• Lloyds

  • Banking

  • Group

[THE CLIENT]

Lloyds Banking Group

[THE YEAR]

2023

[THE PROBLEM]

With over 20 million digital users, Lloyds faced mounting pressure: fragmented journeys, keyword-reliant search, and long wait times left millions of service queries unresolved. Customer trust was slipping.


IBM saw an opportunity to reimagine digital servicing. We proposed a GenAI-powered Search Bar, the UK’s first in banking to deliver fast, human-like support at scale.

[THE ROLE]

As User-Centred Design Lead, I shaped IBM’s experience strategy and led the design of four conversational AI proof-of-concepts, collaborating across design, development and data science over 4 weeks in parallel with a full rebrand.


I drove discovery, co-creation, and C-suite demos, helping IBM secure funding and scale GenAI within financial services for future endeavours.

[THE IMPACT]

  • 91%+ AI accuracy, up from 60–75% industry average

  • £1M+ in annual savings through automation

  • 25% uplift in resolution rates within 3 months

  • 6-figure contract extensions secured quarterly until go-live

  • LLM design patterns now reused across FS accounts

  • 🏆 Highly Commended – MCA Awards 2024

  • 🏆 Finalist – MCA Awards 2023

Designing the UK's First

Conversational AI in Banking

Designing the UK's First

Conversational AI in Banking

IT WASN'T JUST A SEARCH BAR, IT WAS THE FRONT DOOR TO THE BANK

We reimagined the first touch as a conversation, not a menu. One input that understands intent, shows the right context, and gets you to action in a single flow. “Show my direct debits.” “Why £29.99 yesterday?” “Can I overpay my mortgage?” Questions in. Outcomes out.

IT WASN'T JUST A SEARCH BAR, IT WAS THE FRONT DOOR TO THE BANK

We reimagined the first touch as a conversation, not a menu. One input that understands intent, shows the right context, and gets you to action in a single flow. “Show my direct debits.” “Why £29.99 yesterday?” “Can I overpay my mortgage?” Questions in. Outcomes out.

IT WASN'T JUST A SEARCH BAR, IT WAS THE FRONT DOOR TO THE BANK

We reimagined the first touch as a conversation, not a menu. One input that understands intent, shows the right context, and gets you to action in a single flow. “Show my direct debits.” “Why £29.99 yesterday?” “Can I overpay my mortgage?” Questions in. Outcomes out.

DESIGNED FOR REAL MOMENTS. BUILT TOGETHER.

Money is emotional – unknown charges, frozen cards, missed payments. Replies were written to calm first and guide next with plain English and clear verbs. I partnered daily with content, data science and engineering to co-author prompts, validate flows against real scenarios, and keep compliance and accessibility in the room from day one.

DESIGNED FOR REAL MOMENTS. BUILT TOGETHER.

Money is emotional – unknown charges, frozen cards, missed payments. Replies were written to calm first and guide next with plain English and clear verbs. I partnered daily with content, data science and engineering to co-author prompts, validate flows against real scenarios, and keep compliance and accessibility in the room from day one.

DESIGNED FOR REAL MOMENTS. BUILT TOGETHER.

Money is emotional – unknown charges, frozen cards, missed payments. Replies were written to calm first and guide next with plain English and clear verbs. I partnered daily with content, data science and engineering to co-author prompts, validate flows against real scenarios, and keep compliance and accessibility in the room from day one.

From Frustration to Confidence.

One Conversation at a time

From Frustration to Confidence.

One Conversation at a time

SYSTEMS, NOT SCREENS

We built a reusable conversation system with intent-led taxonomy and patterns for Suggest, Confirm, Clarify, Escalate and Recover. With data scientists I co-designed prompt templates, guardrails and fallbacks that kept answers consistent, safe and on brand.

We set up lightweight evaluation with real queries to tune classification and confidence. The result was a scalable design and prompt library in Figma that teams can ship with, whether the model changes or the domain expands.

SYSTEMS, NOT SCREENS

We built a reusable conversation system with intent-led taxonomy and patterns for Suggest, Confirm, Clarify, Escalate and Recover. With data scientists I co-designed prompt templates, guardrails and fallbacks that kept answers consistent, safe and on brand.

We set up lightweight evaluation with real queries to tune classification and confidence. The result was a scalable design and prompt library in Figma that teams can ship with, whether the model changes or the domain expands.

SYSTEMS, NOT SCREENS

We built a reusable conversation system with intent-led taxonomy and patterns for Suggest, Confirm, Clarify, Escalate and Recover. With data scientists I co-designed prompt templates, guardrails and fallbacks that kept answers consistent, safe and on brand.

We set up lightweight evaluation with real queries to tune classification and confidence. The result was a scalable design and prompt library in Figma that teams can ship with, whether the model changes or the domain expands.

Your money.

In words you actually use

Your money.

In words you actually use

JARED SMITHERS

©2025

Next chapter

Goodwood