Streamlining complaints with conversational AI

At Lloyds, the Black Horse customer dashboard app was migrating from a legacy platform to a new platform. Rather than replicating the standard web form used for complaints, I saw an opportunity to design a more empathetic experience that reflected the emotional nature of submitting a complaint.

Tasks
User Experience
User Interface
User Testing
Prototyping

Client
Black Horse
(Lloyds Banking Group)

Year
2025

The challenge

Blackhorse was transitioning to a new platform, but not all journeys were migrated at launch. One key gap was the complaints process—customers could no longer submit complaints online and had to call instead.

This created frustration for customers, who faced long wait times, and increased workload for agents, who spent up to 45 minutes manually logging complaints into Nucleus.

My first task was to map the AS-IS and TO-BE journeys through a service blueprint. I also worked with the analytics team to review data from the legacy platform.

The old 4-page form had a high abandonment rate. It was long, lacked an upload function for evidence, and required customers to send a separate email before submitting—another major drop-off point.

Reimagining the journey

I decided not to simply bring over the old WSS complaints form. After researching competitors and discussing with stakeholders, we agreed that the new experience should use a conversational bot to guide users through the same series of required questions.

The aim was to:

  • improve customer engagement,

  • reduce operational strain,

  • and create a framework for future AI integration.

Powered by Gemini

Gemini was chosen as the LLM since it's already in use across the banking group.

We defined key parameters for the bot, including the types of complaints it handles (e.g. vehicle, service, finance, payments), actions it can take (respond, escalate, refund), and when to involve a human.

To align it with company policy, we provided relevant artifacts like the complaint handling policy, FAQs, escalation procedures, and agent scripts.

A conditional script to cover different scenarios

I worked closely with the content designer to polish the script structure and make sure it followed the tone of voice and policies of the company.

Components

I created new components to support the bot interactions, building on the existing Lloyds bot design. I enhanced it by adding features like voice-to-text to improve accessibility and user experience.

Testing and iteration

Two rounds of usability testing were conducted with real users to validate tone, ease of use, and overall trust. Based on feedback, we refined the bot’s tone to be more empathetic and ensured the interaction felt intuitive and supportive during what can be an emotionally charged task.

Doc upload

Key improvements

  • Conversational UX: I designed a more empathetic dialogue flow that collected all required information. Conditional logic enabled personalised journeys based on user responses.

  • System Integration: The bot captures all responses and feeds them into Nucleus, a colleague-facing portal used to manage and review complaints.

Results we're driving toward

Increase completion rate to

70%

Old platform rate was 50%

Reduce complaint-related calls by

90%

within 6 months

Decrease time spent logging complaints from

45min to 0

thanks to automated data capture