Data Central Platform
Embedded Files
INDUSTRY: Data Management
COMPANY: Amazon, BDT
YEAR: 2020-22
ROLE: Head of Design
Overview
Amazon Business Data Technologies (BDT) accelerates the company’s data-driven decisions by providing the foundational data governance, data catalog, and big-data platform used across Amazon. BDT's mission is to accelerate data-driven business, enabling the next generation of analytics and machine learning technologies at scale
When I joined, the org had no UX practice, and product experiences were built entirely by engineers. My roles were to build the UX function from the ground up, define the end-to-end product vision, and lead the transformation toward a unified data experience that improved productivity, trust, and governance for data engineers, analysts, and ML teams.
Situtation
Challenges
Amazon operated a complex internal data ecosystem:
Fragmented tools
Teams used 3–5 separate applications to discover datasets, understand lineage, govern metadata, request access, and execute platform tasks.
Slow and inconsistent workflows
Common data tasks took 3–7 days (up to 2 weeks) due to unclear ownership, missing metadata, and hard-to-find lineage.
Low data trust
Users lacked confidence in data quality because of no single source of truth, inconsistent metadata standards, and limited visibility into data freshness, dependencies, or quality signals.
Engineer-driven UX
Tools were built for technical correctness, not usability. Most users required human assistance to complete critical workflows.
Before - 5 different data sources/tools & 7 applications
Before - inconsistent UIs across applications
My Role
Build the UX Practice from the Ground Up
As the head of UX, my role was to transform a highly technical, engineering-led ecosystem into an intuitive, end-to-end data experience that scales across Amazon.
Built UX Team
Recruited, mentored, and scaled first design team to 4 designers in Business Data Technologies in one year
Established UX Engagement Mechanisms
Included UX Office Hours, Voice of User, Bar Raiser, monthly UX flash, and UX program review with Sr. stakeholders
Cross-Functional Influence
Represented UX in annual operating planning across five VP above goals, and created shared team goals of establishing UX framework for Data Platform
Defined UX Design Process
Unitized the iterative process to understand the user, challenge existing assumptions, and redefine problems in attempt to ideate strategies and solutions
Goals Settings
Partnered with BDT leaders to define UX Design goals aligning with business
Business Goals
Accelerate Amazon’s data-driven decisions
Reduce time-on-task and reliance on manual support
Strengthen governance + security for data at scale
UX Design Goals
Simplify data discovery and management
Provide centralized workflows (search, notifications, table creation)
Establish UX process & culture in an engineering org
Discovery
User Personal
Data Producers: publish/manage datasets, often overwhelmed by metadata chaos.
Data Consumers: search + join datasets for analytics/ML.
Data Engineering Managers: monitor workflows, resolve breaches, ensure compliance.
Data Producer Journey Map
50+ studies, mapped producer/consumer workflows:
Identify Problems & Data Needs
Discover & Identify Datasets
Ingest, Store and Process
Extract, Process, Store
Post-Analysis Ingress
Socialization & Visualization
Maintenance
Core User Pain Points
This research informed the strategy for a unified platform experience.
Data discovery relied on tribal knowledge, not intuitive tooling, making it difficult to find the right datasets.
Metadata quality was inconsistent, varying widely across teams and lacking standardization.
Users had no clear understanding of dataset ownership, how to request access, or whether data was trustworthy.
Data lineage was incomplete or missing, limiting visibility into data dependencies.
Governance workflows were manual, fragmented, and unclear, increasing operational overhead.
Despite strong technical capabilities, the platform delivered an overwhelming, engineer-centric user experience that slowed productivity.
Strategy
A Unified Data Experience
I defined a cross-organization UX strategy built on three pillars:
Discoverability
Create a cohesive way for users to find datasets, pipelines, metadata, and owners without navigating multiple tools.
Trust & Governance
Increase confidence in data by providing visibility and context.
Productivity
Replace multi-tool workflows with integrated, self-service experiences.
The Hybrid Design System (Katal + BDT)
To scale across 5+ disparate applications without bloating engineering debt, we faced a "build vs. buy" decision. I led the evaluation that selected Amazon's Katal Design System for its superior instrumentation and A/B testing capabilities, rejecting the legacy Polaris framework.
However, Katal lacked complex data capabilities. My team engineered a custom "BDT Theme" layer to handle our specific density needs and contributed critical components back to the core library.
Advanced DataGrids
Handled cell-level sorting and pagination, saving partner teams ~7 days of dev time per implementation.
Code Editors
Standardized SQL/Python inputs across the platform.
Solution
DataCentral
DataCentral is the first unified data experience that brings together data discovery, governance, lineage, and workflows into a single platform.
