Transactions Labelling and Scoring
Summary
Decision making based on the end-user’s transactions was cumbersome and prone to mistakes up until now. Kontomatik aims to make this process automatic using advanced proprietary algorithms that were built upon massive amounts of data. Our data science products can label transactions and calculate the probability of the end-user repayment based on the labels. As a result, they are a perfect extension of our Banking API.
Labeling
Our proprietary algorithm can dissect each transaction and assign appropriate labels from over 60 available (e.g. monthly, salary, compensation, loan, welfare, cash withdrawal, nightlife, rent). The assignment is based on the relationship uncovered by the labeling model in the historical data. The data has been handcrafted by human engineers based on the strict rules of defining each label.
Having labeled transactions gives you a clear and immediate picture of where the end-user spends their money and where they get it. To make this even more seamless, we’ve built the Insight portal that lets you see the results on the web page with a graphical interface.
Labeling is currently available in Poland, Spain, Czech Republic, and Portugal.
Scoring
Having labels is a great advantage, but the end-user’s repayment score is the ultimate goal. Our machine learning algorithm has been designed to achieve exactly this. It can calculate the probability of repayment based on provided labels and suggest the decision to be made. Specifically, the algorithm returns the probability, percentile and tier (e.g. A, B, C, or D).
If you already have a scoring implemented on your side, you can still use our solution - we consider our model to be a great addition to other scoring methods that supports decision making.
Scoring is available in Poland and Spain, only for clients who also use our labeling solutions.
How to integrate
Integration of our data science products is as swift as all others. If you have bought the transaction labeling feature, then the labels will automatically appear next to transactions imported using our Banking API.
To get the score, simply call the owner-scores endpoint using the id you have assigned to the end-user.
FAQ
Scoring is a method that calculates probability of someone’s repayment (score) given the labels and other transaction details. Specifically, the algorithm returns the score, percentile, and tier (e.g. A, B, C, or D). The algorithm is trained using historical data and so is able to recognize very intricate relationships not easily spotted.
If you already have a scoring implemented on your side, you can still use our solution - we consider our model to be a great addition to other scoring methods that supports decision-making.
Scoring is available only for clients who also use our labeling solutions.
If you have labelling enabled, you will get labels along with the transactions automatically. You can also find them in the Insight portal in the section with imports.
To get the scoring you’ll need to perform a separate request after all desired user data has been imported - you can find out more in our documentation.
The scoring algorithm has been designed to help lenders make the decision about their clients given the labels and other transactional data. Despite the fact that the transactions contain all the information that is needed for prediction, the relationship between them and the client’s default may not be easy to uncover for a human. The scoring algorithm was trained on massive amounts of data and so is able to spot these tiny dependencies. Moreover, it is able to explain its result by indicating the labels that were important in its decision process.
