Colleen Park
SELECTED WORK

PSI Dashboard, turning raw payment risk data into decisions partners can act on

Data Dashboard B2B 0 to 1 AI

A client-facing dashboard for Payment Success Indicator (PSI), so external partners can monitor payment-risk performance, understand transaction results, and manage fraud rules in one place.

PSI DASHBOARD IN MOTION
RULE SIMULATIONS REPLAY A RULE BEFORE IT TOUCHES PRODUCTION
PSI API Performance
API PERFORMANCE MAKES SYSTEMIC FAILURES VISIBLE IN ONE PASS

Customize Connect, letting clients brand and configure Connect without engineering

Self-Serve Tool B2B Client Hub Web

A self-serve customization tool inside Client Hub, so clients can align the Open Banking Connect experience with their own branding and user journey without waiting on engineering.

CLIENTS REORDER FINANCIAL INSTITUTIONS FOR THEIR OWN PRODUCTS
Uploading a logo in Experience Settings
CLIENTS UPLOAD THEIR OWN LOGO AND SEE IT APPLIED ACROSS SCREENS
Experience Settings with toggles, logo and CTA color
TOGGLES, LOGO AND CTA COLOR APPLY ACROSS EVERY CONNECT SCREEN

A COVID-19 testing flow built into the Uber app, for riders and drivers

Mobile Service Design Concept

Built upon the Uber app to create a unique experience of covid-19 testing that addresses the concerns and safety needs of the riders and drivers.

Round trip to a drive-thru test, booked in the app
ROUND TRIP TO A DRIVE-THRU TEST, BOOKED IN THE APP
Selecting a testing location on the map — hover to play
TESTING SITES WITH WAIT TIMES AND APPOINTMENT RULES
Driver information and contact tracing opt-in — hover to play
DRIVER SAFETY INFO AND OPT-IN CONTACT TRACING
{{ caseName }}
CASE STUDY / PAYMENT RISK

PSI Dashboard: From Raw Risk Data to Actionable Intelligence

Empowering risk managers to transition from reactive monitoring to proactive, confident decision-making.

B2B DATA PRODUCT RISK & FRAUD SIMULATION SHIPPED
IMPACT Shipped to production
40M → 60M Open Banking users supported (U.S.)
4.7 / 5.0 Clarity and usability, three test rounds
→ Reactive to proactive: rules validated before production
WHAT I DID · JUMP
TEAM
Lead Product Designer End-to-end
Cross-functional PM · Eng · UX Research
TIMELINE
Research → Ship B2B / Payments
Domain Open Banking risk

Executive summary

The Challenge: Payment Success Indicator (PSI) uses AI to predict two kinds of account-to-account payment returns — non-sufficient funds (NSF) and unauthorized fraud — and scores every transaction for both. Those results reached partners only through APIs. PSI had no client-facing dashboard with aggregated user and transaction insights, which left partners without transparency and the product without competitive parity.

The Solution: An MVP dashboard that turns PSI’s technical API outputs and reason codes into a single source of truth for partners: performance monitoring, explainability behind each risk result, and a self-service rules-management experience where fraud rules can be simulated before they go live.

WHAT PSI IS
Two predictions per transaction Non-sufficient funds (NSF) and unauthorized fraud. Every transaction is scored for both.
0–100, higher is safer
0–10 · high risk 11–30 · medium risk 31–100 · low risk
A higher score means a greater likelihood of successful settlement.
What sits behind a score NSF analysis uses real-time balance, forecasting up to ten days, and eight weighted reasons — balance history, NSF history, spending, deposits, transaction amount. Unauthorized-return analysis covers current-day risk, with the Mastercard Identity Risk Network in development.
HOW THIS PROJECT MOVED

Problem: the “Data-Action” gap

PSI delivered NSF and unauthorized-fraud scores through APIs, but nothing helped partners read them. The scores arrived without aggregation, without explanation, and without a way to act.

