Bank statements can predict spending habits: Credrails and Zindi collaborate to categorise spending with AI
According to Thuso Simon, Head of Data and ML at Credrails, the solutions will enable Credrails to build a better prediction model.
“We created a model with many of the insights gained from the innovative approaches of the Zindi community, and we are now testing it against the current production model.”
Credrails is an open banking platform that specializes in enriching the data from banking statements by using machine learning models to tag and classify transactions. Every transaction has some basic information about the movement of money, but getting the context of what the movement was for requires more detailed knowledge about the transaction.
Thuso says that automatic categorisation of bank statements using machine learning can predict consumers’ spending habits and this helps the bank by understanding their customer needs better.
Understanding the merchants or people involved with the transaction can tell us if the transaction was for a purchase of groceries, or used to repay a loan. To get this context, Credrails classifies each transaction to better understand a person’s spending habits.
The Credrails Banking Transaction Categoriser Challenge lasted for three days as part of the #ZindiWeekendz Startup Innovation Series, and 70 individuals or teams submitted solutions to the competition.
Stephen Kolesh (Kenya) and Yohannes Melese (Ethiopia) emerged as the winners of the competition. According to Kolesh, using domain knowledge during feature creation helped them have the best solution. Teamwork also set them apart.
“It is said that God blesses you by sending friends into your life, therefore teaming up with a partner is very helpful in winning competitions. Teaming up with my friend Yohannes has been my biggest blessing this year. Combining our ideas is what bagged us the win.”
At Credrails, Thuso is looking forward to using these new models to generate a better context for future transactions.
“In the near future, we are looking forward to building a solution that will tag the entities in the transaction descriptions to get more context. For instance, when someone purchases food at a fuel station, our current model can assume any purchases at a fuel station is for fuel instead of food,” he says.
Visit our business page to learn how we can help you solve some of your data challenges.