Machine Learning Marketplace for AWS
Amazon Web Services launched SageMaker in late 2017 - an AWS service for building, training and launching ML models and algorithms. Within a year of that launch, I needed to design a digital marketplace for developers and data scientists to publish and sell their products.
The tricky part, is that the new ML marketplace needed to sync into the AWS Marketplace where vendors sell many different kinds of products such as: Dedicated software, SaaS products, CloudFormation, and Kubernetes products. The following project was designed working with the SageMaker team to ensure a seamless transition to publishing from their SageMaker service portal into AWS Marketplace.
The tricky part, is that the new ML marketplace needed to sync into the AWS Marketplace where vendors sell many different kinds of products such as: Dedicated software, SaaS products, CloudFormation, and Kubernetes products. The following project was designed working with the SageMaker team to ensure a seamless transition to publishing from their SageMaker service portal into AWS Marketplace.
The team
Consisted of the SageMaker product manager, his team of about ten engineers and myself, senior UX designer. The engineering team was also supporting the buyer side of the Machine Learning Marketplace.
The plan in steps
Step 1: Define and align
Step 1: Define and align
Where does SageMaker end and AWS Marketplace Management Portal (AMMP) begin? I must create a seamless integration across AWS services for our users, and work with the SageMaker UX designer to include an ingress point, and call to action to link to AMMP.
Step 2: Set expectations
Create a publishing flow to communicate the end to end functionality from the user's perspective. Include all steps involved in publishing: Happy path, possible issues, session expiration, communication triggers, fraud checks, and more.
Step 3: Redesign the information architecture
Adding the new product type is part of a phased release of the entire Seller's Marketplace. To support a division of product types changes to main navigation were required. I planned a redesign of all of the site to implement the new style guide as defined by the AWS branding guidelines.
I reorganized the order of importance in the navigation bar, designed a new logo for the site, created new nomenclature, and added new standardized UI elements. The new products page is consistent across each product type.
After the redesign
Step 4: New Product Overview page
Vendor need to see the status of various product components that make up a product listing: Marketing details, link to software component, and the offer (billing details).
The new Product Overview page uses a card approach to group the various actions a user can take into groups, thus reducing the cognitive load.
New product overview page for a ML product
The Overview page allows for all kinds of feedback to the user. Any errors, missing information, access to software versioning... All available from one quick scan of the page.
Step 5: Deliver all forms required for publishing
Every existing form for all product types were redesigned to meet current information design and UI design standards. Proper visual cues, hints, progressive disclosure and visual groupings were given to convey the information required to complete the form.
Every existing form for all product types were redesigned to meet current information design and UI design standards. Proper visual cues, hints, progressive disclosure and visual groupings were given to convey the information required to complete the form.
The #1 request from our users for the redesign of our publishing flow on AWS Marketplace was to have insight into the publishing process. It could take days (or sometimes weeks!) for a product to go through the verification process and become available for customers. I designed a status bar, with SLA expectations, into the Product Overview page.
Results
The ML Product listing has been a massive success. The AWS Marketplace offers 665 machine learning models and algorithms (as of this writing). Data scientists, developers and product owners enjoy the simplified experience to publishing and editing their products. The ML publishing flow was the first product type to use the proposed UI, and we are currently using it to roll out the redesign for every other product type.