01
WHAT THEY HAD Scores and reason codes over API
NSF
FRD
RET
API
FI
02
WHAT WAS MISSING No actionable context
Scores arrived without aggregation
without explanation
Without a way to act
→
No actionable context
03
CORE PAIN POINTS
01
High Cognitive Load Risk scores arrived with reason codes, but nothing explained why PSI landed on a result.
02
Operational Blind Spots Identifying API health issues or systemic failures was slow and manual.
03
Lack of Predictability No way to test a scoring rule change before production.
04
QUESTIONS THE DASHBOARD HAS TO ANSWER The design challenge was to turn PSI’s API outputs and reason codes into an experience where a partner can answer five questions quickly.
Q1 How are transactions performing?
Q2 Where is NSF or unauthorized-return risk concentrated?
Q3 Why did PSI assign a particular risk result?
Q4 What action or fraud rule should the partner consider?
Q5 How is performance changing across users and transactions?
“We have the data, but we don't know what to do with it.” PARTNER FEEDBACK

Before research: learning the data

I could not interview a risk manager about a score I did not understand myself. So the first week was spent with the product and engineering material, mapping what PSI actually returns for a single transaction — and where a partner would have to make a judgment call.

ONE TRANSACTION
Checking …3456
Balance $5,752.00
Transfer amount $1,000.00
Ex: an account funding transaction.
SCORE OUTPUTS Every transaction is scored for both risks.
1 Non-Sufficient Funds
Settlement risk score 0–100, forecast up to ten days to predict optimal payment timing, plus real-time account balance. 8 weighted score reasons
Recent balanceRecurring NSF Balance historySpend history NSF historyDeposit history Recent NSF historyTransaction amount
2 Unauthorized Return
Settlement risk score 0–100 for the current day, from real-time analysis. Optionally incorporates the Mastercard Identity Risk Network, a feature still under development.
SCORE RANGES A higher score means a higher likelihood of settlement.
0–10
High Risk
High risk of insufficient funds or unauthorized fraud return.
11–30
Medium Risk
Moderate risk of insufficient funds or unauthorized fraud return.
31–100
Low Risk
Low risk of insufficient funds or unauthorized fraud return.
READING A SCORE OF 5
NSF Score = 5 The account is unlikely to have enough funds to cover the transaction.
Unauthorized Return Score = 5 The transaction is likely initiated by a first or third-party fraudster.
Both scores read “5,” both mean high risk, and they mean two entirely different things. That was the moment the design problem became clear: the number is not the answer, and averaging the two would destroy the only information a partner can act on.

Research

LEAN UX
1. Research
7+ interviews
2. Synthesize
Competitive benchmarking
3. Iterate
Hi-fi prototypes
4. Validate
7+ participants

Discovery: 7+ in-depth interviews and competitive benchmarking to map how risk managers think.

Iteration: Hi-fi prototypes tested with 7+ participants for clarity under pressure.

INTERVIEW GUIDE · 6 TOPICS

A structured guide to learn how clients use current PSI visualization tools, and what a dashboard would need to change.

{{ t.num }}
{{ t.label }} {{ t.probe }}
Paired with client outreach templates for scheduling.
CROSS-FUNCTIONAL LEADERSHIP
PMs Aligned design roadmap with business growth goals.
Engineers Weekly feasibility checks on real-time simulation.
UX Researchers Validated that changes reduced cognitive load.

Strategy: Designing for Confidence

Users didn't need more data. They needed more certainty.

THRESHOLD
?
{{ shiftTitle }} {{ shiftSub }} Simulate before shipping {{ shiftDetail }}

The simulation environment

V1 · 03 RULE SIMULATIONS
V1 rule simulations screen
The rule table users could read but not act on. ↑ back to the first pass
→
V2 · SIMULATION ENVIRONMENT
SIMULATED VS. PRODUCTION
The same rules, now runnable: change one, see the impact before shipping.

The answer to “what happens if I change this?” was not a slider. Partners write an override rule against the NSF model, run it over real historical volume, and read the predicted returns before anything reaches production.

Generate Simulations Generate override risk score rules for the NSF Model
Rule 1
If Select FI =
And Transaction Amount
Then Return NSF Model score as
{{ hintSetup }}
{{ hintRun }}
Replaying Rule 1 over 11,972 transactions… 06/01 – 06/07
RETURNS OVERVIEW · PRODUCTION
Total returns predicted by PSI
1,000 84% of total returns
Total returns missed by PSI
197 16% of total returns
1,197 returns of 11,972 transactions, selected period.
SIMULATED RETURNS OVERVIEW RULE 1
Total returns predicted by PSI
{{ simPredicted }} {{ simPredictedPct }} of total returns ↑
Total returns missed by PSI
{{ simMissed }} {{ simMissedPct }} of total returns ↓
{{ simNote }}
{{ applyTitle }} → {{ applyBody }}

First design pass

FIRST DESIGN AFTER RESEARCH · 3 SCREENS

The first pass translated the interview findings into three surfaces. Scroll each frame to see the full screen.

