Close panel

Close panel

Close panel

Close panel

Data> Big Data 22 Nov 2019

Machine learning: how it’s used in banking

Automatic learning is a prerequisite for intelligent systems. It enables data-driven predictions and creates new business opportunities. Banking in general, and BBVA in particular, is already taking advantage of this technology in order to improve products and services for its customers.


Machine learning is a branch of artificial intelligence that uses data to enable machines to learn to perform tasks on their own. This technology is already live and used in automatic email reply predictions, virtual assistants, facial recognition systems, and self-driving cars. Breakthroughs in this technology are also making an impact in the banking sector.

BBVA has identified four main areas of automatic learning around which it organizes its data-driven product development and improvement strategy: automation, personalization, human-machine interactions, and security.

Tailor-made services

Before automatic learning reached the banking sector, (as is the case in other industries) systems executed rule-based business decisions, but only with a partial view of what was a very compartmentalized customer digital footprint. With this approach, it was normal to apply the same criteria across very broad customer segments. This has now changed.

"Today we have a unified, omni-faceted view of the customer footprint (financial status across multiple accounts, products that have been contracted, transactions, etc.), something that is now feasible thanks to new data warehousing and processing platforms. This improvement in data sources, combined with new analytical capabilities based on machine learning models that combine these data sources, learns from them, and is iteratively repeated, allows us to give each customer personalized service in accordance with her specific financial profile,” comments Juan José Divasson, head of service personalization initiatives at BBVA AI Factory.

He also adds that business decisions made at both the corporate level as well as those made by bank managers are increasingly based on information generated from advanced analytics. "Ultimately, we use customer data to add value to the services and products that we offer; concurrently, we deliver this data back to the customer in the form of relevant financial advice or opportunities,” he states. Consequently, business and consumer customers alike can can make better, informed decisions.

"The improvement in data sources, combined with new analytical capabilities based on machine learning models, allows us to give each customer personalized service"

Reducing uncertainty

Divasson points out that improving customer service and increasing productivity, which have the knock-on effect of reduced costs, are the primary areas where machine learning is being used to transform the banking business. Some of the activities that are being worked on in this area have to do with increasing product and service personalization, “adapted to each customer's lifestyle,” and the improvement of financial services, “with reduced uncertainty thanks to a more accurate risk assessment based on a wider customer digital footprint.”

“We also provide this data and artificial intelligence to the bank managers who look after our customers everyday. This frees them from dealing with the most basic activities so they can focus their time on those issues that add the most value to the customer,” adds Ángela Gonzalo, head of human-machine interaction at BBVA AI Factory.

In terms of productivity, activities most especially target the automation of simple, recurring processes, providing bank managers support in their day-to-day work, facilitating customer self-service when it comes to accessing data and performing transactions, and among the more critical aspects, providing greater security.

Fraud detection

A clear example of this last point is fraud detection, a problem that represents a serious challenge given that only one case out of tens of thousands is raised. “Techniques are used — the identification of anomalies or other types of pattern recognition models (unsupervised learning) — that work very well to address this kind of problem,” points out Javier López Peñalver, program manager at BBVA AI Factory. "It is also very common to use methods to tweak the data distribution beforehand (supervised learning), thus achieving a better balance between classes (fraud, not fraud)," he adds.

Last year, BBVA worked with a team of MIT researchers to develop a model based on machine learning algorithms that can reduce the number of false positives related to fraudulent credit card transactions by 54 percent. The new approach, known as deep feature synthesis (DFS), facilitated the extraction of more than 200 additional attributes from each transaction, which served to provide a more detailed description of the credit/debit card transaction behavior, thus improving the fraud detection engine results.