Data Quality
What a Data Quality platform does?
Companies make decisions from data: client reports, risk checks, AI forecasts. A data quality platform checks that data before anyone relies on it. It finds records that are missing, duplicated, out of range or out of date, and flags them so someone can fix them.
Passive→Active
data clean up
Data quality stops being a clean-up job after something breaks. It becomes ongoing work, tied to what the business is trying to achieve.
What should have happened?
The duplicate is caught before the report goes out.
The engineer sees exactly which record broke, and where.
The business lead sees that accurate client reporting is at risk, before a client does.
What if data quality started with the business goal?
Most data quality tools start with the data. Teams write checks, find errors, then try to work out which ones matter.
This platform flips the order. The business sets a goal first, such as accurate client reporting. That goal decides which data matters, which checks to run, and which problems get fixed first.
Research, as a sole designer
I had to understand data quality before I could design for it.
Market Research
I explored four established platforms, Informatica, IBM, Ataccama and Salesforce, through hands-on use, product demos and documentation. Each handled part of the problem well.
None connected a data error to the business priority it affected.
Expert Interviews
I interviewed four people who knew the domain and the client context: a subject-matter expert, a project manager and business analysts. These conversations shaped the personas and journey map that follow.
Workshops with in-house teams
I ran two kinds. In the first, I brought low-fidelity screens and worked through them with the team to decide what mattered and what could be cut. In the second, we built the journey map together, so everyone agreed on how the four roles worked before design started.
Data Steward
Ensures data accuracy, consistency, and trust across business reporting.
↓
Needs to know which errors matter, and who should fix them.
Data Engineer
Builds and maintains data pipelines to enable reliable, scalable data flows.
↓
Needs to find exactly where a record broke.
Data Scientist
Analyses complex datasets to generate insights and support data-driven decisions.
↓
Needs early warning when the data changes.
C-Level / Lead
Use data quality signals to guide strategy, governance, and operational priorities.
↓
Needs to know whether the numbers can be trusted.
Insights
Four patterns…. One root cause
Across the interviews and workshops, the same problems kept coming up. They all traced back to one gap: nothing connected finding an error, understanding why it mattered, and fixing it.
Fixes are slow to confirm.
Teams waited hours to know whether a fix worked.
➔
How might people see the effect of a change before they commit to it?
➔
Too many tools.
Work was split across three systems, so problems fell between them.
➔
If four roles share one platform, how does each see what they need without drowning in what they don't?
➔
No link to business impact.
Even after a fix, no one was sure the data was right for the decision.
➔
How might a business goal decide which errors get fixed first?
➔
Every new report starts from zero.
New reports and regulations meant rebuilding checks each time.
➔
How might people see the effect of a change before they commit to it?
➔
What we designed the Data Quality platform led by business goals?
An AI-assisted platform where business goals set the agenda for data quality.
Leaders define goals and KPIs, and those decide which checks run on the data. AI suggests checks and flags problems, but people approve every change. Each of the four roles gets its own view of the same picture.
Persona mapping
AI as a User
I treated the AI as a fifth persona: a team member with a clear job and clear limits.
Its job was to watch the data, flag anomalies and drift, suggest checks, and show how each suggestion would change the health score. It could not change data or rules on its own.
Every suggestion waited for a person to approve or reject it, and each decision was designed to sharpen its future suggestions.
Defining this early meant every screen answered one question:
Where does the AI act, and where does a person decide?
Journey mapping
Mapping the workflow
The end-to-end data quality journey was mapped across all four personas, tracing how workflows, decisions, and information moved between roles. The map surfaced where support, visibility, or guidance broke down at critical handoff points, the moments this project was ultimately designed to fix.
Building system
Design system
I built a library of 80+ Figma components that kept four role-based views and complex data workflows consistent, so one designer could deliver the full platform in 10–12 weeks.
Design
Key Interventions
Business Layer
How might a business goal decide
which data problems get fixed first?
The tools I reviewed all started from the data: write rules, then sort through the alerts. That left the business at the end of the chain, finding problems in reports. Design asked what would change if the order flipped.
If a leader starts with a goal, the goal can choose the KPIs, and the KPIs can choose the checks.
