BNY Mellon deploys new AI, cloud tools to assist back-office analytics needs

Data science can drive automation, assist with labor-intensive tasks, uncover insights so firms can make better decisions, detect useful anomalies for risk management teams—the list goes on. 

“We believe we can make meaningful contributions by applying advanced analytics across our businesses,” says Mike Demissie, head of digital assets and advanced solutions at BNY Mellon. “With that in mind, we took it upon ourselves to apply AI within BNY Mellon in a number of areas, at scale, and solve some of these really large problems.”

One example of a solution the custodian bank recently worked on uses AI to help predict settlement failures. Demissie says the bank handles more than 100 million trades a year for its clients. A small fraction fail to settle on time, an ongoing industry problem. 

In financial services and other industries, there’s a litany of pilots and proofs of concepts that are just small scale and really don’t go anywhere

Mike Demissie

“Our thesis was we can use AI to look at the historical patterns and give clients real-time probability of how likely a trade is to settle. So if a trade has a high likelihood of failing, they can take action on it, and not wait until after the fact,” he says. 

Another example is the use of machine learning to translate a client’s payment instructions. BNY Mellon receives payment instructions from more than 3,000 clients and the instructions come in a variety of formats.

Rather than have a team of people transcribing client instructions and processing payments, the bank uses machine learning to determine if it is a payment instruction, extract the relevant information, and populate it into the payment system and then present it to the operations team alongside the source information, so they can validate it. That has reduced processing time by 80%, adds Demissie. 

“In financial services and other industries, there’s a litany of pilots and proofs of concepts that are just small scale and really don’t go anywhere. We were very intentional about avoiding that and recognizing the AI capability is mature enough to tackle these really complex industry problems, and focusing on applying and implementing solutions at scale,” Demissie says.

But all this is only possible with the right data infrastructure in place.

“If you think of AI solutions as applications, for them to work effectively, you need that infrastructure that brings the data together,”  Demissie says. “That means quality checking, tagging, and everything else that is needed. That way, you spend more time on the data science work as opposed to wrangling the data. Think of it as a core enabler.” 

With that in mind, BNY Mellon’s team spent the last four to five years building Data Vault, its cloud-native data platform. Built on Microsoft Azure and powered by Snowflake, it is designed to help users rapidly collect, connect, store and distribute data across an organization. 

Mark McKeon, CEO for data and platform solutions at BNY Mellon, tells WatersTechnology that the goal is to provide a platform to clients with the right tools and capabilities to either perform their own data science, as well as connect to other platforms they might use to assist in decision-making activities.

Not as easy as it seems

The right data infrastructure is something that abstracts, aggregates, and normalizes various disparate structured and unstructured data sources into information that is easily “queryable,” according to Steven McCaffrey, managing partner at IT consultancy Proviniti.

This allows data scientists to spend their time and effort looking for unique business value in the data, rather than marshaling and massaging it to prepare it for analysis.  

“As the treasurer for one of the world’s largest banks said to me years ago, ‘We have the most comprehensive and granular financial dataset in the world that would allow us to forecast the US and world economies better than anyone … if only we could get to this data,’” he says.   

In today’s world, projects that go more than 12 to 18 months without showing significant value can be extremely risky

Steven McCaffrey

But setting up such an infrastructure that is “data-science friendly” is challenging for asset managers and banks. Some of those challenges include accessing and normalizing disparate data platforms and structures from different “feeder” systems, as well as properly and legally handling personally identifiable information data abstraction. 

They may also be dealing with dated source systems that have been modified and not documented over the years; concerns by the business units that data science activities will somehow affect their production operations; bandwidth and encryption requirements and certifications needed to move to a cloud or cloud/on-premises hybrid infrastructure; and decisions about the scope of a federated model, where data is abstracted to “appear” as a single source, or is copied to the data science platform or a hybrid of both, McCaffrey adds. 

These challenges are apart from any politics between business units in how these platforms are set up. And once agreed upon, the platforms can take time to get to day-one usability, particularly at a large bank or asset management firm. 

“In today’s world, projects that go more than 12 to 18 months without showing significant value can be extremely risky as management and personnel changes—and different visions over time—can lead to delays in delivering value. In our experience in the field, we have seen that complex infrastructure projects supporting data science and data mining operations are the poster child for this phenomenon,” he says. 

