DataRobot AI Platform redesign

Soon after joining DataRobot in April 2022, we kicked off an ambitious redesign of the platform. Our task was to transform a legacy startup application into a modern, scalable enterprise SaaS platform, poised for DataRobot's next chapter.

Leading the work with our design team, we conducted research, collaborated with PM, engineering, and leadership to define the vision and then deliver it to production. This unfolded amid tech-industry turbulence, company layoffs, a Ukrainian design team dealing with war and the generative AI storm that disrupted everyone’s plans.

Context

DataRobot pioneered the automated machine learning (AutoML) industry, making it easier for data scientists to build and deploy AI models. When I joined in 2022, the information architecture (IA) was under strain from new features added over time. Customers were simply finding it harder to realize value—and our competition was catching up. In addition, a push towards a self-serve SaaS model meant significant changes were needed to the experience if we were to stay relevant and grow.

In March 2023, we announced one of the biggest updates to date — a complete overhaul of the platform UX called DataRobot NextGen. The experience is built from the ground-up to better meet the needs of data-scientists, software developers and ML engineers when experimenting with predictive and generative AI models, and getting them into production.

DataRobot machine learning platform Experiment screen

Challenges with old design

Information architecture limitations
Too many features across a single organizational layer meant users struggled find what they need, or even understand what the platform could do.

Knowing what to do next
Limited asset organization and workflow guidance meant users found it difficult to manage their assets or know what to do next at each step in the flow.

Sales-led model limited growth opportunities
New users were dependent on 1:1 training in order to onboard to the platform, limiting our ability to scale. Enabling users to discover value independently was essential, making the shift to a self-serve SaaS onboarding model crucial.

Redesigned ‘NextGen’ experience

‘Classic’ experience

How we worked

  • Discovery user testing

    Qualitative and quantitative research (using the jobs-to-be-done framework) with end-users and internal customer representatives to understand challenges and pain-points early on. This laid the foundation for the new vision.

  • Describing the system

    Collaborative cross-domain workshops to synthesize research, create alignment and explore high-level information architecture concepts in order to understand how we might evolve and re-think the platform experience.

  • Exploring & validating solutions

    Early concept exploration was an important part of the ideation process. Quick concepts brought research to life and enabled our team to quickly validate directions with internal teams and external users.

  • Align on who we are solving for

    Highly collaborative cross-domain sessions to synthesize research outputs and create high-level information architectures in order to understand how we might prioritize and re-organize features and evolve the experience overall.

  • Validate and iterate solutions

    Qualitative and quantitative research with end-users and internal customer representatives across the company to understand challenges and pain-points early on and then continuous validation testing throughout.

  • Execute on the detail

    Early concept exploration was an important part of the ideation process. Quick concept brought research to life and enabled our team to quickly validate directions with internal teams and external users.

Design principles

Design principles express our shared vision for what makes a great product and what the experience should feel like to the end users. These principles served as decision making guides for our organization, keeping us aligned while driving consistency across the platform as it evolves.

01 A robot, with personality

DataRobot is direct, pragmatic and impartial in how it generates and presents information but it delivers insights with warmth and humanity. Our tone is that of a friendly professor—recognized for expertise and intelligence but yet humble and approachable. This means we use accessible, natural language where possible and we use visuals to inject the brand personality where appropriate.

03 Bias for action

DataRobot is fundamentally a decision-making tool that enables users to solve complex business problems. Everything the product does should help them move forward and get closer to achieving these goals. This means being action-oriented, providing just the right amount of information to make informed choices and making the next steps or options in the journey as clear and unambiguous as possible.

02 Approachable first, then flexible and powerful

DataRobot is for technical users who are comfortable with the inherent complexity of machine learning. Our task is to streamline this complexity, enabling both high and low maturity users to swiftly master the functionality at a pace suited to their experience. While we offer multi-modal interaction, our initial focus is on ease of use, gradually introducing more advanced features and flexibility as users progress.

04 Relentless focus on clarity & context

We don’t strive to over-simplify but we do aim to make the process as clear and efficient as possible. This can mean adopting industry standard language and patterns to minimize friction in the experience. We also help users of all levels to maintain a clear sense of their context in the overall journey journey through clear wayfinding points and simple navigation. We are often breaking new ground however and continue to develop new approaches when it makes sense to do so.

NextGen experience overview

Intuitive navigation informed by key workflows
UX driven by persona-based workflows: Workbench for experimentation, Registry for ML governance and Console for ML operations—catering to how users use different parts of the platform.

Hub for experimentation and collaboration
Workbench Use Cases helps users to organize their assets across both predictive and generative AI workflows.

Intuitive onboarding to enable product-led growth
Tailored onboarding experiences help users with different learning styles see value quickly. Features like the LLM Playground make it easy for users to see the value through exploring, testing and tuning LLMs for production.

Feedback & impact

Since this initiative launched in March 2023, we have made a lot of progress in delivering this into production, with lots of iteration along the way…

"This is like the most beautiful, well designed ML tool I've ever seen. It feels like this is clickable ML. Everything felt super intuitive, it felt like exactly like where my mind is."

— User testing with external participant

"The redesign reduced the project creation time from 54 seconds to 19 seconds! The #1 positive feedback we got was consolidation of all actions into one clear CTA."

— Summary of Maze user testing

"This is like the most beautiful, well designed ML tool I've ever seen. It feels like this is clickable ML. Everything felt super intuitive, it felt like exactly like where my mind is."

— User testing with external participant