WHY THIS SCREEN LOOKS THIS WAY {{ firstLabel }}
{{ firstWhy }}
{{ firstAnswers }} HOVER A DECISION → THE SCREEN HOLDS ON IT
V1 · {{ firstLabel }}
First design, API Performance First design, Score Insights First design, Rule Simulations
{{ r.num }}

Testing

Three rounds of hi-fi prototype testing with 7+ risk managers, each round checking a different one of the five questions: can they read performance and risk concentration, do they understand why PSI scored a transaction, and will they manage a rule themselves.

{{ roundFocus }}

TESTED {{ roundTested }}
FOUND {{ roundFound }}
CHANGED {{ roundChanged }}
4.7 / 5.0 clarity and usability after round 3
7+ participants per round

Iterations after testing

V1 → FINAL

Three changes carried the weight of the feedback. Each one is shown before and after, at the part of the screen it touched.

01 API PERFORMANCE
An FI filter on the page itself Performance could only be read in aggregate. The header now carries a Financial Institution filter, so a partner can isolate one institution and see its volume, success and failure behavior on its own.
BEFORE · NO FI FILTER API Performance header before testing
AFTER · FI FILTER IN THE HEADER FI filter added to the header
02 SCORE INSIGHTS
Risk colors that name their own range Red, yellow and green carried no scale. Hovering a risk label now reveals the score band behind each color — High 0–10, Med 11–30, Low 31–100 — so the chart can be read without leaving it.
BEFORE · COLOR WITH NO SCALE Score Insights risk labels before testing
AFTER · SCORE BAND ON HOVER Risk label revealing its score band on hover
03 RULE SIMULATIONS
A rule you compose, and a result that comes back The builder no longer pre-names its own fields. Every condition is picked as attribute, operator and value, so one rule can be written against any attribute instead of the two the form assumed.
BEFORE · FIELDS FIXED BY THE FORM Rule builder before testing
AFTER · ATTRIBUTE / OPERATOR / VALUE Rule builder after testing
ALSO A shadow test now returns its result a week later, in the same place the rule was written, so a rule is judged on a week of real traffic before it reaches production.

Final design

SHIPPED · 3 SCREENS

Three screens under one Insights rail, each answering one question and handing the user to the next. Pick a screen, then hover a decision to hold the frame on it.

{{ shipTitle }} {{ shipLede }}
{{ rsWhen }} · {{ rsLabel }} {{ rsBody }}
FINAL · {{ shipLabel }}
Final design, API Performance Final design, Score Insights Rule Simulations step 1 Rule Simulations step 2 Rule Simulations step 3 Rule Simulations step 4 Rule Simulations step 5 Rule Simulations step 6 Rule Simulations step 7 Rule Simulations step 8 Rule Simulations step 9
{{ rsCta }}
{{ rsCta }}
Top 5 Failed API Reason Code Distribution Count and Percentages of Top 5 Error Code Types
Total Failed API Reason Code Distribution Count and Percentages of Total DS HTTP Status and Error Code Types
Download CSV
Rank DS Https Status Error Code Description Count Percent
Total 697 100%
{{ r.rank }} {{ r.status }} {{ r.code }} {{ r.desc }} {{ r.count }} {{ r.pct }}
{{ codeHint }}
{{ r.num }}
{{ rsCta }} Submit rule to shadow test confirmation Submit rule to production confirmation

Impact

The dashboard shipped as the decision layer for PSI, and the behavior it was built to change did change.

REACH
40M → {{ ocUsersBig }}
Open Banking users supported (U.S.)
SATISFACTION
{{ ocScoreBig }} / 5.0
Clarity and usability, across three test rounds
BEHAVIOR
Reactive → Proactive
From fixing errors to optimizing in a sandbox
OBSERVABILITY Isolated incidents are now distinguishable from systemic API issues at a glance.
INTERPRETABILITY Users read macro-trends and anomalies instead of isolated data points.
PREDICTABILITY Rule changes are validated against historical volume before they touch production.
CASE STUDY / CLIENT SELF-SERVICE

Customize Connect

A self-serve customization tool in Client Hub, so clients can tailor the Open Banking Connect experience themselves — branding, FI search order, and publishing — without engineering support.

