How Bud updates its categorization taxonomy

One of the most important aspects of financial transaction enrichment is categorization. At Bud, we are proud to build and operate what we believe to be the most advanced and accurate categorization engine available. It utilises a hierarchical, three-level categorization taxonomy containing more than 300 elements.

The engine itself is constantly evolving, adapting to new tokens found in the data and to the shifting financial habits of end customers. We are also continuously monitoring its performance, working to find the right balance between coverage and accuracy. Sometimes the changes are small, resulting in subtle improvements in accuracy and shifts in confidence distributions. In other cases, the impact is more significant.

The most impactful changes to the categorization engine are related to the taxonomy itself — the list of categories that transactions are matched against. These changes are typically driven by larger trends observed in the data. Broadly, there are two primary types of changes.

The most obvious are reactions to major shifts in financial behaviour. Adding inflows and outflows related to cryptocurrencies is one example. Another is the emergence of “buy now, pay later” loans. These changes allow us to more accurately capture the true meaning of transactions, something that would not be possible without the flexibility to introduce new categories.

Another group of changes reflects the increasing amount of data available within the transaction information we receive from clients. The infrastructure landscape and payment rails are constantly evolving, often allowing us to extract additional insight from each transaction. However, sometimes the opposite occurs, with similar transactions becoming harder to distinguish due to shifts in processing.

There is more nuance to both groups, and a lot depends on transaction volumes and how those transactions are perceived, both by financial institutions and by end users. While we have full flexibility in how the taxonomy evolves, frequent changes are discouraged. New or disappearing categories can create confusion for users and make analytics more difficult, especially when there is no clear mapping between old and new categories.

A recent example relates to prediction markets. For some end users, these transactions align with the gambling category, while others prefer a clear distinction. From a financial institution’s perspective, customer sentiment matters, but these transactions may also signal changes in a customer’s risk profile, as well as an affinity for higher-risk, higher-reward investments.

Importantly, categorization is only one of the tools available to us. Our engine can also apply tags to transactions, allowing each statement entry to carry multiple layers of meaning depending on the use case. When combined with other insights, this approach maximizes our understanding of each transaction and each customer; ultimately contributing to better outcomes for all parties involved.