How this digital banking provider uses LLM-s to build a single source of truth for retail customers

The Need

This digital banking platform is serving the world’s largest brands enabling them to turn on financial products and services. 

Often, both brands and community banks have the same problem. Integrating a new financial services suite is a huge undertaking, which they cannot afford to do internally. 

This is the requirement, however, if a non-financial brand wants to start making revenue from financial services. Not everything is fintech, but for many companies fin-services revenues are a natural user experience and a way to augment their existing products and boost unit economics. 

BaaS has a bad reputation these days, but platforms offering digital financial services as components have seen success serving both banks and non-banks for a while now.. The need is there and has always been.

This digital banking provider came to us with the problem of unifying siloed customer data, as a result of introducing a new banking suite to their customers, in order to build more personal and better financial products and services.  

Existing customer self-service capabilities were limited and the data was held in silos as a result of legacy technology. Customer’s onboarding information and transactional data was poor quality and fragmented  across different platforms.. The experience was functional, but made it harder to upsell new products or even have a single source of truth about customer retention, satisfaction and risk. . 

Unifying this data across different sources and normalizing it across multiple formats was key in order to unlock better NPS, improve retention and increase consumer LTV-s. 

Evaluation

They had previously evaluated other providers, yet most evaluations had failed to meet expectations, presenting at best results of varying quality. 

The customer ran 1000s of transactions through our API and we had our team of human evaluators sample and label 500 of them to gauge the quality of the results.  We looked at the accuracy numbers of four fields that were critical for them : labels, merchant names, merchant websites, intermediaries. The results were significantly better than anything they had evaluated previously: 

labels : 94.8%

merchant : 95.6%

website : 93.2%

Intermediaries : 95.8%

This level of accuracy meant that they could use this data to create clean customer audience personas and ship better models across marketing (upsell and cross-sell) and risk (underwriting).  

This data acts as a source of truth for any financial product or activity further up the stack.  

When we started Ntropy, our insight was simple. Any financial process, from the beginning of your journey (onboarding, KYC and KYB) to the end of it (accounting and reconciliation) relies on this information. 

ROI

One of the biggest use cases in banking with AI has always been around personalization and better recommender systems. This is a conversation around personalizing money that has been relevant for at least the last 15 years. 

In a competitive setting with high CAC like banking, user satisfaction is a huge lever, increasing LTV is another one. Until now, despite the aspiration, the data quality has never been there in order to be able to do anything even comparable to other industries, such as entertainment, social media and e-commerce. 

These generate billions of dollars thanks to their recommendation engines. Your Netflix and Spotify know you way better than your bank, despite your purchase history being more precise and telling, than any of the data they collect. 

With a centralized source of truth and high quality of the data, we are paving the way to finally unlock this future where your money experiences are true to what you need.