Analytics Platform — Shape your ML models with ease
AI, Machine Learning, Models Configuration
MY ROLE
UX/UI designer
I was responsible for research, design strategy, ideation, user flows, visual design, interaction design and prototype.
RESULT
MVP: SaaS platform where ML models can be created, configured and changed.
DESIGN IMPACT
Team efficiency rose by 30%.
Dependency on Data Scientists declined by 60%.
COLLABORATED WITH
Data Scientists
Developers
Product Owner
Project Manager
TOOLS USED
Figma
InVision
Zeplin
Adobe Illustrator
PROBLEM
BUSINESS GOALS
Expand client base
Increase customer satisfaction to retain clients
ROADBLOCK
We could not expand the client base because of the deep involvement of Data Scientists with every customer. We either had to hire more people or remove this bottleneck so we chose the latter.
ALL SCREENS
USER NEEDS
EFFICIENCY
Changes can be made quickly with fewer steps
CLARITY
User understands what system does and how it works
ROBUST SYSTEM
No change by user can break the system
DESIGN STRATEGY
UX GOALS
Simple and clear design
Error prevention
UI GOALS
Follow existing Design System
HOW IT SUPPORTS BUSINESS GOALS
By taking out the middleman we removed the bottleneck that allowed for more customers to be served by the same team.
Customers no longer had to wait for Data Scientists to make every update for them. Clients can quickly and easily change parameters themselves, increasing their satisfaction.
Remove bottleneck to:
Serve more clients
Increase satisfaction
CHALLENGE 1
WHAT DOES THE SYSTEM DO?
At the start of this project I did not know anything about the ML models configuration system. It was not my first time working with Data Scientists though. So I went directly to the people who created the underlying machine learning system. They worked me through it.
I carefully drew diagrams to visualize dependencies. Each diagram was reviewed by Data Scientists to ensure I got it right and nothing was missing. It took us a few rounds to nail it.
Through this process I learnt how the existing system worked and noticed some weak points that could have caused issues for users.
WEAK POINTS
Complexity
Interdependence of parameters
Specialized terminology
CHALLENGE 2
WHO SETS UP THE SYSTEM?
The major weakness of the ML models config system design was its complexity. As much as I tried to simplify it, there was no work around a long and tedious first run of the system. New Organization User Flow shows for many steps a user has to go through to set up a new organization.
User testing showed that users could not go through the setup process on their own. Even with onboarding guidance, the error rate was 30%.
SOLUTION
Set up every new organization internally
NEW ORGANIZATION FLOW CHART
CHALLENGE 3
HOW TO PREVENT ERRORS?
Complexity also led to another issue — every parameter had to be in place for the system to work. Take one out and everything falls apart. So the huge challenge was to ensure that errors were prevented instead of fixed afterwards.
SOLUTION
Use modals for every new parameter creation
SEPARATION
Every parameter creation process is separate from all other activities.
NO INTERRUPTIONS
User’s attention is fully focused on the current task.
If the process was interrupted, no data was saved.
ERROR PREVENTION
A user had to go through the whole process to add any new parameter so no faulty ones with missing information could be created.
FEATURE ENGINEERING STRATEGY CREATION PROCESS