PRODUCT DESIGNER LAUNCHED DEC 2023 FIGMA WEB
IMPACT Shipped to clients
200 Unique patterns configured
70% Less reliance on support teams
→ Integration turnaround cut from 2–4 weeks to self-serve
WHAT I DID · JUMP
TEAM
Product Designer End-to-end
Cross-functional PM · Eng · Client Success
TIMELINE
Research → Ship Launched Dec 2023
Domain Open Banking / B2B self-service
Customize Connect experience editor in Client Hub

What were we trying to solve?

Customization required engineering support, long turnaround times, and back-and-forth with internal teams. Three pressures made that untenable.

ONE BRANDING CHANGE, BEFORE CUSTOMIZE CONNECT 2–4 WEEKS END TO END
STEP 1 Client requests a change
→
STEP 2 Sales engineering picks it up
→
STEP 3 Internal build and review
✕NO PREVIEW
First look, already live
DAY 0 88% OF THE TIME SPENT WAITING ON INTERNAL TEAMS LIVE
FRICTION No preview before launch, and no way for a client to iterate or test on their own.
USERS More control over branding Clients needed the Connect experience to match their own branding and user journey, and could not get there alone.
LOW EFFICIENCIES 2–4 week integration delays Every change ran through internal teams, driving higher costs and inefficiencies on both sides.
COMPETITION Third-party alternatives Without a first-party option, clients turned to external tools to get the control they wanted.
BEFORE — CURATOR Clients had to contact sales engineering for every change, and could not preview the experience until it went live.
AFTER — CUSTOMIZE CONNECT One place inside Client Hub to configure, preview, publish, and share the experience.
Curator experience settings before Customize Connect Customize Connect experience settings inside Client Hub BEFORE · CURATOR AFTER · CUSTOMIZE CONNECT
◀ ▶
DRAG TO COMPARE

Research & discovery

I built a persona and journey map from client conversations, then ran a survey to find where customization actually broke down.

USER PERSONA Persona: Flynn Laxton, Program Manager at Acmelending Persona: David Rhie, Sales Engineer at Finicity
{{ jrnLabel }}
{{ jrnPct }}
Journey maps for Flynn, the client program manager Journey maps for David, the sales engineer
FINDINGS
Problem Limited control — aligning Connect with their branding and user journey ran through internal teams. Lack of clarity — with no preview before launch, clients could not tell what was customizable or what a change would look like.
→
What to solve Put branding controls in the client's hands, and make every change visible before it goes live.

Strategy: a self-serve tool, not a service request

Making ‘how might we’ question

“How might we create a self-serve customization tool that reduces reliance on third-party solutions, minimizes internal workload, and remains competitive by offering a customizable and cost-effective experience?”

How self-serve customization matters in business

Customizing without engineering
influences
Letting clients customize themselves means business profits
Lower internal workload & faster onboarding
results in
Retention over third-party tools
WIREFRAMES — FINDING THE BEST INTERFACE
{{ wfPct }}
Wireframes of the Customize Connect editor, upload flow and publish dialogs
WIREFRAME RESOLVED INTERFACE
The wireframes resolved into four decisions that carried into the shipped product.
Experience settings panel with toggles for back button, exit button, logos and brand color
01 Easy customization controls Clients adjust branding elements like logos, colors, and button styles without coding.
Live preview of the Find Banks screen beside a drag-to-reorder institution list
02 Live preview Changes are instantly reflected, ensuring a smooth and intuitive user experience.
Landing page editor with an inline tooltip explaining the data transparency copy
03 Guided interactions Tooltips help users navigate the customization process effortlessly.
Share dialog with demo link, production link and email tabs
04 Publishing workflow Clients can review, test, and publish updates seamlessly.

Testing

PHASE 1 Research & discovery Competitive review of Curator, Canva and Squarespace to shape the journey.
→
PHASE 2 Usability testing 12 participants from Client Success and Sales Engineering, observed live.
→
PHASE 3 Beta testing Weeks of real use, with weekly check-ins on where testers got stuck.
TESTING DOCUMENTATION Internal testing and beta testing research documents
WHAT TESTING CHANGED
FINDING 01 Users struggled to find the entry point in Client Hub. Fix  · Improved entry point visibility in Client Hub.
FINDING 02 Testers wanted one place to change settings all at once. Fix  · Added a list view option so settings can be edited together.
FINDING 03 Navigation issues — users didn’t realize they needed to scroll. Fix  · Highlighted editable screens for easier navigation.
PUBLISH
FINDING 04 No publishing flow — users lacked clarity on finalizing changes. Fix  · Added guidance pop-ups and a clear publishing flow.

