello
• Figma
• Slack
• Trello
Discovery
PRFAQ Summary
• Lowers entry barrier to Machine Learning (ML) as a low-no-code service
• Allows data scientists to focus on more complex problems
• Democratizes ML and reduces bottlenecks for data scientists
• Aggregates, compiles and cleans data from multiple sources
Goals + Objectives
The product solves less complex prediction model needs. Simplifying the nine steps of the data modeling experience per the current SageMaker experience was a top priority. My challenge was simplifying the "Import Data" and "Join Data" processes, collaborating with the other designers to make sure that the entire process was seamless and interconnected.
• Get output in front of test customers immediately
• Deliver key experiences for data analysts
• Fits it into Amazon Web Services Ecosystem
• Focus on designing the "happy path”
Metrics for Success
• Functions with other data scientists tools
• Streamlines the data modeling process
• Informs end user via heuristics
• Simple, consistent and polished
User Persona
Data Analyst: Wants to predict business churn. Has the data and needs Data Science support.
Role
• Business Manager
• Operations Manager
• Financial Manager
• Accountant
• Business Manager
• Operations Manager
• Financial Manager
• Accountant
ML Experience
• No coding experience
• Familiar but not technical
• Comfortable with data
• Understands ML value
• No coding experience
• Familiar but not technical
• Comfortable with data
• Understands ML value
AWS Experience
• No experience needed
• Familiar with BI tools
• Works with Data Scientists
• No experience needed
• Familiar with BI tools
• Works with Data Scientists
Functionality
1. SSO Login
2. Home
3. Import Data
2. Home
3. Import Data
4. Join Data
5. Create Model
6. Train Model
5. Create Model
6. Train Model
7. Post-Train Model
8. View Predictions
9. Collaborate
8. View Predictions
9. Collaborate
Features
Data Join
With guidance from Amazon’s Product Manager and Front-End Manager who provided screen references from Amazon Snowflake, I gained an understanding of data joining using SQL. Data joining is the process of combining datasets from multiple sources side by side where at least one column in each data set must be the same. There are four types of Joins: Left, Right, Inner and Full.
With guidance from Amazon’s Product Manager and Front-End Manager who provided screen references from Amazon Snowflake, I gained an understanding of data joining using SQL. Data joining is the process of combining datasets from multiple sources side by side where at least one column in each data set must be the same. There are four types of Joins: Left, Right, Inner and Full.
User Stories
"As a user, I want to join multiple datasets without code and train a new model."
"As a user I want to preview the data in order to validate it's the right data."
Wireflows
Design Challenges
Due to the short timeframe of this project, it was difficult as first to gain and understanding of the data modeling process and SQL and apply it immediately to low-fidelity wireframes for the "Join Data" user experience. There were several early iterations where I missed the mark, including:
Due to the short timeframe of this project, it was difficult as first to gain and understanding of the data modeling process and SQL and apply it immediately to low-fidelity wireframes for the "Join Data" user experience. There were several early iterations where I missed the mark, including:
1. Starting the process from the incorrect screen
2. Focusing too heavily on data visualization
3. Utilizing too many screen transitions
Style Guide
Design Aspirations
Beautiful, Different, Accessible, Extensible, Breathable
Beautiful, Different, Accessible, Extensible, Breathable
The style guide created was based on the AWS design system, and shares some components from its library. But SageMaker Canvas required new components that were simpler, more colorful and engaging. Illustrations were also created for landing pages.
Project Output
• Streamlined primary data modeling features from SageMaker
• Simple, consistent and polished interface
• Completed on schedule
Project Impact
• Solved data prep and model build gap
• 3x faster machine learning training time
• Added to the popularity of SageMaker Studio
• Fits seamlessly into the AWS ecosystem
In Retrospect
Learn and unlearn.
Having spent the previous several years working with more autonomy on projects where more time could be dedicated to user flows and journey maps, it was a welcomed challenge to work on user interactions starting with wireflows.
Having spent the previous several years working with more autonomy on projects where more time could be dedicated to user flows and journey maps, it was a welcomed challenge to work on user interactions starting with wireflows.
Ask the difficult questions.
With the data modeling and SQL coding process, I assumed or tried to figure out technical aspects independently due to the project timeframe and working remotely. I could have saved time and relied more on the Technical Program Manager for insight.
With the data modeling and SQL coding process, I assumed or tried to figure out technical aspects independently due to the project timeframe and working remotely. I could have saved time and relied more on the Technical Program Manager for insight.
Media Links
1. Understanding the Supply Chain Business Model
