Supporting financial health with a GenAI chatbot
KickstartAI x ING
ING and KickstartAI are leveraging Generative AI (GenAI) to deliver two key outcomes: 1) making personalized financial health insights more accessible, and 2) transforming static financial information into actionable, conversational guidance. This initiative will help customers manage their financial wellbeing with greater confidence and ease through natural language interaction.
The Problem
Managing personal finances is often perceived as complex, time-consuming, and , for many, overwhelming. Within the Netherlands, the scale of this challenge is significant: 47% of households struggle to keep a clear overview of their financial situation or feel they lack the tools to manage their money effectively. While banks provide a wealth of data – such as transaction histories and spending categorizations – this information is often fragmented or presented in static formats that require manual interpretation.
In practice, understanding one's financial health requires a customer to manually track expenses, compare budgets, and identify patterns over time. When users have specific questions about their spending habits or are looking to be educated on improving their financial resilience, they often have to navigate complex FAQs or wait for human support. This friction contributes to the high percentage of people feeling disconnected from their financial data.
With the rapid advancement of Large Language Models (LLMs), there is a significant opportunity to bridge this gap. ING and KickstartAI are partnering to develop a GenAI solution that can understand financial queries and provide tailored, human-like responses, ultimately helping the nearly half of the population currently struggling to regain control over their financial health.
Goals
The goal of the project is to develop a GenAI-powered assistant that can analyze a customer’s financial data in real-time to provide personalized insights, answer complex financial health questions, and eventually perform actions to simplify money management.
Approach
The approach consists of building three important parts:
Architecture & Customization: Designing a Retrieval-Augmented Generation (RAG) framework that allows the LLM to access specific documentation on financial health from ING, Nibud & Geldfit.
Guardrails & Privacy: Rigorous testing is necessary to ensure the model functions as an educational tool rather than a source of financial advice. A critical requirement is obtaining user consent before the assistant can access the user's financial data. This dual approach guarantees the provision of accurate and compliant financial information.
Scaling & Agency: Moving from an informational tool (answering questions) to an "action-oriented agent" that can execute tasks like turning automatic savings on or setting up budget alerts directly within the chat.
The KickstartAI x ING team has focused heavily on the evaluation phase, testing the model with "wrong" or difficult questions to ensure reliability.
Key features
The AI system we are developing consists of the following components:
Natural Language Understanding: Allowing users to ask questions about their finances in plain language (e.g., "How much did I spend on groceries compared to last month?") rather than searching through menus.
Contextual Data Retrieval: Seamlessly connecting the assistant to the user’s transaction history and financial profiles to provide answers rooted in real, personal data.
Action-Taking Capabilities: Developing "agentic" features where the assistant can perform concrete actions, such as resolving issues end-to-end without redirecting the user.
Expected Results
The two main questions we hope to answer are:
Can we provide more tailored financial support to more people through a Financial Assistant? Assessing the assistant's effectiveness in supporting users who engage proactively, measuring the successfully diverted volume of users who avoid entering a formal arrears state.
Can we help reduce financial stress for ING customers? Measuring the long-term financial stability of users following their interaction with the assistant, specifically tracking their ability to maintain a healthy account status over an extended period.