Below are key features I led:
Unified dataset search
Integrated task flows replacing 3–5 distinct tools
Rich dataset profile pages with lineage, quality, ownership, and metadata
Governed access request workflows
Standardized metadata authoring for data stewards
Clear, consistent interaction patterns across teams
Universal Search
“I search separate things in separate sites, they don’t work all the time. Metadata is scattered across at least 7 different places.” -Data Pipeline Creator Aug 2021
Problem Statement
Data producers and consumers are frustrated to search in multiple applications because metadata are not useful, and on average it takes 7+ different places to find dataset he needs.
Solution
Build universal search that enables data producers and consumers to get their task destinations efficiently, reduce time on data discovery.
Data Producers are working with his partner team looking to join datasets related to Warehouses. They are asked to source, identify, and subscribe to the right dataset.
Data Producer reviews Table Details Page for “WAREHOUSE_ATTRIB UTES” looking at Details of the Table, Schema, Support Level and Schedule Type.
Notifications
“A single large scale breach issue created ~1,000 tickets in April 2021. ~60-70 tickets could have been avoided if alarm aggregation was in place.” - Data Engineering Manager, BDT
Problem Statement
Data producers are overwhelmed at managing the large volume of email and SIM tickets from each DataCentral application.
Solution
Develop a centralized location that simplifies all communication exchange for data producers
After data producer submitted request for permission access, the provider of WAREHOUSE_ATTRIB UTES will get a notification for review.
Notification dashboard shows a consolidated view of all communications.
Data producers have a need to run periodic communication campaigns to multiple recipients where status of the task is tracked.
Current workflow: Manual tickets are individually created to notify fleet owners. To set up a campaign would take upwards of 2 weeks.
Create Table
“I cannot actually create a schema within a cluster. And so a team member created one and now I have read/write access to that I just haven't really gotten as far as I've created a table, but I don't know how to load the data in it” - Program Manager, Data Oct 2021
Problem Statement
A data producer is exasperated to spend 3+ days creating a table using 5 different applications and services and categorizing data through Kale.
Solution
Simplify table creation workflow for data producers by reducing time on task within one application
Data categorization contains three steps including customer personal data, categorize column data, and roe data filter.
The review page shows new table detail and column data categorization detail. Data producers are allow to edit before submitting for approval.
Once the table is created, the page links to the table dashboard where displaying the Warehouse_data on the top.
Self-Service Governance
Problem
Data governance was our biggest bottleneck. Users faced "shifting requirements" for GDPR/PII compliance, often waiting days for manual approvals via tickets.
Solution
Design a fine-grained access control framework that turned complex compliance rules into a simple wizard.
Data Consumers discover , evaluate and start requesting access.
Data Consumers define the scope of data, which they need and request access to.
Dataset Owners manage approval requests.
IA & Navigation
The goal of the DataCentral IA & Navigation project is to reorganize the navigation IA to improve findability and discoverability of relevant information. It aims to focus on the tasks that data producers and consumers undertake on a daily basis.
The new navigation allows users to complete each of their tasks in less time by providing use-case-based workflow pages that simplify complex processes and are application-agnostic.
UX Mechanisms & Culture Change
Introducing UX into a deeply technical team required scaling influence across orgs. I established mechanisms to make UX part of the operating rhythm.
UX Office Hours
Provided design consultation to teams without dedicated designers (105 sections in 18 months)
Voice of User Program
Monthly structured insights shared with engineering and product leads
UX Bar Raiser Program
Ensured consistency and set a quality bar across the BDT organization
Experience Principles for Data Tools
Adopted across teams to guide future product decisions
These mechanisms helped shift BDT from an engineer-driven model → user-centered model.
Results
Archived Business & Design Goals
Catalog 1.5K petabytes of data for 62K monthly active users
(1 petabyte = 1K terabytes, FB generates 4 petabytes of data per day)
95% Efficiency Gain
Consolidated 5 disparate apps into 1. Reduced critical workflow time from 5-6 days to <15 minutes.
Operational Savings
The self-service models for governance and table management are projected to reduce support tickets by 80%.
Trust Improvements
Higher data trust through more complete metadata, clearer ownership, and accurate lineage.
Widespread adoption of the unified DataCentral platform and standardized workflows across teams
Organizational Influence
Stronger UX influence, with design embedded in roadmap planning and mechanisms adopted across 10+ product teams
Secured 3-year UX roadmap funding through demonstrated impact
What I learned
Data platforms succeed when trust and clarity are built into the UX
Data governance must feel like enablement, not a compliance burden
Engineers adopt UX standards when they understand the “why” behind them
The most valuable impact comes from building mechanisms beyond designing screens
Page updated
Google Sites
Report abuse