Transaction Intelligence
The Problem
One of Australia's largest banks was losing ground to its own self-service platform. Despite significant investment in digital banking, over 11,000 calls per month — 14% of total business banking call volume — were coming in for issues that should have been resolvable without human support. Inquiries took up to two weeks to process, customers received no automated updates, and the platform ratings reflected the frustration. The technology existed to do better. What was missing was the design thinking to make it work.
The Brief
Lab49 brought me in as the sole UX lead to conceptualize and prototype an AI-powered transaction intelligence tool, built in collaboration with AWS Bedrock and Anthropic's Claude LLM. The goal was clear: give customers real-time access to the kind of transaction insight and analysis that previously required a phone call — and design an experience that made that intelligence feel instant, trustworthy, and genuinely useful.
The Design Challenge
This wasn't a standard dashboard problem. The core challenge was designing for AI-generated, dynamic interfaces — where the UI itself needed to respond differently for every transaction, every user, and every query intent. There was no fixed screen to design. There was a system to design.
Three design problems needed to be solved simultaneously: how to surface the right transactions without overwhelming the user; how to present AI-generated analysis in a way that felt clear and credible rather than opaque or alarming; and how to reduce cognitive load at exactly the moment a customer is anxious about their money.
The Design Response
The solution centered on a concept called Transaction Intelligence — a priority-first transaction view that uses AI to identify which transactions genuinely need a customer's attention, based on behavioral patterns, merchant familiarity, location, timing, and risk signals.
Rather than presenting a flat list of transactions, the interface surfaced a curated priority view. Each flagged transaction was paired with an "intent" — the specific reason a customer might call support — which was then used to prompt the AI to generate a bespoke analysis of that transaction, assembled from all available data. The result was a unique, dynamically generated screen for each transaction: not a template, but a genuinely contextual experience.
To make risk legible without creating alarm, I introduced a trust score — a visual indicator inspired by familiar consumer heuristics like Google Flights' price guide — that communicated transaction confidence on a spectrum. Critically, the trust score was designed to learn: as customers resolved queries, confirmed known transactions, or raised disputes, the system adapted its future prioritization accordingly. The UI became progressively smarter the more it was used.
A conversational AI layer sat beneath the surface for moments requiring escalation — allowing customers to resolve disputes, lock cards, and get further help through natural language interactions, without ever needing to call.
The full design scope covered: AI-generated dynamic UI components, transaction prioritization framework, trust score system, conversational AI flows, merchant intelligence panels, and the path-to-production considerations required to move from concept to a hardened, accessible, bank-grade product.
Outcome
The six-week proof of concept was completed on schedule and received strongly positive feedback from the client sponsor. A Path to Production document was produced to guide next steps — including data integration, security hardening, accessibility validation, and channel integration. Follow-on workshops were scheduled to move the concept toward production.
The engagement demonstrated something important: that AI in financial services doesn't have to feel impersonal or intimidating. Designed well, it can be the most reassuring thing a bank has ever put in front of a customer.
Comprehensive case studies and design assets are available for private review upon request. These documents contain sensitive client information and are withheld from public display to ensure data security and confidentiality.
Transaction Intelligence
The Problem
One of Australia's largest banks was losing ground to its own self-service platform. Despite significant investment in digital banking, over 11,000 calls per month — 14% of total business banking call volume — were coming in for issues that should have been resolvable without human support. Inquiries took up to two weeks to process, customers received no automated updates, and the platform ratings reflected the frustration. The technology existed to do better. What was missing was the design thinking to make it work.
The Brief
Lab49 brought me in as the sole UX lead to conceptualize and prototype an AI-powered transaction intelligence tool, built in collaboration with AWS Bedrock and Anthropic's Claude LLM. The goal was clear: give customers real-time access to the kind of transaction insight and analysis that previously required a phone call — and design an experience that made that intelligence feel instant, trustworthy, and genuinely useful.
The Design Challenge
This wasn't a standard dashboard problem. The core challenge was designing for AI-generated, dynamic interfaces — where the UI itself needed to respond differently for every transaction, every user, and every query intent. There was no fixed screen to design. There was a system to design.
Three design problems needed to be solved simultaneously: how to surface the right transactions without overwhelming the user; how to present AI-generated analysis in a way that felt clear and credible rather than opaque or alarming; and how to reduce cognitive load at exactly the moment a customer is anxious about their money.
The Design Response
The solution centered on a concept called Transaction Intelligence — a priority-first transaction view that uses AI to identify which transactions genuinely need a customer's attention, based on behavioral patterns, merchant familiarity, location, timing, and risk signals.
