.png)
Cnvrg.io by Intel
Optimizing Flows creation for an AI/MLOps platform by reducing friction and enhancing usability
%201.png)
Flows
Overview
Redesigning Flow Creation: Improving Usability and Efficiency for ML Teams
Cnvrg.io is an MLOps platform that helps data scientists and ML developers build, run, and manage models efficiently. It offers tools for ML pipeline management, automation, collaboration, and infrastructure, aimed at Data Science, DevOps, AI teams, and product managers looking to streamline processes and deploy models to production.
Flows in Cnvrg.io allow the construction of complex pipelines using a drag-and-drop interface, simplifying the process of moving a model into production.
Problem
Too Many Support Tickets is a Red Flag
When we noticed a large volume of customer support tickets related to Flow usage, it was a clear warning sign that we needed to take action and improve the creation process on the platform. I spoke with our support team to understand the frequently asked questions from users regarding Flow issues, in order to identify potential frictions.
I found that many users struggle to create a new Flow, which leads them to reach out to our support team for assistance.
Research
Understanding the Users
User interviews
I approached the research phase to understand the current state of Flows and the needs of our customers.
I conducted user interviews to gain a deeper understanding of their challenges and pain points during the Flow creation process.
User quantity and profile: I spoke with 6 users – half of them were advanced and experienced in the field, and the other half were relatively new with basic technical knowledge.
Key Insights from the interviews
Data analysis
-
40% of users did not identify the "New Task" button as the primary action.
-
47% of new users abandoned the Flow creation process halfway through due to a lack of guidance.
-
20% of users reached out to technical support due to difficulties in creating or connecting tasks within Flows.
Mapping pain points
I mapped out the pain points of the current Flows page, based on support tickets, user interviews, and data analysis. These are the main issues I have focused on:
-
Different Skill Levels - One of the main issues with the Flows page stems from the complexity of the interface. Experienced data science users sought more flexible and advanced options, while i identified a new persona with less advanced skills requiring simpler and faster processes. This gap in skill levels creates frustration and barriers to using the system effectively.
2.
Unaligned Mental Model - Analysis revealed that users are not clicking on the "New Task" dropdown, indicating that they either missed it or misunderstood that it offers additional options. This mismatch between users' mental model and the interface design prevents them from utilizing the available functionalities.
3.
Overwhelming Initial Choices - Users who clicked on create "New Task" felt overwhelmed by the wide range of options. This leads to cognitive overload and slows down the workflow, particularly for new users who are unsure which options are best suited for them.
4.
Lack of Guidance for Beginners - New users feel uncertain when starting from scratch without clear direction or examples. They need guidance to help them understand how to use the various features and create Flows effectively.
Design Principles
User testing (internal+external)
First, I defined the important issues we need to examine in our new design and the correct way to test whether the solution is good enough.
Next, I created a script, in which we walked through our user within the prototype. In every few tests, I made changes in my design, following the conclusions we had. Next, we continued with an updated version for future tests and tested the improved solutions on new users.
UX improvements I wanted to test
Equipped with insights from user interviews and analysis of pain points, as well as considering the proposed solutions, I began to contemplate the changes and improvements I wanted to test in the new design of the Flows page:
-
Redesign of the "New Task" button - Transforming it into a primary and clear button, with a dropdown menu, and examining the impact on user behavior
2.
Presenting skill-level adapted options - Testing the effectiveness of displaying two options in the empty state: 'Example Flows' for beginners and 'Create your own Flow' for advanced users.
3.
Adding 'Example flows' - Evaluating the impact of adding examples of common flows on users' understanding and ease of system use. Additionally, we explored the optimal way to present a large amount of information, including descriptions and requirements for each example, to help users choose the most suitable example for them without overwhelming them with information.
4.
Strategic placement of entry points - Testing the efficiency of positioning 'Example flows' buttons in both the empty state and the feature header.
5.
Implementing progressive disclosure - Implementing a system where advanced options are initially hidden but accessible when needed, reducing cognitive load for new users while maintaining flexibility for experienced users.
UX & Design
Following numerous user testing iterations, copy revisions, and UX refinements, we developed an improved version of the Flows page.
1.
New Task Button
I transformed the existing 'New Task' dropdown menu into a prominent, clear button that directly offers task creation options. This change makes the primary action more accessible and visible to users, streamlining the process of starting a new Flow.
2.
Empty State Options
The empty state now presents two clear options:
a. Example Flows: Allows beginners to choose from existing templates, helping them understand the structure and function of Flows.
b. Create your own Flow: Suited for advanced users who want to build a unique Flow from scratch.
Flows page - Empty state new design
3.
Example Flows
I introduced 'Example Flows' for common processes, enabling less experienced users to intuitively understand the feature and easily set up ML pipelines. This addition bridged the gap between beginners and advanced users, providing a smoother onboarding experience.
4.
Strategic Entry Points
I identified that adding clear entry points for Example flows would improve accessibility and convenience. I positioned these entry points in both the empty state and the feature header, making the functionality accessible to users both when starting a new process and while working on an existing Flow.
5.
Example Flow Button in Header
I added an Example flow button in the action panel (header) as an additional option. This button is defined as a secondary, independent button, without direct relation to other task actions, designed for users who want to learn from an example.
6.
Visual Hierarchy Redesign
I reorganized the layout to help users better understand the relationship between different elements in a Flow. I used colors and icons to differentiate between various task types and emphasize the workflow.
Flows page - new design
Initial Results Analysis
What I Learned
-
Less is more - Keeping things simple and clean, even in the wording, makes a huge difference for users. We tested a few versions with people in the office, and it was amazing to see how small changes impacted resolution time for the same tasks.
-
Testing is everything - When you're deeply involved, it’s hard to see the full picture. Watching users test the product revealed weak spots and unclear instructions.
-
Priority is Key - Sometimes it's worth letting go of minor details to focus on the bigger picture and core project goals.