Outcome

Four capabilities shipped in Client Hub, each replacing a request that used to go through engineering.

01 Customized FI search

Clients can reorder financial institutions based on their products and preferences. This ensures the most relevant options appear first, streamlining user journeys and improving discoverability.

02 Experience settings

By customizing the exit button, uploading logos, and applying brand colors, clients create a consistent branded experience. This reduces friction for end users and strengthens brand trust.

Editing logo, brand color and button settings in Experience Settings
03 List view

A centralized list view displays all active configurations, giving clients full visibility and control in one place. This transparency reduces errors and simplifies ongoing management.

04 Share experience

Clients can instantly share demo or presentation links, making it easier to align internal stakeholders and showcase configurations without additional setup.

Impact

Customize Connect shipped in December 2023, and clients started configuring experiences without opening a ticket.

ADOPTION 200
Unique patterns configured, showing strong adoption and early traction
SUPPORT LOAD −70%
BEFORE
AFTER
Clients configure and customize independently, reducing reliance on support teams
ENGAGEMENT Increase
Personalized FI search and branded experience improve relevance, satisfaction, and trust
CASE STUDY / SERVICE CONCEPT

Uber COVID-19 Testing

A COVID-19 drive-thru testing and safety validation feature built upon the Uber app, enhancing user safety for riders and drivers during the pandemic.

PRODUCT DESIGNER APR — JUN 2021 FIGMA, ILLUSTRATOR, MIRO MOBILE
IMPACT Usability testing results
+20% User satisfaction scores
+12% / −11% Engagement rate / abandonment rate
→ 50% fewer design inconsistencies across platforms
WHAT I DID · JUMP
TEAM
Product Designer End-to-end
Collaborators Ben · April · Jacob
TIMELINE
Apr — Jun 2021 10 weeks, research to prototype
Tools Figma · Illustrator · Miro

Problem

Riders and drivers carried four concerns into the same ride.

Rider and driver exposing each other to the virus
01 Safety Issue HIGHEST CONCERN
RIDER Worried about exposing themselves and the driver to the virus.
Driver waiting in the car
02 Increase in Price
RIDER Uber charges by time in the ride, so charges could keep rising during the wait.
DRIVER Unclear what happens to their rates if riders are not charged for that wait.
Sanitization spray bottle
03 Unsure Vehicle Sanitization
RIDER Unsure the other party wears a mask, or that the car is properly sanitized.
DRIVER Rising equipment costs, and no guidelines for handling riders’ belongings.
Rider and driver waiting together in the car
04 Unfamiliar Environment
BOTH Long, unfamiliar waits in the Drive-Thru line leave both anxious.

Solution

A Testing service inside the Uber app: one round-trip booking that takes a rider to a drive-thru site and back, with the safety terms stated before the ride is confirmed.

Uber home screen with the Testing tile in the service menu
01 Testing enters the service menu

Testing sits beside Ride, Food and Vaccine on the home screen, so the flow starts where riders already are instead of on an external booking site.

02 The terms come before the booking

The entry screen states what the service is: a round trip to a drive-thru test, added safety measures for both parties, and contact tracing as an option rather than a default.

COVID-19 drive thru testing intro screen with three terms and a Book Ride button
Selecting a testing location on the map with wait times and appointment requirements
03 Testing sites carry the details that decide the trip

Sites appear on the map with wait time and whether an appointment is required, and the trip is entered as pickup, testing location and drop-off in one form.

04 Driver information and anonymous contact tracing

The rider sees who is driving and what the car has in place before the trip, and can opt into anonymous contact tracing so an exposure can reach the other party later.

Contact tracing opt-in and driver information screens

Outcome

+20% User satisfaction scores User research informed product strategy and optimized the experience for complex logistics-based services.
+12% Engagement rate High-fidelity prototypes tested for usability raised engagement across the flow.
-11% Abandonment rate The same testing rounds cut the share of riders who dropped out before booking.
-50% Design inconsistencies Established and maintained UI patterns and contributed to the design system for cross-platform consistency.

Leave a note. Only you and Colleen can see yours.