The context
Visa and Deloitte partnered to create Visa Data Works, an enterprise data platform that consolidates scattered workstreams and prepares the infrastructure for agentic AI. Driven by extensive cross-functional user research, my team delivered a high-fidelity design that seamlessly transitioned into development.
Following strong executive feedback, the project secured 3 contract extensions, generating over $1.3M in its first two renewals alone.
My role
Directed end-to-end design for 3 workstreams, delivering comprehensive design systems, interaction patterns, and final high-fidelity deliverables.
Scaled design maturity by coaching executives on design thinking and establishing high-efficiency workflows that boosted team velocity.
Optimized research delivery by co-facilitating user interviews and spearheading a structured data-cleaning and annotation framework.
The process
While I crafted the solution, I prioritized simple, high-utility designs that made it easy for data practitioners to leave behind their old tools and adapt to our new workflow.
How did I identify pain points & validate my solutions? My research process was as follows:
Click each tab to dive in.
Please note images may be minimized or blurred due to privacy restrictions.
The solution
Visa Data Works is a platform designed as modular building blocks rather than a rigid pipeline, allowing data scientists to move flexibly thorugh their workflow while maintaining consistency and governance.
Breaking it down
Get Data
The standard, default model.
Connect approved sources, standardize intake, and make governed datasets accessible for analysis.
Conversational Solutions
The agentic model.
Turn trusted data into guided answers, summaries, and natural-language workflows.
Build BI
The analytics, report-based model.
Create dashboards, reports, and reusable metrics that help teams monitor performance and act faster.
Key features
Workspace
A comprehensive dashboard giving data scientists an immediate, bird’s-eye view of active projects, incoming project requests, and newly landed data assets within the ecosystem.
My Assets
A library of all data assets the user has saved within Visa Data Works or used in recent projects.
Asset Center
A centralized catalog directly connected to Visa's Data Center. Users can dive into granular asset metadata, view details, and bookmark datasets directly to My Assets for future reference.
Smart Search
An intelligent search engine supporting natural language queries and inline token tags (like '@Project Name'). It proactively highlights key metrics from relevant datasets based on search results and surfaces context-aware query suggestions on the fly.
Guided Workflow
A flexible pipeline that dynamically adapts to the user's specific objective:
Project Overview: Captures foundational project metadata (name, audience, collaborators).
Data Discovery (different per workstream):
GetData & Build BI: Users select data assets for their project from data assets via AI recommendations, My Assets, or the Asset Center catalog.
Conversational Solutions: User select agent model options.
Build ETL: Bridges to an external IDE for running ETL tasks while tracking execution health on a unified dashboard.
Development (different per workstream):
GetData: Bypasses secondary steps to move straight to publishing.
Build BI: Prompts users to define semantic models, build reports, and assemble comprehensive business intelligence dashboards.
Conversational Solutions: Opens an Agent Marketplace to select, configure (LLM instructions, validation, hosting, model hyper-parameters), and orchestrate custom AI agent workflows.
Project overview
Data discovery for: Conversational solutions
Data discovery for: GetData & Build BI
Lands in Build ETL, which is Development for: GetData
Development for: Build BI
Development for: Conversational solutions
Project Outputs
A final staging review area that lists all generated technical artifacts before deployment. The layout automatically pivots based on the goal, utilizing a clean list layout for standard data tracks or an interactive accordion format for complex BI structures.
Publish
The final gate where users review their compiled artifacts and hit launch, instantly deploying their finalized data product to the enterprise ecosystem.
Comparing solutions
Let's take a look at the user journey diagrams for each.
User journey map
User’s Steps
Define requirements, success rules, intended use, and proposed solution. Align stakeholders on scope and pre-scheduled materials.
Familiarize with data structure and content, connect sources, set access permissions, and confirm privacy/governance requirements.
Clean, merge, process, and aggregate source data for analysis; perform ETL and establish data quality routines.
Store data in appropriate solutions and organize into data meshes when applicable. Create subject-specific marts as needed.
Visualize data to extract insights, train ML models where applicable, and ensure ongoing data quality.
Conduct user acceptance testing; share findings and data for validation and facilitate external reviews.
Release the data product to market or client; maintain ongoing access for future collaboration.
What platforms/tools are used?
Jira
External third party data
Visa Net
CNS
ANA
DPI
OPCert
FDM
CSAuth
Regional Data Marts
Goro (VIDA Review)
Visa Data Catalog
GBI
Airflow
Python
PySpark
Tuber
FDM
Citrix (For Local Remote Server)
Trino
Denodo
ThinkCell Python Package
QA Environment (if available)
EF
Sharepoint
Regional Data Marts
Hadoop
Hive
DB2
MLP
Local Remote Server
Jupyter Notebook
GitHub
Cline
CoPilot
Python
Microsoft Excel / CSV
PPT
PowerBI
Tableau
Microstrategy
How long does this step take?
4–6 sessions minimum
Data is often very specific; stakeholders should be available for quick context checks.
2–3 weeks (depending on client)
Getting review & approval can take several weeks
~1 week for data pull
1 month, at least
Customization requests take up to 1 week for existing subscription-based data products
8-10 weeks for re-engineering for compliancy (6 weeks is the minimum time for a fix)
1–2 weeks to generate code for automating/documenting/publishing/quality check
~1 month for modeling
Up to 20 business days
1 month
User journey map
User’s Steps
Define requirements, success rules, intended use, and the proposed solution.
Review the data structure, connect sources, and confirm access, privacy, and governance requirements.
Clean, merge, transform, and aggregate data for analysis and downstream workflows.
Store data in appropriate solutions and organize marts or meshes for analytical use.
Visualize data, build models where applicable, and maintain data quality through the process.
Validate with users, facilitate reviews, release the product, and maintain ongoing access.
What platforms/tools are used?
Visa Data Works
*with integration with user's IDE for development
How long does this step take?
As little as one day, total.
Reflections
Designing for large-scale and data-driven users means prioritizing efficiency, tackling high cognitive load, and understanding that there is zero-error tolerance.
Working with developers and data practitioners exposed me to technical systems and terminology up-close. I am now more comfortable speaking to my designs in the context of data and technical systems, allowing me to work in cross-functional teams even more seamlessly.
I’m open to full-time opportunities, coffee chats, or just yapping about the latest trends. Get in touch or explore my resume.
graceerya@gmail.com
Connect on LinkedIn