All data shown is for illustration purposes only.
Summary
Walmart Data Catalog holds millions of internal data assets, and analysts couldn’t tell which ones to trust. I designed natural-language search and AI-generated asset descriptions, so people can ask for data in plain words and judge a result without opening it. I also designed Spotlight, a curated tier of trusted assets, with a readiness checklist that shows owners how to qualify. In UAT, task time dropped from 15–30 minutes to 3–5, and description coverage went from 40% to 100%.
What I owned
Led research: 4 workshops with 12 users.
Designed the search and Spotlight-tab directions.
Defined with business and engineering what the AI descriptions should include.
Problem
Walmart Data Catalog is Walmart’s internal platform where analysts and data users find and use company datasets. When users searched, all results appeared equal, and reliable tables were mixed with duplicates, deprecated versions, and poorly documented assets. Users had no signal for which result to trust, so many bypassed the catalog entirely and asked colleagues on Slack instead.
How did I know it was a problem?
We ran research sessions where participants shared their daily tasks.
4
One of the workshops
User statement analysis
Here I observed that…
A typical search performed in the session looked like this.
…users weren’t struggling to search, they didn’t trust what the search returned.
Understanding the opportunities
These opportunities directly shaped the four solutions I designed.
Key decisions
I explored two directions: a single list where Spotlight assets carry a tag, and a two-tab layout separating Spotlight results from everything else.
Descriptions emerged from research as the single most useful attribute for identifying an asset. I collaborated with business and engineering to define which details the AI-generated descriptions should surface, so the output was consistently useful rather than generic.
Final designs
NLP Search — users could describe what they needed in plain language instead of knowing the exact table name, making the catalog usable for people new to the data.
Spotlight Tab — trusted assets get their own tab, with automatic fallback to all results when no Spotlight assets match…
Automatically falls back to other assets when no Spotlight assets match.
AI Descriptions — every asset gets a readable, accurate summary, reducing the need to open an asset just to evaluate it.
Readiness Checklist — Owners follow a step-by-step checklist that makes the qualification criteria explicit and the path to nomination clear.
Early Results
Task time
15–30 min → 3–5 min
Description coverage
40% → 100%
Preferred the tab layout
8 of 12 users
Launch metrics
To follow
Most of the time saved came from the finding phase, not the task itself. Users also requested bulk nomination during testing, which is the next gap to close.
Learnings
We started this project trying to fix search. Watching users showed us the real problem was trust. That distinction shaped every decision that followed.
The readiness checklist works well for nominating individual assets, but bulk nomination remains unsolved. Users flagged this in testing, and it’s the most immediate gap to address post-launch.
Two areas worth exploring next:
Using AI to review data lineage.
Simplify data quality setup, reducing manual effort for asset owners.