This is what BNY Mellon hopes to help clients solve, McKeon says. 

“As we think about data science and analysis, it all comes down to having the right datasets to be able to draw insights from so we really focus around that infrastructure and that connectivity—whether it’s with BNY Mellon data, vendor data or third-party data. Then we bring all of that together, start to draw some insights or create insights from that, and then connect that data back into the platforms our clients are using, particularly in the front-office space.” 

Leave it to us

As part of its goal to better facilitate data science and analysis work, part of BNY Mellon’s proposition is to do some of the heavy lifting on behalf of its clients. 

For example, in September 2022, BNY Mellon and Aviva Investors announced that BNY Mellon will provide a fully integrated operating model for certain front-office support services, as well as middle- and back-office activities. 

The solution will see BNY Mellon provide Aviva with traditional asset services, including custody, fund administration, and depository capabilities, as well as front-office support such as mandate monitoring and performance management. 

Central to this is Data Vault, which will allow Aviva to gain insights and accelerate access to analytics for teams across the investment lifecycle. Outsourcing those services to BNY Mellon will allow Aviva to enhance its client proposition and improve operational efficiency. 

As firms look toward interoperability and transformation, they’re taking a more horizontal approach, looking at potential issues across the front, middle, and back office, instead of by specific business function. McKeon says the only way to do that is by moving toward a data-centric operating model, as a client’s typical infrastructure is often fragmented with different data solutions for different functions.

“Having a horizontal, data-centric operating model in place not only allows you to deliver scale and efficiency for all your core activities—like your back-office, middle-office, front-office activities—to various tools, but more importantly, it brings a client’s data and third-party data into one place,” he says.

Having data in one place and because Data Vault is built on Snowflake, it allows connectivity to others in the industry. 

We’re now focused on sharing data and in real time

Mark McKeon

“The cloud-based Data Vault solution has all of the underlying raw data. Then our Data Fabric [platform] is where we actually connect to these tools dynamically so if a client wants to understand the risk on their portfolio relative to an attribution breakdown, they can do that by leveraging our Data Fabric capability,” says McKeon.

He adds that, like other providers, BNY Mellon is speaking with FactSet, and has agreements in place with other service providers and data vendors that use Snowflake. 

“So rather than us thinking about integrating data, we’re now focused on sharing data and in real time; it’s not taking a snapshot at a certain point and then moving it over to a different provider. Everyone’s working off this common data set,” he says. 

McKeon adds that over the last three to four years, service providers like BNY Mellon, and others like BlackRock’s Aladdin, other trading platform vendors, and data providers, are finding it more challenging to work independently of each other. 

“There’s been this kind of acknowledgment or recognition that everybody can do better if there is this kind of interoperability and integration and partnerships,” he says.

Beyond integration

A differentiator for BNY Mellon is that it not only integrates with other providers; it can also operate those services for its clients. 

“If we take Aladdin, for example, we operate Aladdin on behalf of our clients—we don’t just integrate with them, but we actually will operate Aladdin whether it’s for middle-office or for some of the front-office services that we’re taking on for Aviva,” McKeon says. 

“We will do all of that integration and take on all of that operational activity. But you can still have access to these tools to look at the output, we’re just helping you get there. Rather than clients having large teams on their side just running the systems and operating them, we will take that on and then provide insight into the output,” McKeon says. 

As part of the Aviva partnership, BNY Mellon is taking on 150 people from Aviva that have front-office investment management experience. 

“As we take on some of these functions for clients, it may end up being that we take on a team from a client. These are strategic capabilities, strategic relationships that we will provide to clients,” says McKeon. 

Proviniti’s McCaffrey says as more employees either retire or look for more “satisfying” work, it is difficult to find the skills necessary for some non-core functions that are not attractive to work on, especially in a competitive labor market. 

This will lead banks to separate their non-core functions and turn them over to a third party so they can focus their efforts on more profitable functions. 

He sees a trend for banks and asset managers to provide more value-added capabilities to their clients in the form of advanced analytics on their customer’s data and/or provide a platform for their customers to apply data science and analytics on their own data. 

“This is a win–win for banks like BNY Mellon because they can monetize their platforms and at least cover their operating costs and provide new products to differentiate themselves and give their clients the ability to differentiate themselves,” he says. “This function will apply to almost all commoditized data elements that just about everyone has, including customer data that is generally available for those who seek to utilize it.”