Trusted intelligence for everyday banking

Getting started

Canadia Bank serves millions of retail customers through branches and digital channels. When I joined as Senior UX/UI Lead, the mobile app worked—but it did not keep pace with how people expect money to feel on a phone. This case study focuses on the redesign of core banking journeys and the introduction of AI-driven insights built for trust first.

The Challenge

Canadia Bank’s mobile app carried fifteen years of product history in a single codebase. Transfers worked. Bill pay worked. But everyday customers still called the branch for questions the app should have answered: Where did my salary go? Why is this fee here? Am I on track this month?

The product team had a mandate to modernize—not a cosmetic reskin, but a platform customers could rely on for the next decade. That meant untangling dense legacy flows and introducing intelligent features in a market where “AI” often reads as “risk.”

[!NOTE] Insert 16:9 hero image: before/after of the home dashboard—legacy dense menu vs. simplified “Move money / Pay / Account” structure with a subtle insights entry point.

In discovery, we learned conservative banking users do not reject smart features—they reject surprise. Anything that moved money, changed limits, or surfaced advice without context triggered anxiety and support calls. Our design problem was not “add AI.” It was earn permission for AI, one screen at a time.

The Breakthrough

We chose one high-value moment instead of a chatbot everywhere: predictive spending insights on the account home screen—plain-language summaries of where money went, what looked unusual, and what was likely coming next (recurring bills, typical payday patterns).

I led UX for the insights module end-to-end: information hierarchy, confidence states, and escalation paths when the model was uncertain.

Designing for trust, not novelty

• Show the source. Every insight linked to underlying transactions—not a black-box score. Tap “Dining up 22%” and you land on filtered history, not a marketing paragraph.
• Confidence as UI. High-confidence patterns used calm, affirmative copy; low-confidence used “We’re not sure yet” language and softer visual weight so users never felt lectured by a guess.
• Human exit ramps. One tap to dismiss, snooze, or talk to support. We measured support volume weekly; the goal was insight without interrogation.

[!NOTE] Insert 16:9 image: AI insights card on account home—annotated callouts for (1) linked transactions, (2) confidence label, (3) dismiss / “Not helpful” actions.

The moment that changed the room

In a stakeholder review, compliance asked what happened when the model was wrong. Instead of a slide, I walked through a wrong insight prototype: incorrect category, clear “This doesn’t look right” feedback, and an audit trail for the bank’s risk team. That demo shifted the conversation from “Can we ship AI?” to “How do we ship it responsibly?”

Behind the scenes, I paired with data and engineering on prompt and guardrail copy—short, testable strings for edge cases (empty accounts, first-time users, joint accounts). We A/B tested headline tone: advisory (“You may want to review…”) beat imperative (“You should cut spending…”) by 18 points in comprehension tests with existing customers.

[!NOTE] Insert 4:3 image: side-by-side of two insight copy variants from usability testing—highlight the advisory tone winner.

The Outcome

We shipped the modernized navigation and insights module in phased release:

Signal	Result (pilot cohort, 12 weeks)
Insights weekly active use	34% of eligible accounts
Support tickets on “unknown transactions”	−12% vs. prior quarter
Task success on transfer + bill pay (retest)	91% (up from 74%)
Design system adoption across squads	4 product teams on shared Figma library

The four-person design squad I led delivered reusable patterns—insight cards, confidence chips, transaction drill-down—that became the bank’s template for future intelligent features (including a limited pilot of in-app Q&A that reused the same trust patterns).

[!NOTE] Insert 16:9 image: design system page in Figma—insight card component with variants for high / low confidence and loading states.

What this case study shows: enterprise-scale mobile UX, legacy modernization without a “big bang,” and design for AI where transparency and reversibility matter more than novelty.