Transaction labelling

What is transaction labeling?

The labels are like hashtags used widely on the Internet. Each transaction can receive multiple “hashtags” that are supposed to characterize the financial movements on the account.

We label transactions from bank accounts, so you can easily define their structure and detect trends in your clients' financial behavior.

How can you benefit from this service?

The metadata contained in the Kontomatik labels allow to determine, among others, customer lifestyle, healthcare expenses, travel, kids, nightlife and more.

You will learn what payment methods your clients use, how much their salary is, and what obligations they have. Deeper analysis allows:
detailed understanding of specific areas like e.g.: insurance, scanning behaviour patterns and trend analytics. This helps with user profiling and precise targeting of your clients.

Multidimensional analysis over
categorization

A typical categorization creates some friction when analyzing data - you can only have one category assigned, maybe decide which is more important or whether any is actually appropriate.

That’s why we went for a more complex solution. Each operation can receive multiple labels, making both transaction classification and information aggregation more accurate. As a result, customer profiles and analysis of their financial behavior become even more precise.

Some transactions will not receive a label which in our experience is far better than a forced assignment which would skew the results and lead you to analyze behaviors that simply aren’t there.

How are the labels assigned?

The solution is based on a machine learning system. Multiple people we work with manually assign labels to hundreds of transaction examples, we then create algorithms that are being taught the patterns and as a result we obtain a statistical model that for a single transaction reviews all fitting labels.

We’re taking into account data such as the title, amount of the operation or its regularity. Every single transaction is reviewed for all labels separately.

Each label has different characteristics that determine numerical values for specific words, amounts alongside some algorithmic rules that also verify the dates. If the final crosscheck value for the transaction and the label is high enough, the label is assigned.

And even though the amount of factors taken into account is quite large, the labeling process takes just seconds even for an enormous chunk of transactions!