From configuration maze to a guided model-design canvas
I consolidated a multi-screen model-configuration workflow into one guided canvas, helping analysts understand dependencies, start with a runnable setup, and iterate independently.
The guided canvas: target definition, table relationships, and validation in one workspace.
Users could complete steps—but not predict their impact
Creating a model task required users to import data, define schemas, select a target, connect tables, and configure advanced settings across separate screens.
Users could learn the workflow. What they could not see was how one setting affected the next—for example, how target selection changed required relationships or how table connections affected the model.
“I don’t know what these settings will affect next.”
Model design is an iteration loop, not a linear checklist
The issue was not simply where to begin. Users needed to understand the relationships between their decisions, run a first version with confidence, then return to adjust it without relying on a data scientist.
So I replaced the accordion-style flow with a single canvas. Configuration panels keep the work organized, while the canvas makes target tables, source tables, and their relationships visible together.
Disconnected configuration steps became one canvas users can return to and adjust.
Visibility alone was not enough
The first canvas made data relationships visible, but testing showed that some users still did not know how to begin. They hesitated over whether tables needed connecting and how to choose a target.
That changed the direction: the canvas needed to preserve its global view while also giving users a credible starting point.
A canvas with a confident starting point
The final canvas keeps the whole data model in view and opens with a recommended setup rather than an empty plane. Users can run a first model from it, then return to the same canvas to refine it.
Target tables, source tables, and their relationships, visible together on one canvas.
Auto-schema suggests an initial structure from sample data.
Auto-connect proposes table relationships so users do not start from a blank model.
Suggested target, time-related fields, and advanced settings reduce setup effort.
Configuration time was reduced by 50% in usability testing.
My involvement
As Lead Product Designer, I drove the redesign end to end — problem framing, interaction design, prototyping, and usability testing — in close partnership with engineering and PM.