Coventry Building Society

The Situation 

One of the leading financial providers in the country 

Trusted by over 2 million people, Coventry Building Society (CBS) is the second largest building society in the UK. 

As a financial institution operating in a highly competitive market, the CBS marketing team creates a steady flow of customer communications. These must be prepared very carefully, encouraging customers to take up new products, while simultaneously providing information about complex topics like interest rates and financial regulations. 

The Problem 

Marketing financial products effectively requires grappling with complex problem 

The CBS marketing team regularly evaluates new communication strategies using carefully constructed persona profiles. Compiled using detailed information about their customer base, these personas reflect the backgrounds and perspectives of typical customers. The marketing team uses them to consider how likely different customer types will respond positively to a proposed message. While extremely useful, this process is ultimately subjective and open to bias. 

The Task 

Creating a customer persona who can have conversations with marketers 

After seeing a demonstration of Rocketmakers’ AI-powered Portrait technology, the marketing team enquired about using similar technology to animate one of their customer personas. They wondered if having an actual dialogue with a customer persona would make evaluating marketing communications easier and less subjective. 

After some initial discussions, they asked Rocketmakers to build a prototype application which, using 3D animation and AI-powered language processing, would create an animated version of one of their core customer personas - “Elizabeth.” 

Our Discovery 

Understanding the needs of the Coventry Building Society marketing team 

To fully understand the needs of the marketing team, four members of the Rocketmakers design and development teams travelled to Coventry for a half-day workshop. 

Led by one of the Rockemakers user experience (UX) designers, the workshop’s goal was to ensure the animated Elizabeth persona as user-centric as possible. Discussions focused on how the application would be used by the marketing team, what information it would need access to, and how it would help them do their jobs. 

The workshop also explored possible user journeys for employees using the app, and what part of the project lifecycle the app would be used for. 

Our Approach 

Creating a bespoke version of advanced technology 

Converting Elizabeth from a detailed text description into an animated character that has conversations was a complex task, but one that Rocketmakers’ developers were already very familiar with. 

Using its Portrait technology, Rocketmakers has been creating 3D avatars capable of real-time conversations since 2023, including Santa Claus, Pride & Prejudice’s Elizabeth Bennet, an airline check-in attendant, President Barack Obama, and Steve Jobs. 

Creating CBS’ Elizabeth persona presented a similar challenge, but with slightly different constraints and opportunities. 

To maximise access to the system, CBS wanted Elizabeth to run on a web browser. Portrait, the technology Elizabeth would be based on, uses a powerful software for rendering 3D avatars called Unity, which web browsers cannot currently support. To overcome this, the team sourced a third party integration that can render realistic looking 2D animations online in real time. 

Hosting Elizabeth on a web browser did come with an advantage the original Portrait lacked: web browsers are equipped with integrated speech-to-text readers as standard. This sped up development and reduced operating costs. 

Elizabeth’s conversations were created by a Large Language Model (LLM), which processed answers based on CBS’ in-depth persona profile. Keeping track of questions and responses was important, so the developers added Superbase, an inexpensive Postgres database, to store them. 

The Result 

Elizabeth listens and responds to questions in real time 

After four weeks of development, Rocketmakers delivered the prototype version of Elizabeth to the marketing team. 

Users press a button to record a question, which is uploaded to Elizabeth like a voice note after the recording is finished (this prevents either the user or Elizabeth from interrupting each other). 

Elizabeth generates her responses using the following steps: 

● The question is then converted to text and sent for processing by the LLM. 

● The LLM generates a text response using the CBS’ extensive personality profile, which has been carefully optimised for AI use by the Rocketmakers team. 

● This text is then converted into an audio file using Elizabeth’s virtual voice, and sent to the 3rd party integration that animates Elizabeth’s head to match the audio file. 

● Once this is complete the audio and video are played for the user on their web browser. 

The Impact 

Testing and preparing for next steps 

Elizabeth is currently being tested by fifteen members of the CBS marketing team during H1 2025. At the conclusion of the test, user data will be analysed and evaluated for ways to improve future versions of Elizabeth.