The accuracy uplift of an specialised transaction enrichment provider

Transaction enrichment uplift

Two questions that come up most often when we talk with prospective customers are focused around;

1. Are you able to enrich transactions from any bank?
2. Our open banking provider offers enrichment, why do we need a separate solution?

Both valid questions that we wanted to address and our answer to the first question parlays into the second.

Any data source, any language

⁠At Ntropy, we train language models to understand financial data at scale and our use of language models means that we are agnostic to where the transaction data is from. 

Our models have been trained on a diverse dataset of financial data from different banks and different countries, and have been built to parse and enrich bank data in any format and any language. ⁠

Much like how open banking providers and data aggregators connect to different banks and standardise the output, we do the same for bank data.

It doesn’t matter which bank it is from, which open banking provider or data aggregator our customers use or if it is in a spreadsheet, PDF or from your Snowflake/AWS/Databricks (name any database solution) data lake or even what language it is.

If it is bank data, we can enrich it with a high degree of accuracy.

Why do we need a specialised transaction enrichment solution?

Open banking providers are great at their core function, providing connectivity to a wide range of financial institutions so consumers and businesses can consent to sharing their bank account data with third parties. 

Their core focus and competency is building bank integrations and providing access to the data to their customers. Our customers would struggle to access and send us bank data if they did not exist.

Most open banking and data aggregation also offer some transaction enrichment as a secondary offering but from what we have heard from our customers, it often does not meet their needs for a number of reasons we’ll outline below.

Accuracy - 95%+ or bust

Poor accuracy of both merchant enrichment and categorization, the two core components of transaction enrichment, is the biggest reason we see with customers who come to Ntropy after using or testing open banking enrichment solutions from their existing open banking provider.

It is just not accurate enough for their use-case.

Focus on consumer data enrichment

Firstly, this lack of accuracy is most apparent with business data.  Open banking providers mostly provide access to consumer data as their customers are mostly consumer facing apps and services.

They have less experience with business data which is inherently different and more complicated to enrich. Businesses are a lot more heterogeneous than consumers and the context and understanding of the difference requires experience which they don't have. 

No business specific category hierarchy

Open banking providers typically try to fit their existing consumer focused categories to business data but this simply doesn’t work and makes their categorization very inaccurate.

The context of transaction data is highly dependent on use case and customer and so a consumer category taxonomy like "deposit" are not granular enough for businesses with different sources and types of revenue.

We have two different hierarchies, one for consumer data and one for business data to provide more granular and specific context to bank transactions.

You can easily see our business category hierarchy and consumer category hierarchy which each have over 100 categories tailored to each data type. 

Building bank connectivity

Additionally, there are around 10,000 financial institutions in the US and over 5,000 in Europe. Open banking providers are in a highly competitive industry and their core focus is expanding and maintaining their bank connectivity to be the broadest provider to win customers.

Their focus is not in enriching bank data with a high degree of accuracy and they devote less resources to their enrichment capabilities.

Accuracy uplift from switching to a specialised solution

Customers and prospects that we speak to that have benchmarked the accuracy of our solution with that of their data aggregator or other solutions on the market, found that we are the most accurate solution they tested.

Many customers in our core markets (US, UK, EU) experience accuracy of 95%+ compared to the ~60% accuracy of their existing open banking provider

This is one of the reasons why Yapily partnered with us last year, to provide their customers with the most accurate enriched bank data.

The Ntropy Method

Rather than using a rule-based approach that many other enrichment providers use, we use a combination of approaches to ensure high accuracy and high coverage.

We combine the use of language models, merchant databases, search engines, rules and a human-in-the-loop feedback loop to provide the best results for our customers. 

Our CTO wrote a comprehensive blog post on the Ntropy method and compared our results to enrichment from ChatGPT. (Spolier alert: we had similar accuracy but 100x cheaper and 200x quicker).

While mediocre accuracy is workable for some use cases where enriched data is a “nice to have”, when making big decisions or building revenue generating products, it simply is not good enough.

If you are not getting over 90% accuracy with your transaction enrichment, come talk to us!

The need for speed

Our customers have reported that they can spend seconds waiting to get enriched and categorized transactions from their open banking provider. There are some use cases in which latency of this magnitude is not a problem but for many use-cases it is not feasible when other workflows and processes run in near real-time.

For example if you are trying to use enriched and categorised bank data in the payment authorisation flow or you want to provide borrowers with an instant lending decision, adding seconds to that process will severely impact the user experience.

At Ntropy our standard latency to enrich and categorise bank data is 250 milliseconds. We can also optimize for speed to reduce this latency even further if required for your use-case.

Conclusion

We can enrich any bank data, with a higher level of accuracy and specificity and at a quicker speed than open banking providers, whose core focus is not enrichment, whereas it is our raison d’etre.