Building a Chatbot Prototype with Progressive Credit Union

Introduction 

Progressive Credit Union is a Dublin based Credit Union with branches in Balbriggan, Baldoyle, East Wall, Fairview, Glasnevin, Rush and Swords. Founded in 2013, Progressive Credit Union has grown considerably over the past number of years, merging with local branches and massively increasing their community reach across central and suburban Dublin. They currently cater to a membership of over 60,000 people, providing valuable financial services across a wide range of community-focused products. They provide local communities with affordable financing options through student current accounts, car loans, personal loans, mortgages, death benefit insurance etc.

Challenge 

As membership numbers grew, so did the volume of routine queries reaching Progressive’s customer agents. Questions about branch hours, account requirements, and general policy information that, while simple to answer, took up a disproportionate share of member service staff time. This left less capacity for the staff to close out more complex customer queries.

The challenge Progressive Credit Union faced when deciding to explore AI for their operations was to make sure it benefited their members directly, and in a user friendly, accessible way. Through consultation with CeADAR, the decision was made to build a Retrieval Augmented Generation (RAG) chatbot prototype which could sit on the organisation’s website. This could be queried by members to provide context-driven answers to queries related to credit union policies and general branch information.

Solution Implemented

After an initial consultancy phase, it was decided that the RAG chatbot prototype would be built using a basic LLM architecture, with the source documents used to direct the prototype’s answers consisting of a variety of FAQs and standard information related to the credit unions day-to-day operations and policies. A deliberate design boundary was set from the outset: the chatbot would not be used to determine an individual’s eligibility for loan approval, as this would trigger EU AI Act high-risk obligations. Instead, it was designed to provide informational support only, with complex or unresolved queries being escalated to a human member service officer where appropriate. The prototype supported different user roles for members, member service staff and administrators, ensuring users only accessed information appropriate to their role. 

Users would be able to query the chatbot and get quick, up to date, relevant information regarding different branches of Progressive Credit Union across Dublin. To ensure members were not left without support when the chatbot could not satisfactorily answer a query, the prototype also included an intelligent escalation mechanism. Where a member continued to ask questions without receiving the information they needed, the system prompted them to escalate the conversation to a member service officer for a call back. The full conversation history was then made available to staff through a dedicated administration portal, allowing them to quickly review the context before responding. 

The aim of the project was to ease the administrative workload of member service officers. PCU staff tested the prototype and found that approximately 90% of general inquiries could be answered by the chatbot, and only 10% of these questions actually required downstream engagement with a live agent. Responses referenced the underlying policy documents used to generate the answer, helping users understand where information originated and increasing confidence in the chatbot’s responses. 

Examples of commonly asked questions the chatbot was positioned to answer included:

• What are the opening hours for the Balbriggan branch?
• What documents do I need to open a student account?
• What are the current personal loan interest rates?
• How do I apply for death benefit insurance?

The system was purposefully built using a document ingestion pipeline in SharePoint alongside an administration interface that allowed authorised users to manage the knowledge base. The prototype also provided visibility into the questions being asked, helping identify gaps in documentation and opportunities to improve customer information over time. This helped to future proof the system, allowing document content to be improved and old documents to be easily replaced with new ones so as to guarantee customers weren’t receiving out of date information. This is especially important when it comes to information regarding interest rates and loan criteria.

Results and Benefits

This prototype has been a successful proof of concept, allowing Progressive Credit Union to test a practical framework for future implementation of a RAG chatbot to be hosted on their website. It has successfully demonstrated the time and labour saving opportunity of developing a fully functioning RAG chatbot. Some of the benefits experienced during testing are listed below:

• Reduced administrative workload: Agents spent less time on repetitive, low-complexity questions and more time on higher-value member tickets.
• Human-in-the-loop support: Rather than attempting to answer every question, the chatbot recognised when additional assistance was required and prompted members to escalate to a live agent. Agents could review the complete conversation history through an administration portal, allowing them to continue the interaction with full context while maintaining member service quality. 
• Future-proof architecture: The SharePoint-based ingestion pipeline let administrators update source material directly, keeping chatbot responses accurate as policies and rates changed.

By validating the technical approach before committing to full-scale development, the prototype has reduced implementation risk and provided Progressive Credit Union with a clear blueprint for a production solution. The organisation is now assessing the most appropriate production implementation approach.