XiNG: Inside Citi’s all-encompassing risk platform

When Jon Lofthouse first joined Citigroup (then Salomon Brothers) in 1997 as a junior developer, most firms were focusing on Y2K preparations. The bank was just starting to dabble in complex financial instruments. Risk management was performed at the desk level, with different individuals and systems for each asset class.

Then came the 2008 financial crisis, and banks realized they needed to understand the risk they held across all their silos, both for internal management and to respond to regulators paying closer scrutiny to their businesses.

So, it was only natural when this came about that Lofthouse, who had already held almost every tech role in Citi’s Markets division (and since February of this year has served as chief information officer), would be picked to play a pivotal role in building the bank’s response.

That response began as a series of standards for its fixed-income trading group, which represented a significant portion of the bank’s business and the former Salomon Brothers business. The XiNG Platform, pronounced “zing,” is an acronym of Xi Next Generation, or XiP for short, where Xi was the internal nickname for an existing library of analytics. This set of standards would be the basis for a consistent suite of models for risk, stress testing, pricing, and quantitative analysis.

Though fixed income was the starting point, the platform was always intended to encompass as much of the bank’s business as possible and present it in a clear and uniform way.

“We didn’t want to build an app—we wanted to build a platform with a portfolio of applications,” Lofthouse tells WatersTechnology.

But that didn’t happen overnight. From the origination of the platform circa 2014, Lofthouse and his team—which included developers as well as professionals from the Markets division and the bank’s quant function—spent several years figuring out that initial standards component.

In fact, Lofthouse credits the involvement of representatives from the Markets group—and the physical proximity of the key parties that made it possible for them to be involved—as being pivotal to an effective development process.

“Andy Morton—who ran rates at the time and is now head of Markets—and I were in rooms next door to one another. The quant team was 10 feet away, and the tech team was in London or this office. We were working together physically for much of the time,” he says.

As the project moved forward across different asset classes, the key was to identify use cases that XiP could seize upon and use to expand its functionality, breaking down asset class-specific data siloes and normalizing that data and making it more widely available as an input to the models being built on the platform for other business lines. Anchoring the project to specific business problems allowed the firm to justify the continued investment as it added support for more asset classes and functions incrementally.

“We learned lessons as we went along,” Lofthouse says. “For example, the API wasn’t as simple as it needed to be. And we learned a lot about data and keeping market data and developers together. And as we evolved on our journey, we started solving bigger and more complex problems.”

We try to make these apps as simple as possible. And making them appear simple is actually more complex to do

Jon Lofthouse, Citi

As that evolution took place, it became clear that a platform this broad simply couldn’t fall into the trap of becoming a monolithic system, so Citi took the approach of building an API services platform that could make the same data services available to different areas for different uses, such as for pricing by the front office, or for stress testing.

About five or six years ago, the bank started building apps on top of the platform, and over time, developed a “self-service” approach where the consistency of the platform could allow users to craft their own models with certainty that the look and feel would be the same as other apps on XiP. For example, if a quant wanted to build a new model to price callable bonds, they could do that without needing to tie up developers.

“Because we spent time getting the standards and API parts right, we can now build apps very fast,” Lofthouse says. “We try to make these apps as simple as possible. And making them appear simple is actually more complex to do. We wanted to make them only require a small amount of code, and to extract complexity so that it’s hidden in services that run behind the scenes.”

Indeed, behind the scenes, thousands of lines of code are required to make the interface appear easy to users. But that apparent ease came about only as a result of vision and persistence, and by anchoring XiP to specific business needs, rather than an aspirational “build it and they will come” approach.

“The first lesson is to get the foundations right and to stay the distance,” he says. “There’s a big difference between engineering and PowerPoint.”

Every trade

It’s a similar approach to the “reuse and recycle” model adopted by Bank of America in our second Voice of the CTO article. Though executed differently, the end results are a better consolidated view of technology investment across the organization, fewer disparate data siloes, and more sharing of data and services, reducing the amount of new build or tech spend. There, BofA had developed a portfolio and cash trading front-end that it was able to tweak and roll out to its repo trading business, and was able to repurpose an equities internal crossing engine to grow its rates internalization business.

But starting from scratch, Citi began with its largest footprint and expanded from there over time.

“We started with fixed income because our Markets business is biased toward fixed income and because fixed income requires more compute. Then we applied it to credit, commodities, and equities—all our trading businesses and our quantitative and risk functions, and for equity mark-to-market accounting, and derivatives,” Lofthouse says, adding that Citi is currently exploring additional use cases throughout the bank. “Probably a year ago, we got to the point where every trade we do in the firm goes through XiNG.”

That was the point at which Citi CEO Jane Fraser gave the system a shout-out on the bank’s Q3 2024 earnings call, where she described the bank’s simplification transformation and infrastructure modernization—including consolidating technology and retiring some 1,250 platforms since 2022—and its efforts to demonstrate greater resiliency to regulators.

Consolidation and retiring applications isn’t something unique to Citi: Other institutions are also focusing on simplification and streamlining. These include Deutsche Bank, profiled in our third Voice of the CTO article, which has been navigating its own data transformation as part of its transition to Google Cloud and to “redefine how it develops and offers its financial services” and has retired 41 applications in its corporate bank in the past year alone. 

These efforts to simplify firms’ architectures are in direct response to the application overload that has sprung up as the desire to embrace new technologies has arguably outpaced firms’ ability to manage the underlying data. The more silos and siloed data you have, the more applications are needed to deliver that data to the users or customers who need it. 

Compare this to Citi’s approach: Leverage the same underlying platform and allow, encourage and mandate sharing of underlying data. Fewer silos, fewer disparate applications to support, more value.

Below the radar but in the cloud

Yet even after Fraser’s comment, no one outside of the bank seemed to pick up on XiP. Inside the bank, however, the results were already clear: The platform can scale upward to potentially process billions of calculations per day and has—though the bank declined to share specific timeframes—delivered a 10x improvement in the time taken to perform them.

“Some of the calculations we wanted to do take hours, so we had to thread them across different Kubernetes pods,” Lofthouse says.

The bank used Kubernetes as a way to make the platform operate across multiple clouds because any platform handling so many trades, calculations, and models requires a lot of supporting data and a corresponding amount of compute power.

That said, the core of the platform and its underlying data resides in-house—“I don’t have anything in a third-party center where, if it went away, it would be a problem for me,” Lofthouse says—but was expanded to allow it to burst to cloud when it needed access to more compute power, and now parts of XiP run across multiple clouds as required, depending on a number of factors.

“Calculation workloads can be executed on-premises, and across multiple public cloud vendors,” Lofthouse says, though he declines to name specific clouds that XiP uses. “Placement of these workloads is dynamically based on factors such as resource availability, latency and cost. For example, if one cloud region is experiencing issues, workloads will seamlessly run in another region to ensure continuity of service.”

Indeed, when a bank as big as Citi runs a platform as big as XiP, it may spend millions of dollars per year on compute services, so being able to switch between providers can yield significant savings.

When asked whether Citi could offer the platform to clients or even spin it off, Lofthouse seems to suggest that the bank sees greater value in keeping XiP in-house.

“Very few, if any, financial institutions running at Citi’s scale, have a modern platform that can execute on the entire universe of Markets,” Lofthouse says. “The combination of Citi’s diverse product portfolio, global reach, and the modern efficient platform in XiP represents a distinct competitive advantage among our competitors.”