<img height="1" width="1" style="display:none;" alt="" src="https://dc.ads.linkedin.com/collect/?pid=575722&amp;fmt=gif">
Open Banking featured Tech deep dives

How it works: Our new approach to merchant ID

Our new Merchant ID service adds the ability to return the merchant name, logo and even added company information to a given transaction, to power insights that drive further actions.

Our new Merchant ID service adds the ability to return the merchant name, logo and even added company information to a given transaction, to power insights that drive further actions. Doing this without the need for any added information or system modifications, such as relying on merchant IDs, will make it easy for application builders to add enrichments that improve the end-user experience. These features can all add significantly to the look and functionality of banking applications, giving the user an added feel of confidence and driving engagement with the product.

How does it work?

Identifying names, locations and merchants from transaction descriptions falls under a branch of Natural Language Processing called Named Entity Recognition (NER). Given just a sentence of text or a transaction description, how do we know which characters are part of a merchant name, a location, a date, or just the name of a friend? And just as importantly, how can we solve this problem in a scalable way that will easily translate into new markets or additional merchants?

Bud entity recognitionNER seeks to solve this problem, leveraging deep neural networks and state-of-the-art networks called Transformers, specialised for reading text. These networks read a transaction description and understand the context of each element just like a person would. They are then able to classify the words in a transaction description, and tell us which parts are the merchant name, which are the names of people, and which are just dates or random numbers.

So problem solved, right? Well, not quite. As anyone who’s ever seen a bank statement knows, the merchant name often isn’t written cleanly or completely. We need to rely on more machine intelligence to learn which known entity a given word is referring to. This falls under a related but distinct research area called Named Entity Matching (NEM). Bud is working on various approaches to solving this problem, including deep learning models, teaching them to match identified merchant text against our known list even as that list grows with new merchants. 

All of this means that we will be able to go from just a transaction description, to a match with a specific merchant, even as the list of possibilities grows or we expand to cover new markets.

 

What are the benefits?

This feature means that application builders can retrieve merchant identification and enrichment directly from Bud, without having to analyse transaction descriptions themselves, or change their systems to record a merchant ID. 

It is another product from Bud that adds genuine transaction enrichment, allowing our clients to easily build more feature-rich products that will provide a better experience to their users, and consequently better represent their business.

How it works: Our new approach to merchant ID