Rather than presenting a flat list of transactions, the interface surfaced a curated priority view. Each flagged transaction was paired with an "intent" — the specific reason a customer might call support — which was then used to prompt the AI to generate a bespoke analysis of that transaction, assembled from all available data. The result was a unique, dynamically generated screen for each transaction: not a template, but a genuinely contextual experience.
To make risk legible without creating alarm, I introduced a trust score — a visual indicator inspired by familiar consumer heuristics like Google Flights' price guide — that communicated transaction confidence on a spectrum. Critically, the trust score was designed to learn: as customers resolved queries, confirmed known transactions, or raised disputes, the system adapted its future prioritization accordingly. The UI became progressively smarter the more it was used.
A conversational AI layer sat beneath the surface for moments requiring escalation — allowing customers to resolve disputes, lock cards, and get further help through natural language interactions, without ever needing to call.
The full design scope covered: AI-generated dynamic UI components, transaction prioritization framework, trust score system, conversational AI flows, merchant intelligence panels, and the path-to-production considerations required to move from concept to a hardened, accessible, bank-grade product.
Outcome
The six-week proof of concept was completed on schedule and received strongly positive feedback from the client sponsor. A Path to Production document was produced to guide next steps — including data integration, security hardening, accessibility validation, and channel integration. Follow-on workshops were scheduled to move the concept toward production.
The engagement demonstrated something important: that AI in financial services doesn't have to feel impersonal or intimidating. Designed well, it can be the most reassuring thing a bank has ever put in front of a customer.
Comprehensive case studies and design assets are available for private review upon request. These documents contain sensitive client information and are withheld from public display to ensure data security and confidentiality.
Transaction Intelligence
The Problem
One of Australia's largest banks was losing ground to its own self-service platform. Despite significant investment in digital banking, over 11,000 calls per month — 14% of total business banking call volume — were coming in for issues that should have been resolvable without human support. Inquiries took up to two weeks to process, customers received no automated updates, and the platform ratings reflected the frustration. The technology existed to do better. What was missing was the design thinking to make it work.
The Brief
Lab49 brought me in as the sole UX lead to conceptualize and prototype an AI-powered transaction intelligence tool, built in collaboration with AWS Bedrock and Anthropic's Claude LLM. The goal was clear: give customers real-time access to the kind of transaction insight and analysis that previously required a phone call — and design an experience that made that intelligence feel instant, trustworthy, and genuinely useful.
The Design Challenge
This wasn't a standard dashboard problem. The core challenge was designing for AI-generated, dynamic interfaces — where the UI itself needed to respond differently for every transaction, every user, and every query intent. There was no fixed screen to design. There was a system to design.
Three design problems needed to be solved simultaneously: how to surface the right transactions without overwhelming the user; how to present AI-generated analysis in a way that felt clear and credible rather than opaque or alarming; and how to reduce cognitive load at exactly the moment a customer is anxious about their money.
The Design Response
The solution centered on a concept called Transaction Intelligence — a priority-first transaction view that uses AI to identify which transactions genuinely need a customer's attention, based on behavioral patterns, merchant familiarity, location, timing, and risk signals.
Rather than presenting a flat list of transactions, the interface surfaced a curated priority view. Each flagged transaction was paired with an "intent" — the specific reason a customer might call support — which was then used to prompt the AI to generate a bespoke analysis of that transaction, assembled from all available data. The result was a unique, dynamically generated screen for each transaction: not a template, but a genuinely contextual experience.
To make risk legible without creating alarm, I introduced a trust score — a visual indicator inspired by familiar consumer heuristics like Google Flights' price guide — that communicated transaction confidence on a spectrum. Critically, the trust score was designed to learn: as customers resolved queries, confirmed known transactions, or raised disputes, the system adapted its future prioritization accordingly. The UI became progressively smarter the more it was used.
A conversational AI layer sat beneath the surface for moments requiring escalation — allowing customers to resolve disputes, lock cards, and get further help through natural language interactions, without ever needing to call.
The full design scope covered: AI-generated dynamic UI components, transaction prioritization framework, trust score system, conversational AI flows, merchant intelligence panels, and the path-to-production considerations required to move from concept to a hardened, accessible, bank-grade product.
Outcome
The six-week proof of concept was completed on schedule and received strongly positive feedback from the client sponsor. A Path to Production document was produced to guide next steps — including data integration, security hardening, accessibility validation, and channel integration. Follow-on workshops were scheduled to move the concept toward production.
The engagement demonstrated something important: that AI in financial services doesn't have to feel impersonal or intimidating. Designed well, it can be the most reassuring thing a bank has ever put in front of a customer.
Comprehensive case studies and design assets are available for private review upon request. These documents contain sensitive client information and are withheld from public display to ensure data security and confidentiality.