Introduction

Cubematch is an Irish digital transformation consultancy founded in Dublin in 2002. With offices in Ireland, the UK, Singapore, Spain, Germany, Switzerland and India, Cubematch service an international client base through their bespoke offerings in the areas of strategic change management, digital transformation, automation, compliance, data and AI. Cubematch employs AI consultants, but this does not remove the need for independent innovation support. CeADAR’s role was not to provide routine AI consultancy but to support test-before-invest experimentation, access to expertise, and the de-risking of new AI capabilities.

Challenges

The challenge Cubematch faced was building an AI-based banking chatbot prototype which could be showcased to multiple banks to demonstrate the the possibilities of AI-based user interactions and user interfaces. The motivation for developing this prototype was to demonstrate the feasibility of an AI chatbot capable of providing quick access to customer banking information and insights. Traditional digital banking interfaces often require users to navigate multiple menus, dashboards, and account views to access information such as balances, spending summaries, transaction histories and recurring debit status. This creates friction for users who may prefer a simpler conversational interface, and ultimately quicker access and insights.

Solution Implemented

The CubeMatch AI banking chatbot prototype is a web-based conversational banking prototype that demonstrates how a Large Language model (LLM) can be used to provide customers with easy to access account information, transaction insights, spending analytics, and simulated banking actions through a simple chatbot interface. This prototype was developed to address some of the following problems:

1. Limited ability to access banking data through conversational UI navigation
2. Lack of conversational access to account-level financial insights
3. Difficulty demonstrating to enterprise stakeholders what future AI-enabled banking experiences will actually look and feel like in practice.

On login, the user arrives at a landing page made to look and feel like a real online banking interface. From there, they head to the chat page, where they can ask their own question or use one of the suggested prompts to get started. As they chat, the assistant understands what they’re asking, whether it’s a general banking question, a look at their account balance or recent transactions, a breakdown of their spending with simple charts, or a request to transfer money. The prototype responds naturally by retrieving the relevant information from the test banking database as needed.

If the user asks to make a transfer, the assistant guides them through it step by step and includes a mock security verification screen, simulating a real banking app user authentication experience. Users can also look back at previous conversations or start a new one at any time. Overall, the experience feels like texting a knowledgeable customer service agent who can see and explain your bank account for you. The prototype incorporated governance controls aligned with the principles of the EU AI Act. It was designed to provide factual account information and assist with routine banking enquiries, while explicitly preventing functionality that could constitute high-risk AI, such as creditworthiness assessments or lending decisions.

While the prototype demonstrates functional feasibility, it has been developed using a combination of low-code/no-code orchestration tools and cloud-based services optimised for rapid prototyping rather than production deployment. Scaling this solution to a production grade environment would require further engineering work, including migration to the hosting infrastructure required by the bank (dedicated cloud, private cloud, or on-premises), implementation of enterprise grade security controls, performance optimisation, and integration with core banking systems. For production roll-out, the assistant functionality would include the mechanism to escalate queries that can’t be fulfilled by the assistant to a real customer service agent for resolution.

Results and Benefits

The prototype provided Cubematch with a successful proof of concept, enabling the company to assess the feasibility of an AI-powered banking chatbot and inform decisions on future product development. The resulting demonstrator has subsequently been showcased at a number of high-profile government-led events across Europe, helping to showcase the innovation to international stakeholders.  The prototype can function as an AI-driven assistant for handling common banking queries. It demonstrates how AI can enhance customer service by supporting routine enquiries, improving response times and consistency, and enabling staff to focus on higher-value customer interactions.

Disclaimer: The prototype described in this case study was developed and evaluated using synthetic test datasets designed to replicate the structure and characteristics of banking data. No real customer, financial, or personally identifiable information (PII) was used during the development, testing, or validation of the solution. Any references to banking scenarios are illustrative and intended solely to demonstrate the prototype’s technical capabilities. This ensured that experimentation can take place in a controlled environment, while still allowing stakeholders to explore the impact of AI-enabled banking interfaces.