The "Nooxes" Intelligent Loan Platform
This project focused on reducing decision cycle time in a regulated, data-heavy loan review workflow without sacrificing trust, control, or auditability.
The platform was used daily by senior loan officers operating in high-stakes environments where errors are costly, decisions must be explainable, and confidence matters as much as speed.
My role:
Combined UX research leadership, AI strategy, and experience design. I guided problem framing, led research and validation, partnered closely with product and engineering, and ensured the solution was measurable, buildable, and aligned with regulatory constraints.
The Impact:
By shifting from a manual spreadsheet-style experience to an AI-assisted decision dashboard, we achieved measurable business outcomes:
-
Processing time: 4.2 days → 1.1 days
-
Throughput: +45% approved volume week over week (approvals per officer)
-
Adoption: 92% of eligible officers switched voluntarily within the first week of beta
The Business Challenge
Loan officers were spending an average of 4.2 days reviewing and approving a single application. The system behaved like a passive data warehouse: fragmented screens, manual calculations, and constant tab switching created cognitive overload and slowed decision-making.
Baseline cycle time was taken from ops reporting and system logs before redesign.
Officers had to hunt for critical information, reconcile missing documents manually, and determine priority without system guidance. This increased processing time, error risk, and drop-off, while senior resources were consumed by low-value manual work.
A dense, passive interface where critical metrics like '4.2 Days Processing Time' (bottom right) were stagnating.
The Strategy
Moving from "Data Entry" to "Exception Handling"
The core strategy was to invert the workflow from manual data entry to expectation handling.
Instead of forcing officers to review every application in full, clean and low-risk cases could progress automatically while the system surfaced exceptions that genuinely required human judgment. The goal was not automation for its own sake, but focus: helping officers spend time where expertise actually mattered.
We moved in short cycles: draft with AI support, prototype quickly, then verify with users before committing to build.
Research & Method
Research approach
-
Primary research goal
Reduce analyst decision time without reducing confidence or regulatory safety. -
Key decisions the research needed to inform
-
What must be reviewed by humans vs AI
-
How to prioritize work on first scan
-
Where trust breaks down in current workflow
-
-
Methods used
-
Semi-structured interviews (loan officers, supervisors)
-
Task-based usability testing
-
Workflow and journey mapping
-
System log and behaviour analysis
-
-
Participants
-
12 senior loan officers
-
4 supervisors
-
2 compliance reviewers
-
-
Outputs
-
Prioritized pain points
-
Workflow changes
-
Testable prototypes
* Why these methods
Methods were selected based on decision risk and time constraints, not academic completeness.
The Solution
The Intelligent Dashboard The new design shifts the cognitive model from "Search" to "Action."
My strategy was to invert the workflow. Instead of officers reviewing every loan, the system would auto-approve clean applications and only flag exceptions for human review.
Key Director-Level Features:
1. The "Smart Recommendations" Widget:
-
The AI proactively identifies opportunities.
-
Example: "Refinance Opportunity Detected". The system analyzes market rates against the applicant's fixed rate and calculates the spread automatically.
2. Visual Prioritization:
-
Loans are tagged clearly (e.g., "Exception" vs "Approved"). The "Latest Updates" feed uses icons to distinguish between Compliance Risks (Red Triangle) and Routine Tasks (Blue Check).
Example: "Refinance Opportunity Detected". The system analyzes market rates against the applicant's fixed rate and calculates the spread automatically.
* AI was used to summarize inputs, flag exceptions, and suggest next steps, allowing officers to quickly scan, orient, and act. Critical signals were elevated visually, while low-risk cases required less manual attention. Every recommendation is explainable and overrideable so the officer stays in control.
-
AI does
-
Summarize and contextualize application data
-
Flag exceptions and missing information
-
Suggest next best actions
-
AI does not
-
Auto-approve decisions
-
Hide actions or logic
-
Block manual overrides
An AI-powered workspace where 'Smart Recommendations' guide decision-making, reducing manual calculation.
.png)
AI Trust & Failure States
AI behavior by confidence level
-
When AI is confident: rationale and supporting signals are shown
-
When AI is uncertain: cases route to human review with gaps highlighted
-
When AI is wrong: users override decisions and actions are logged for audit
This directly supports regulated decision-making.
Scaling the System (Design Ops)
To support scale and consistency, the solution extended the existing design system with standardized AI states, severity indicators, and accessibility defaults.
The design system ensured predictable behaviour across products and allowed teams to ship faster without visual or interaction drift. Components were built to work seamlessly with React, reducing rework and improving engineering confidence.
We evolved from a fragmented 'Style Guide' (Left) to a tokenized 'Design System' (Right). By defining semantic tokens like AI-Accent-Purple, we ensured consistent branding for all future AI features."
Validation & Accessibility
Validation focused on speed, comprehension, and confidence.
We measured time on task, error recovery, and decision flow clarity through moderated usability tests.
Officers completed reviews faster while maintaining confidence in their decisions.
We also checked comprehension and confidence:
users could explain why a case was flagged and felt comfortable overriding AI when needed.
Key Director-Level Features:
1. 1. Usability Testing (Time-on-Task Analysis):
-
Test: We timed 10 Loan Officers performing a "Risk Assessment" on a complex loan application.
-
Result: The average time dropped from 14 minutes (Legacy) to 3.5 minutes (New Dashboard).
-
User Quote: "I used to need a calculator and three screens to check DTI. Now the dashboard just tells me 'Low Risk' and I verify it. It's hours of my day back." Senior Underwriter
2. Accessibility Audit (WCAG 2.1 AA):
-
Challenge: The legacy portal relied heavily on color (Red/Green text) to show status, failing color-blindness standards.
-
Solution: We implemented a "Double-Coding" system. Every status now uses both Color + Shape (e.g., Green + Checkmark, Orange + Triangle).
-
Outcome: Passed external accessibility audit with a 98% score, ensuring legal compliance for internal staff accommodation.
.png)
The Impact
ROI & Outcome By shifting from a manual spreadsheet view to an AI-assisted dashboard, we achieved:
Processing Time: Reduced from 4.2 days to 1.1 days.
Throughput: Increased "Approved" volume by 45% week-over-week.
Adoption: 92% of officers switched to the new dashboard voluntarily within the first week of Beta.