Hello Bank runs entirely on its digital experience. When that experience started cracking under the weight of new features, an overloaded support team, and a chatbot nobody trusted, the work became less about adding things and more about fixing the foundation. I led two connected initiatives across mobile and web:

• Relaunching an AI-assisted customer support system
• Redesigning the banking homepage experience

The work extended beyond interface redesigns. It required aligning product, engineering, legal, compliance, and support teams around a more responsible and scalable approach to AI, accessibility, and customer trust in a regulated environment.

The Chatbot Nobody Trusted

Reframing the brief strategically.

The original brief was to build a sophisticated conversational assistant that could handle a wide range of financial queries autonomously through generative AI. However, user feedbacks and technical discovery pushed back on that pretty quickly:

• Hallucination risk was too high for sensitive financial interactions
• Long response times damaged perceived reliability
• Ambiguous AI behaviour reduced trust quickly
• Unsupported requests lacked clear escalation paths
• AI capabilities were overestimated internally
• Users valued clarity and predictability more than conversational sophistication.

Rather than positioning AI as a replacement for human support, I reframed the experience around:

• Rapid information retrieval
• Task-oriented guidance
• Deterministic support flows
• Escalation assistance
• Transparent system limitations

The pivot from "autonomous assistant" to "confidence-aware guide" came from understanding that the design problem wasn't capability, it was transparency. This shifted the product strategy entirely, becoming a critical alignment point between product, engineering, and compliance teams.

AI Confidence Routing

Three paths structured around what the AI was certain about.

Instead of designing a purely conversational assistant, I worked with engineering teams to structure the experience around confidence-aware interaction patterns.

User submits a request → Intent is captured. Loading state starts immediately — no blank screen.
AI evaluates confidence → Does this request map to a known, safe category? Is the answer retrievable without guessing?

A. High confidence → Guided response
Structured answer, clear next actions.

B. Low confidence → Clarification prompt
Ask a specific follow-up rather than returning a vague answer. User stays in control.

C. Unsupported / risky → Human escalation
Explicit handoff with context passed forward. User knows why they're being transferred.

This reduced ambiguity while improving trust calibration during sensitive interactions.

Beyond The Happy Path

Most AI chat design shows the happy path. These are the states that actually defined trust:

Loading
Progressive reveal with expectation-setting copy. Users knew something was happening and roughly how long it would take.

Guidance
Deterministic response with clear next action. No open-ended "does that help?" dead ends.

Clarification
Specific follow-up question, not a generic "can you rephrase that." The system had to earn the retry.

Escalation
Named escalation path. "I'm connecting you with the team who handles this" is better than silence.

Reducing Perceived Latency

Redesigning the waiting experience before the back end caught up.

Backend optimisation takes time to ship, so I redesigned the waiting experience itself. Improvements included contextual status indicators, lightweight motion feedback, and microcopy that set expectations instead of asking users to just wait. 

The problem wasn't the time, it was the uncertainty. People can wait 20 seconds for a complex request if they believe something useful is happening. These changes reduced user uncertainty during long-running requests and significantly improved perceived responsiveness.

Conversational Design That Brings Confidence

Navigating between ongoing improvements and outdated branding.

I crafted language that would build trust through thoughtful UX writing. I chose to build the chatbot character's identity by defining its persona, its tone of voice.

AI Support Outcomes

The redesign of support experience led to:

→ Perceived response time dropped from 20 to 6 seconds — measured across usability session before backend changes shipped

→ Escalation clarity improved — unsupported requests now had a named path rather than a dead end ex. ticket volume reduction target: ~30% unconfirmed

→ Stakeholder confidence in AI-assisted workflows rebuilt — compliance and legal signed off on the revised interaction model

→ Conversational patterns became reusable across additional support flows

→ Earlier accessibility integration within delivery workflows

Most importantly, the initiative shifted organizational thinking from: “How do we automate support?” to: “How do we responsibly augment customer assistance?

An Outdated Homepage

What research has revealed.

The homepage experience had been designed primarily for quick account checks, but platform capabilities had expanded significantly over time. As new services were added, the information architecture became increasingly difficult to navigate:

• Important features lacked discoverability
• Advanced actions required excessive navigation depth
• Cross-platform inconsistencies increased cognitive load
• Users struggled to prioritise financial information
• Homepage density reduced task clarity

At the same time, the interface needed to support very different behaviours:

• Frequent daily banking tasks
• Occasional high-stakes financial actions
• Both digitally confident and less technical users

Four inputs shaped the decisions: competitor benchmarking, behavioural synthesis from support tickets and app store reviews, accessibility audit, and stakeholder workshops.

UX Strategy & Organisational Influence

From a visual execution workflow to a collaborative, research-driven practice.

 I introduced more collaborative and research-driven product practices, including:

• Kick-off interviews
• Workshop-based discovery
• Cross-functional ideation sessions
• Collaborative prioritisation exercises
• Data-informed decision frameworks
• HMW facilitation workshops

These practices improved alignment between teams and reduced downstream design debates during implementation.

Navigation Hierarchy Evolution

Designing for progressive disclosure.

Business stakeholders initially pushed to add more products and services to the homepage under the principle that visibility drives adoption. That's not wrong. The conflict came from usability data showing that more density was reducing task completion. The negotiation was about reframing what "visibility" actually meant: a feature buried under cognitive load isn't visible.

Before:

• Dense homepage, everything surfaced at once
• No priority hierarchy between actions
• Complex features 3–4 taps deep
• Inconsistent patterns across iOS and web
• Cognitive overload on first view

After:

• High-frequency actions surfaced first
• Secondary content contextualised, not hidden
• Reduced navigation depth for complex tasks
• Shared interaction patterns across platforms
• Progressive disclosure by task type

Design Operations For Scalable Systems

Consistency and implementation scalability as major operational concerns.

As the redesign expanded across platforms, I partnered closely with the design system lead to:

• Standardise responsive behaviours
• Evolve reusable navigation patterns
• Improve accessibility documentation
• Align interaction behaviors across platforms
• Reduce implementation inconsistencies between web and mobile

 Rather than designing isolated screens, I focused on scalable interaction systems engineering teams could reuse efficiently.

Accessibility-First Design Culture

I led a shift toward accessibility and inclusivity as core design values rather than afterthoughts.

Focus Areas

• Screen reader support
• Keyboard navigation
• Responsive scaling
• Reduced cognitive load
• Voice interaction compatibility
• WCAG 2.1 compliance
  

 I partnered with engineering and the accessibility referent to document reusable accessibility behaviours directly within the component library. This reduced implementation ambiguity and improved cross-platform consistency.

Homepage Redesign Outcomes

The redesigned homepage experience:

• Improved discoverability of complex banking actions
• Simplified high-frequency banking workflows
• Reduced navigation friction across platforms
• Established reusable cross-platform interaction patterns
• Increased accessibility consistency across components
• Expanded adoption of research-driven product workflows

Beyond the interface itself, the project helped mature how the organization approached:

• AI integration
• Accessibility
• Product discovery
• Systems-based design
• Cross-functional collaboration