Technology provider Broadridge Financial Solutions is testing OpsGPT, an application powered by generative artificial intelligence, with clients in operations after launching BondGPT for the front office in 2023.
BondGPT was Broadridge’s first application powered by GPT, a language prediction model.
Vijay Mayadas, president of capital markets at Broadridge Financial Solutions told Markets Media: “I have been in this business for many years and I have never seen innovations come to market this quickly with this level of transformative potential. It is an incredibly exciting time to be in technology and in capital markets.”
Broadridge has always believed that AI will transform operations but the recent emergence of generative AI tools made that potential come into focus. BondGPT is powered by OpenAI GPT-4, and offers a large language model chat function that allows users to ask questions and identify specific corporate bonds on LTX, Broadridge’s electronic bond trading platform, in order to meet traders’ real-time liquidity needs.
In addition to the emergence of generative AI tools, another factor for accelerating automation in the back office is the upcoming shortening of the settlement cycle in the US and Canada. In May this year the settlement period for equities and fixed-income instruments will be cut from two days after execution, T+2, to T+1.
“T+1 was definitely a catalyst as we realised that clients are going to have a lot more operational complexity,” added Mayadas. “We have many other use cases for generative AI across our capital markets business, but this is one area where we felt there would be a lot of interest.”
In a white paper, Efficiency Unleashed: Capital Markets Operations Reimagined with Generative AI, Broadridge estimated that within a subset of areas impacted by T+1, the technology could improve operational efficiency by up to 50% while mitigating risk, rising costs, and enhancing overall governance.
Mayadas said: “I think 50% is just the beginning.”
Using the right data
One of the risks of AI models is that they can generate ‘hallucinations,” which are false or misleading results, and therefore the integrity of the underlying data being used is critical. To overcome hallucinations in BondGPT Broadridge created an adversarial, large language model which tested the output and compared it to a set of rules that the firm has been developing for many years.
Mayadas explained that Broadridge trained OpsGPT on data from Broadridge’s global multi-asset post trade systems which clear and settle $10 trillion in trades every day. The underlying models are derived from Broadridge’s investment in creating BRX, a data ontology which harmonises data across asset classes so that it is AI friendly.
BRX structures and distills trade data so that the most important trade attributes are captured in a platform-agnostic modelling language. This has resulted in a repository of clean, normalized data that can be used for both trading and post-trade functions.
“I think BRX is a kind of special sauce in terms of our ability to launch AI products very quickly,” added Mayadas.
He continued that BRX was developed following the acquisition of Message Automation, a UK post-trade fintech, in 2017 to broaden Broadridge’s post-trade and data analytics capabilities. Message Automation built their business by simplifying how multi-asset class transactions were represented to automate and scale regulatory reporting.
As a result of this data ontology and using AI to help develop the product, BondGPT was developed from concept to introduction in about eight weeks, which Mayadas said was a real record for Broadridge.
BroadRidge also has a managed services business, who run outsourced operations for clients who were critical in developing OpsGPT. The application is being tested by clients and their feedback will be incorporated before it is launched shortly.
OpsGPT is also built with regulatory requirements and industry-specific laws in mind and includes a compliance layer in the form of an AI agent that monitors the application’s output.
Use cases
Broadridge has an extensive roadmap of use cases for OpsGPT that the firm is working on with clients. For example, under T+1 it becomes important to reduce trade failures and resolve problems more quickly.
There are often many underlying causes for trade failures, which require pulling in data from disparate databases to resolve and identify the root cause. Mayadas explained that one of the powers of GPT-enabled products is that clients can use a very simple chat interface to write a complex, expansive query, and the model can access the relevant databases.
“OpsGPT can collapse all that complexity into a couple of questions from the user and give them actionable insights,” he added. “Something that used to take a couple of hours can now be done in 5 to 10 minutes and that is a real uptick in operational productivity.”
For example, clients can ask questions such as “Where can I find my failed trades?”, “How do I find my open positions?”, or “Do I need special permissions?”.
In addition, the model has a deeper reasoning capability so that becomes smarter as it is used more often. very high level metrics.
“We want to roll OpsGPT out and get real data,” said Mandayas. “In some specific cases, we can see a 10x increase in productivity.”
Growth of generative AI
Mayadas continued that Broadridge has had feedback from clients on adapting the model to inventory management, because trades usually fail because of a lack of inventory in the right location. Clients want to predict depot inventory ahead of time so they can predict what is going to happen in the settlement process.
There has been strong interest in OpsGPT across the spectrum of broker dealers from Tier One firms to smaller firms who may be more constrained in their ability to invest in operational resources to mitigate T+1, according to Mayadas.
Broadridge provides technology across the trade lifecycle and the firm sees opportunities to apply generative AI in the front office to better understand trading behaviours, trading dynamics and make trade analytics easier to access and Mayadas said the firm is quickly developing a number of use cases in these areas.
“There is demand for the equivalent of BondGPT in credit markets,” he added. “We are also seeing opportunities in equity markets, and in futures and options.”
The firm has partnerships with financial market infrastructures and Mandayas said they have been excited about the OpsGPT announcement and have had conversations around applying this technology more broadly across the industry.
Broadridge also expects the shift to T+1 settlement to pave the way for future advancements in market infrastructure.
“In 2024, we can expect further innovation and digitalization of market infrastructure, including same-day and block-level clearing, tokenized atomic settlement, and bilateral settlement, to name a few,” added Broadridge. “Technologies such as distributed ledger technology and AI will be at the forefront of executive minds as they look to further streamline their post-settlement operations.”
Bloomberg has announced the availability of AI-Powered Earnings Call Summaries on the terminal. The new generative AI tool was trained by Bloomberg Intelligence analysts to more accurately understand the nuances of financial language and anticipate what’s most important to investors.
Joyce Meng, partner at Fact Capital, said in a statement: “Most of my team’s job is reading and synthesizing trends across companies, so the quality and accuracy of the summarization tool gives us a big edge. It also distills the contentious points so we know where in the material to look for insights on the important debates.”
Risks
Consultancy Coalition Greenwich identified AI in general, and generative AI, as onto its top 10 market structure trends for 2024.
“In 2024, the ability to search through complex datasets and huge databases by simply telling the computer what you want in plain English (or French or German…) is the killer app,” said Coalition Greenwich. “Are you trying to find the most actively traded stocks in Switzerland with a PE ratio over 10, an issue size over $1 billion, a CEO over 40, and a debt load that has declined over the past decade? Plain language search is in—and it might just make market participants smarter and faster than they already are by putting data to use that last year remained buried in the cloud.”
Consultancy GreySpark Partners said in a blog that generative AI could prove to be a “game changer” by automating trade execution and algorithmic trading strategies to enhance risk management and provide real-time insights. Adaptive learning capabilities can also ensure that trading strategies evolve in line with market dynamics, leading to enhanced performance over time.
“However, it is important to acknowledge that although generative AI has immense transformative potential, the basket of new risks it presents could lead to foreseen and unforeseen challenges,” said GreySpark.
Issues include unavailability of accurate and high-quality training data, the unexplainable behaviour of complex AI models, biased results, potential systemic risks and ethical concerns so human expertise and oversight in running AI systems and interpreting the generated insights is likely to remain indispensable.
“The introduction of regulations such as the EU AI Act reflects the urgency of acting to counter the evolving risks as well as the speed of adoption of this technology,” said GreySpark. “This framework should allow banks to establish controls and implement governance measures to ensure both safety and effectiveness of the application of the technology in the financial services.”
