← Product / Field Descriptions

Fighting subconscious shortcuts

Focus:

UI, interaction design

Spring 2023

Timeline:

Overview:

As designers, we’re constantly weighing the tradeoffs of existing interaction patterns, reviewing the competitive landscape, and making calls on the right approach. But sometimes, when the call seems so obvious, you might come to realize that you’ve taken the easy way out.

The task at hand

To respond to customer requests to improve data literacy and shared understanding following the release of Reusable Datasets, we decided to invest in field descriptions.

Field descriptions were a very common feature within the data space, allowing users to provide additional context via text descriptions for each individual column/field within a given table of data.

System architecture

Before introducing a new reusable dataset object, I had to think about where and how it would fit in in relation to the rest of the system. Similarly, engineering had to start thinking about how we might evolve our legacy dependency on queries.

Organizational principles going into the project:

  • Workspace: an individual customer instance, housing users and their content

  • Spaces: folders of Reports within a given Workspace

  • Reports: a project containing SQL queries, visualizations, and an optional dashboard view of visualizations

  • Queries: the underlying building block for Reports

Early usability testing

After a lot of team discussions and early customer conversations, we decided that Datasets would co-exist alongside Reports, allowing users to organize them across relevant Spaces. Reports could then be powered by queries, Datasets, or a mix of both.

To help address a lack of data governance in Mode, there was also a desire both from stakeholders and our customers to make Datasets feel elevated and differntiated from ad-hoc Reports.

With all of that input, I conducted rapid iterative prototype tests using an exhaustive click-through prototype. During the tests, we asked analysts to show us how they would:

  1. Create a new Dataset
    Analysts could write a query, and then see/prep its results for broader usage

  2. Set up a refresh schedule for the Dataset
    Similar to Tableau extracts, we wanted Datasets to run independently so that customers could leverage its cached results across dozens of Reports. This workflow would be both more performant and cost effective than running every query/Report individually, but it was a big change

  3. Create a Report using the Dataset
    We wanted to give analysts and business users a way to create a new Report from a Dataset without having to see the SQL editor

The prototype contained multiple ways to accomplish the same task to help us uncover the most intuitive workflows.

One of the many, many prototype screens from the Marvel prototype. Our design toolset changed significantly over the course of this project, as we transitioned from standalone tools (Sketch, Abstract, Marvel) to Figma.

The road to Alpha

The success rates for the flows were pretty high, and we gained some much needed clarity around where our power users would look to create new Datasets. We used these findings to solidify an MVP candidate, focussing on the core workflows of creating a Dataset and using it within a Report. I spent a lot of time working on the Datasets editor itself, which leveraged a simplified version of our Report editor behind the scenes.

To give analysts a better data prep and consumption experience, we also worked on a new component called the Data View that replaced our legacy query results table grid.

Alpha, Beta and beyond

Given the complexity of the project, we opened things up to customers initially through a closed Alpha. This allowed us to work with customers that were understanding of the incomplete state of the project and vocal about what was and wasn’t working.

One early challenge we uncovered was that users had a really strong expectation that Reports built off of Datasets should operate just like Reports built off of Queries. Rightfully so, they didn’t want to feel like Dataset-powered Reports offered a degraded experience. However, this dramatically increased the scope of the project as engineering had to rewire nearly every Report feature to use a new abstracted data object rather than expecting the data to always come from queries within the Report itself.

In some cases, 1-1 feature parity wasn’t possible. I spent a lot of time facilitating team conversations and brainstorms around how to ensure Report behavior would feel consistent and understandable, regardless of what data was backing it.

Entrypoint challenges

During Alpha and Beta, we also received usability feedback on our updated content creation workflows. Analysts wanted business users to have a clear and approachable way to build their own Reports using Datasets, but they also didn’t want that entry point to come at the expense of slowing them down.

Rebrand rollercoaster

To add to all of the change, we also decided as a business to undertake an enormous rebrand out ahead of the Datasets launch. As I worked to ensure our new CTAs and workflows could satisfy both user parties, I also had to make visual updates to the designs to align with our in-flight branding guidelines.

It took me a few rounds, but the final designs struck a balance between maintaining power user shortcuts and exposing friendlier, code-free starting points.

“Write SQL, get self serve”

We launched a new brand and Datasets together at the end of 2022. The introduction of Datasets meant reintroducing ourselves as a tool that could fit both the needs of code-first and code free users. Business intelligence, built around data teams.

I partnered with Marketing to produce video and static content for a standalone Datasets product page.

A continued effort

Since the initial MVP, we’ve shipped a handful of feature improvements to Datasets including a usage modal to help analysts understand which Reports are relying on a given Dataset and field descriptions to help build data literacy and shared understanding as more business users dig in and explore.

Usage stats

Early Datasets adoption was slower than we anticipated, as a lot of current customers viewed it as an ambitious data ops undertaking that had to be prioritized accordingly (fair). But as time went by…as more enhancements landed and we worked to share more enablement resources, usage began to climb.

Total # of Datasets made: 6K+

Total # of times a Dataset has been used in a Report: 18K+

Percentage of usage by non-technical, business users: ~40%

Appreciations

From 2020-2023, oh so many people had a hand in this project. The “Datasets team” has had many faces over the years, but I’d like to thank the following folks:

  • Mike DeCarlo (PM)

  • Sarah V (PM and Head of Product)

  • Todd P (PM)

  • Neha Hystad (PM)

  • Nan (Head of Product)

  • Jay Leandro (Engineering)

  • Anthony Simone (Engineering)

  • Sarah Raasch (Engineering)

  • Jared Smith (Engineering)

  • Ravi Patel (Engineering)

  • Tommy (Engineering)

  • Chris Haught (Engineering Manager)

  • Megan Kard, Jennifer Gieber, Simran and the entire design team for their invaluable design review feedback