Every year in Australia, the average household quietly drops around $528 on streaming services alone, and that doesn’t count music apps, gaming passes or all the weird memberships you forgot you ever signed up for. Now imagine an average uni student renting their own place, juggling Netflix at $20 a month, Spotify at $15 a month, a mindfulness app, a fitness app, cloud storage and whatever else seemed like a great idea at the time. It’s easy to rack up well over a couple of thousand dollars a year on subscriptions before you even blink. Aussies are currently paying for more than five subscriptions at once on average, and millions are wasting up to $600 a year on services they don’t even use. Our project is all about uncovering these hidden wallet‑leaks, revealing what you’re really paying for, and turning that scattered spending into clarity and control — because right now most people don’t even know how much their subscriptions are costing them.
WTFAIPF uses advanced machine learning to scan your bank records (CSV format) and identify all likely recurring payments, from subscriptions to memberships. It doesn’t just flag them — it gives you detailed insights for each payment, including merchant, number of transactions, first and last seen dates, average amount, cadence, next billing date, status, annual cost, and subscription age. With this, you can finally see exactly where your money is going and take control of hidden recurring expenses. To test this program there is a demo file sitting in the root folder of the repository.
Our team leverages Python and a suite of powerful packages to power Subscription Leak Detector. Raw CSV bank information are managed and transformed using SQL, then analysed using scikit-learn and XGBoost to identify recurring payments. Payments with the same or similar transaction names are flagged and assessed by our custom model to determine if they are subscriptions or memberships. The entire system is wrapped in an intuitive web interface built with Streamlit, allowing users to easily upload their eStatements and gain actionable insights into hidden recurring costs.
One of our biggest challenges was handling user data. Bank statements vary widely. Different columns, inconsistent names, messy string formats, and sometimes no headers at all. Gathering accurate training data under these conditions was complex, and designing a model that could efficiently process it required careful consideration. On top of that, while existing subscription trackers already exist, we needed to innovate quickly to deliver an app that adds real value. We had to pivot ideas, build a robust model, and ensure our web interface is intuitive, aesthetically pleasing, and comfortable to use, all within a limited timeframe.
We are incredibly proud of completing our first hackathon. With a team of just four, we leveraged our skills to their fullest and built a solution that is both practical and genuinely beneficial to others.
Almost everything we accomplished was new to us, from creating a full-fledged Streamlit platform, to performing data engineering and data science, to collaborating effectively as a team. This project provided an invaluable learning experience, and each of us took away a wealth of skills and insights.
This program represents the first step in transforming user data into meaningful insights. Bank transactions hold a wealth of information, and with further development, we can help people better understand their spending habits, track their money, and discover ways to save more.
References:
DECLARATION OF AI: Our team used Google Gemini, Copilot, Claude, and OpenAI for coding assistance, and bug troubleshooting. Some code was written with AI assistance.