About the client
We have been successfully collaborating with our client, SGB-Bank, for years, continuously developing the SGB Mobile application. The results of this partnership have been consistently recognized by the Mobile Trend Awards jury, earning multiple nominations over the years—including this year for Ailleron's transaction classification solution.
One of the key features available in SGB Mobile is personal finance management (PFM), helping users track and organize their spending. While the bank already offered a PFM tool, it didn’t fully meet customer expectations:
- The rule-based logic misclassified 3-5 out of 10 transactions.
- The presented data required interpretation and wasn’t always intuitive for users. To improve this experience, SGB-Bank partnered with Ailleron to enhance its PFM capabilities and make financial management easier for customers.
Our experts designed and implemented the Ailleron Transaction Classifier (ATC) – an AI-powered solution. By leveraging Machine Learning (ML) models, the bank’s customers can now manage their budgets more effectively with automatically categorized transactions and clearer insights into their finances.
Ailleron Transaction Classifier analyses all customer transactions:
bank transfers
card payments, including subscriptions
market-specific payment types, such as BLIK in Poland
After that, it assigns the transactions to the appropriate categories. Private customers are provided with:
structure of their spending over time
active hints, alerts, or notifications to master their budget management
The main idea behind the project was to improve the classification of a bank’s private customer transactions.
The cooperation
business workshops with the bank’s team
to chose the relevant transaction categories
Ailleron Team designed the AI-powered solution
to match customer transaction attributes with predefined categories
setting the scope of Machine Learning model training
and start labeling anonymized data to in line with desired categories
preparing UX/UI design
to blend the new personal finance management (PFM) features into banks’ existing mobile application
Solution:
Our goal was to increase engagement among mobile banking app users by presenting customer transactional data conveniently. The project will bring significant short-term business benefits to the cooperative bank, including:
- increased mobile app engagement with a boost in logins and enhanced in-app activity after introducing new features,
- strengthened customer loyalty and trust by prioritizing their needs.
From a business perspective, structured customer spendings data are useful for:
- advanced analytical purposes - better customer segmentations and product recommendations
- enriching customer data
- building event-driven notifications for customers based on the history of their transactions
Implementation of ATC also allows for enrichments in customer segmentation.
Understanding resemblances in spending patterns enables identifying groups of similar customers accurately.
List of deliverables for the bank during the Ailleron Transaction Classifier project
- ML model as a microservice was prepared for implementation on the bank’s infrastructure
- Integrations with
- customer's transactions sourced within bank systems
- mobile banking app
- Event-driven architecture for delivering real-time categorizations
- Data streaming platform based on Apache Kafka for gathering customer transactions and delivering to transactions classifier
- No-SQL data base –MongoB
- Tools responsible for monitoring, tracing, security, and log management: Graphana, Proetheus, ISTIO, Zipkin, Flunent ID etc.
- UX/UI to blend the new personal finance management (PFM) features into banks’ existing mobile application.
Ailleron Transaction Classfier assigns customer transactions to one of the defined groups:
How it works?
car & transportation
cash
entertainment, hobby & travels
education
daily spending's
health & beauty
clothes
savings and investments
income
other incoming transfers
Some of the transaction attributes analyzed by ATC include:
amount spent
description or title
recipient
currency
Merchant Category Code (MCC)
Key results of the project – top-notch accuracy
ATC divides customer transactions into 13 groups
- 11 groups for spending
- 2 groups of incoming transactions
The accuracy of the ML-model powering ATC is
- 92%, which means 92% of customer transactions are properly qualified (for transactions with or without MCC)
- 98% for the value of correctly qualified customer transactions, which means minor mistakes could occur for low-amount transactions.
We love data challenges! Let’s talk about how we can help your organization turn data into value.
Dawid Klempka
General Manager Financial Services Tribe