Case study
A Gen-AI powered tool to reduce cognitive load for fraud analysts and improve detection workflows during live customer calls.
Client
Major UK retail bank
My role
Lead Designer
Type
Enterprise Application Design
Duration
6 weeks
AI Product Design
Design Leadership
Human-AI Interaction
Enterprise UX
Workflow Design
Overview
Fraud detection relies heavily on the quality of human judgement under pressure. During customer calls, fraud analysts are required to listen carefully, identify suspicious signals, take detailed notes, and summarise outcomes — all in real time.
At one of the UK's largest retail banks, this created a fundamental tension. The more time analysts spent capturing and summarising information, the less cognitive capacity they had to detect and assess potential fraud.
I led the design of a Gen-AI powered Call Assist tool for the Economic Crime Prevention team, helping analysts reduce cognitive load during calls while improving their ability to focus on fraud detection. The solution combined AI transcription and summarisation with structured outputs designed to integrate into the bank's existing enterprise systems.
The problem
Fraud call handlers were required to multitask heavily during customer interactions. They needed to listen actively, identify potential fraud signals, take detailed notes, and summarise each call for downstream processing.
The more analysts focused on documentation, the less capacity they had for the thing that mattered most — detecting fraud.
This created a high cognitive load at a critical moment in the workflow. Manual note-taking and summarisation reduced the analyst's ability to focus on detecting fraud, increasing the risk of missed signals and inconsistent documentation. From a business perspective, this introduced both operational inefficiency and potential financial risk.
Why this matters
Fraud has a direct financial impact on the organisation. Improving detection capability can deliver significant cost savings. At the same time, improving efficiency in call handling reduces operational overhead and allows analysts to focus on higher-value decision-making.
The opportunity was not just to automate note-taking, but to improve the quality of human judgement by reducing unnecessary cognitive load.
My role
I was the Lead Designer, responsible for directing the design workstream. I worked closely with the engagement lead, AI solution experts, and technical leads to shape the overall solution. I led the design team, briefing and delegating work to two designers, reviewing outputs, and ensuring alignment with both user needs and technical constraints.
My responsibilities spanned interrogating the initial brief, structuring the design approach, leading discovery sessions, aligning with stakeholders, and presenting outputs to both internal and client teams.
Working within constraints
The project operated within a tight set of constraints that shaped every design decision.
Understanding the workflow
To design effectively, we needed to understand the real working environment of fraud analysts. We conducted discovery sessions, live call listening, and user reviews with analysts at different levels of seniority.
This helped us understand how calls were handled, how notes were taken, how summaries were created, and how information flowed into downstream systems. The key insight was clear: analysts were forced to split their attention between listening, analysing, and documenting — reducing their effectiveness at the most critical moment of fraud detection.
Designing for AI-assisted workflows
The goal was not to replace analysts, but to support them. We defined a set of principles to guide the AI-assisted experience.
1
Reduce cognitive load during live calls
2
Support, not replace, human judgement
3
Provide clear visibility into AI outputs
4
Allow users to verify and correct AI-generated content
5
Integrate seamlessly into existing workflows
The solution
We designed a Gen-AI powered Call Assist tool integrated into the bank's fraud dashboard. The solution included automated transcription of customer calls, AI-generated summaries, a familiar chat-based interface for reviewing conversations, and structured outputs tailored for integration with enterprise tools.
The interface leveraged familiar interaction patterns to reduce the learning curve, allowing analysts to focus on content rather than interface complexity. Customisable exports based on downstream system requirements ensured the tool worked across the full fraud operations workflow.
Designing for trust and error
Designing for AI required careful consideration of trust. We knew the system would not be perfect, so we designed for transparency and control — users could review and edit AI-generated summaries, the interface made it clear when transcription and summaries were available, and outputs could be validated before being used downstream.
Because real-time transcription was not technically feasible, we designed alternative feedback mechanisms to maintain user confidence — visual cues that reassured analysts recording and processing were in progress.
Trade-offs
The most significant trade-off was the lack of real-time transcription. This limited our ability to support in-the-moment validation during calls. Instead, we focused on post-call workflows and designed clear feedback mechanisms to reassure users that processing was underway.
Given the six-week timeline, we prioritised delivering a high-quality, implementable solution over exploring more advanced but less feasible capabilities.
Impact
The designs were well received by stakeholders and were implemented by the bank's internal development teams.
Analysts freed from manual note-taking during live calls, improving their capacity to detect fraud signals.
Designs handed off to and built by the bank's internal development teams for wide use across the bank.
Structured AI-generated outputs replaced inconsistent manual summaries, improving downstream data quality.
Established a pattern for AI-assisted tooling that could be extended across fraud and wider financial crime workflows.
Reflection
This project reinforced the importance of clarity and structure when leading design in fast-moving, ambiguous environments. With a six-week timeline, it was critical to define the problem clearly, structure the work effectively, and delegate with confidence.
I focused on creating the right conditions for the team to succeed: clear briefs, regular reviews, and alignment with technical and business stakeholders.
I also developed a stronger understanding of where I add the most value as a design leader — shaping direction, aligning teams, and ensuring quality — rather than trying to own every design output.
My contribution
Design leadership
Discovery
Stakeholder alignment
Team direction
Project planning and management
Senior reporting
Collaborators
Engagement Lead
Project direction & client relationship
AI Solution Experts
Technical feasibility & AI architecture
Fraud Operations
Journey design
Design Team
UI design & detailed execution
Methods Used
Discovery sessions
Live call listening
User reviews
AI workflow design
Stakeholder presentations
Enterprise integration design
Oscar Choi
© 2026 · Irvine, California