SC GPT
SC GPT
The first bank-wide conversational AI chatbot deployed across Standard Chartered.
Problem
Building the bank's first enterprise-wide conversational AI tool - with no playbook to follow
When conversations around large language models began reshaping industries, Standard Chartered saw an opportunity to explore what they could unlock for the bank. The challenge wasn't simply building a chatbot - it was figuring out how to do so responsibly, at enterprise scale, in one of the world's most regulated industries. No team at the bank had gone down this path before, and it was left to us to blaze the path.
My role
Sole designer from concept through launch, responsible for research, wireframes, prototypes, and iteration.
Team
Cross-functional team spanning product, engineering, security, and data.
Process
From conversations and constraints, to a live product
4 parallel tracks shaped what SC GPT became, and each one pulled design in a different direction at the same time.
Landscape research
We scanned leading LLM products, and learnt from internal chatbot teams to build an initial feature wishlist, to understand what was possible before deciding what to pursue.
Security set the boundaries
No chat history, no file uploads, no persistent memory. Each was ruled out by our security teams. No precedent existed, and there wasn't enough time to evaluate them safely. Those constraints shaped everything that followed.
Built the foundation from scratch
No conversational UI components existed in the bank's design system. Every pattern was designed from the ground up, validated with the design system team, and contributed back - creating a foundation future AI products could build on.
Design beyond the screen
An over-rigid sensitive data detection model was blocking legitimate user inputs. Working with data and security to fix it was an early signal that in this environment, design decisions rarely stop at the interface.
Outcome
Shipped to 40,000 colleagues across 40 countries
40k
Peak users at launch
~20k
Active users post-launch
40
Countries shipped across
SC GPT launched as a fully realised product, covering onboarding, the chat experience, error handling, and sensitive data controls. The conversational UI components developed for this project were later adopted into the bank's central design system, creating a shared foundation for future AI products.
Reflection
What designing at the frontier actually teaches you
The designer as customer advocate
Being the sole designer meant handling customer support too. Responding to queries, managing frustration, and advocating for fixes put me closest to what was actually breaking - and made every design argument far more grounded in real evidence.
Developer rapport isn't optional
When a director pushed for chat history without design sign-off, a developer built a static table version - technically 'chat history', but unreadable and unusable. Keeping designers in the loop prevents well-intentioned but misaligned decisions from shipping.
Don't assume your users' baseline
We assumed users unfamiliar with chatbots were a minority, since chatbots have been around since the early 2000s. They weren't. The response was tiered support: videos and learning pathways for self-directed learners, and 1-on-1 sessions for those who were truly new to chatbot interfaces.