So the first step is a plain-language prompt: the leader types a goal, and the platform suggests KPIs to track it.
Every rule then shows which KPI it protects, so priority comes from the goal, not from how many alerts fire.
Implementing AI
Where could AI reduce manual effort
without removing human judgment?
AI could spot patterns faster than people, but data teams answer for every number in a report. Letting AI make changes would save time and cost trust. So I drew the line at decisions: AI suggests, people approve. The real question became what a person needs to approve with confidence. The answer was evidence up front.
Every suggestion shows how many records it affects and how it would move the health score.
Users can accept, reject or skip, and those decisions were designed to make future suggestions more precise.
Personalisation
How can four roles share one platform
without drowning in each other's work?
Putting four roles on one platform risked one crowded screen, each role needed a different depth, different urgency, different language. Design asked what each role actually needed to see and act on, and what would just be noise. Personalisation was built around role, adapting dashboards, workflows, and access automatically based on who was signed in.
Each role sees a version of the platform shaped for their responsibilities, a data steward sees accuracy and trust metrics, a C-level lead sees governance and strategic priorities.
Access and workflow steps adjust the same way, so no one wades through detail that isn't theirs to act on.
Rule Workbench
How might a rule written once
work for every dataset that needs it?
Every new report meant writing the same checks again. But many datasets need the same basics: no duplicate IDs, no empty fields. So I made checks reusable.
A team can write a check once and copy it to any similar dataset.
AI can also suggest new checks. The platform then shows which checks catch the most problems, so teams know which ones are worth keepi
Data Visualisation
How do you make complex data
understandable at a glance?
Raw data quality metrics meant little without context, a number alone doesn't tell a user if something needs attention now or can wait. Design asked how to turn tracked metrics into something scannable, so users could spot what mattered without interpreting a spreadsheet first.
Complex data is translated into charts and visual patterns, trends, anomalies, and critical issues, so users can identify what needs attention at a glance, not after analysis.
Outcome
What Changed
1.
Consolidated 3 disconnected platforms into one coherent workflow, connecting business priority and technical action in the same system for the first time.
2.
Defined the role of AI - so every suggestion shows its impact first and nothing changes without approval.
3.
Built a reusable component library that held design consistency across every role-based dashboard and complex data workflow, within a 10-week Agile timeline.
Data Quality Expert
Designed an AI-powered data health platform where every data problem is tied to a business goal, so every role knows what to fix first
Role
UX Researcher
Product - Service Designer
Team
Design Director
Data and AI Team
Business Analyst
Developers
Timeline
2023 - 10 weeks
Full Time
OVERVIEW
Fragmented work-tools, data teams found problems in one tool, ranked them in another and fixed them in a third. Nothing connected an error to the business decision it affected. Every alert looked equally urgent, and the business often found out last.
This platform brings it into one place. Business leaders set goals, and those goals decide which data gets checked. AI flags problems and suggests fixes. People approve every change.
MY ROLE
I was the sole designer on this project, embedded in a cross-functional team of data engineers, business analysts, and technical leads. Everything you read here is my work, from scoping the research to shipping the final prototype.
Over 10 weeks, I designed an end-to-end platform, from 0 to 1, that turned three disconnected systems into one coherent workflow, delivered as an MVP.
OUTCOME
OVERVIEW
Fragmented work-tools, data teams found problems in one tool, ranked them in another and fixed them in a third. Nothing connected an error to the business decision it affected. Every alert looked equally urgent, and the business often found out last.
This platform brings it into one place. Business leaders set goals, and those goals decide which data gets checked. AI flags problems and suggests fixes. People approve every change.
MY ROLE
I was the sole designer on this project, embedded in a cross-functional team of data engineers, business analysts, and technical leads. Everything you read here is my work, from scoping the research to shipping the final prototype.
Over 10 weeks, I designed an end-to-end platform, from 0 to 1, that turned three disconnected systems into one coherent workflow, delivered as an MVP.
1.
Consolidated 3 disconnected platforms into one coherent workflow, connecting business priority and technical action in the same system for the first time.
2.
Defined the role of AI - so every suggestion shows its impact first and nothing changes without approval.
3.
Built a reusable component library that held design consistency across every role-based dashboard and complex data workflow, within a 10-week Agile timeline.