How AI could shape the future of investment banking

By Alex Hamilton | 24 June 2019

Artificial intelligence (AI)'s potential is difficult to pin down. In a competitive market like investment banking every advantage is needed to come out on top in terms of profitability, lead generation and customer experience.

Yet according to a June 2018 survey from Greenwich Associates, just 17% of investment professionals are currently using artificial intelligence as part of their processes. 40% of respondents from the same report indicated they would be increasing the budgets for AI deployment in the future. A February study from efinancialcareers found that major US banks were advertising for 93 positions related to AI.

Automation isn’t something new to investment banking. Junior bankers working in Excel can create macros to give them a hand in financial modelling, but the process can still be time-consuming and labor-intensive, and while auto-calculating formulae and crunching numbers frees up small amounts of time, it still means many hours staring at a screen.

BNY Mellon announced in October 2017 that it was using more than 250 robotic process automation (RPA) systems across its portfolio to automate tasks and reduce the amount of manual work put in by employees. 220 of the 250 have been developed by the bank’s partner Blue Prism. The bank claims to have achieved an 88% improvement in processing time, 66% in trade entry time and quarter of a second reconciliation.

Artificial intelligence’s most promising benefit might come from the freeing up of analyst time, but there are other crucial advantages the new technology could give an investment bank and those in both junior and senior roles.

AI for market data collection

A simple implementation of an AI system could remove the threat of information overload for an investment banker. Associates are expected to know the news as it happens, and in some cases before it happens. An artificial intelligence platform could help analysts by systematically collecting, analyzing and classifying news, as well as sentiment within the market.

According to an August 2018 report from Qualtrics and Amazon, 97% of researchers believe that AI will eventually make the role of market research redundant, alongside statisticians (95%) and analysts (94%). While AI systems have traditionally needed large pools of data with which to feed themselves, advances in the technology mean that some systems can now auto-feed themselves with market data collected from numerous sources.

The latest generation of analytic bots can pull information from interactions with humans and proactively request new data without having to be prompted by an operator. Examples can be seen in AI-driven online surveys, which react to sentiment and context in the answers to determine what to respond and what further questions to ask. That data can then be plugged into an analytical AI system for analysis.

AI systems have become sufficiently advanced that they are able to autonomously verify and cross-reference data inputs to maintain high data quality and ensure that any anomalies are removed. This prevents the collection of duplicate data across multiple pools and ensures that the information flow maintains accuracy and validity.

Market data firm Refinitiv (then still under the Thomson Reuters title) and AI firm Squirro partnered in September 2018. Both offer an integration with proprietary customer data, market data and augmented intelligence. The two firms claim that the system analyzes data from across all three pools and generates real-time insights.

AI for predictive analytics

Predictive analytics is the process by which computer models are used to predict future events. Using historical data, an artificial intelligence algorithm attempts to predict future scenarios under certain conditions. It is one of the more target-specific branches of data science.

French bank BNP Paribas utilizes an AI-powered tool it launched in 2017 called Smart Chaser to streamline the trade matching process through predictive analytics. Smart Chaser uses data records of trades made through BNP Securities Services and identifies patterns that led to a failure. The system is then used to predict the likelihood that a particular trade may need manual intervention in future.

According to Thomas Durif, the bank’s global head of middle office products, up to 30% of trades processed at the time required manual intervention to complete. The Smart Chaser tool was said by BNP Paribas to have a 98% prediction accuracy on release.

ING launched its Katana tool in December 2017, designed to help bond traders make better pricing decisions. The bank believes that the system can learn from historical and real-time trading data to create a statistical forecast, something traders can use on top of their natural skills.

Annerise Vreugdenhil, head of innovation for wholesale banking at ING, says in a bobsguide video that the platform allows the trader to pick within the system’s predicted bandwidth or go outside of it. “Helping them really adds something thing to the experience and creates a better proposal. The great thing is it seems to work, with the hit ratio of the traders going up by 25%.”

AI for trade processing

Developers of AI solutions like Man AHL – part of alternative investment firm Man Group – believe that the technology can identify systemic investment strategies and then automatically execute trades, even across multiple markets. When a trader places a buy or sell order, routing algorithms are used to find a match from stock brokers, exchanges and trading systems that can fulfil the order.

Man AHL claims that the routing of trades can be done via internal execution algorithms, external dealer algorithms, or the bank’s trading desk. Whenever a trade is executed, the AI system will break it into smaller components and then move it in the most optimized manner possible. Man AHL has also collaborated with Oxford University in the creation of the Oxford-Man Institute (OMI), providing funding to the tune of £13.75m.

When it comes to trades made through straight-through processing (STP), things usually involve little to no action from the middle or back office. When they fail, they are moved from the STP system into an unconfirmed queue that must be settled manually. Middle office staff must then perform exception processing by identifying the reason for rejection.

Resolving these failed trades can sometimes mean calling counterparties and confirming details over the phone. On other occasions it can require communication between the back office and trading desk on which discrepancies must be resolved.

As a result, exception processing of rejected trades becomes a labor-intensive process. An AI system using a pool of historical data would be able to identify why trades have failed in the past and apply that logic to new instances, while a supervised learning algorithm would allow human operators to adapt the conclusions it reaches.

AI for regulatory compliance

Risk and compliance departments suffer from large data loads and increasing regulatory requirements. Attempting to solve these issues with manual processes leads to repetitive and time-consuming processes. While AI in the form of algorithms has been around for nearly 20 years, what separates it in the compliance field is its ability to create an iterative learning process. An AI system can adapt inputs and perform tasks automatically.

Almost every reporting process which financial institutions engage can be simplified with the application of natural language processing (NLP) and robotic process automation (RPA). Having to implement and integrate the content from thousands of regulatory publications every month creates a multitude of complex interactions between different sectors of the business.

Beyond the simplification of complex processes, AI can aid in entity resolution and network analytics. If an institution deals in data, it will very likely have entity resolution challenges. Finding duplicates, correlation, and disambiguation within data, as well as understanding relationships between sets, is crucial in proper reporting.

According to a October 2018 report from IBM and Chartis Research, financial services firms are implementing AI-powered regulatory systems to meet compliance requirements (64%), as well as costs reduction (56%) and greater accuracy of processes and analysis (44%). Around 70% of all respondents to the survey were using AI in risk and compliance in some shape or form. Despite this, a lack of skills and data was cited as the largest challenge to greater AI adoption in the industry.

From the same study, professionals were most concerned about regulatory sign-off for their AI projects, but the 36% that cited it were behind the group that cited data availability (58%) and insufficiently knowledgeable staff (49%) as concerns.

Recent announcements, like the December 2018 statement from the Financial Crimes Enforcement Network (FinCEN), the Office of the Comptroller of the Currency (OCC), National Credit Union Administration, and the Board of Governors of the Federal Reserve System in the US encouraging the use and experimentation of AI (albeit in this case to tackle money laundering) have also emboldened financial services to deploying automated services in greater numbers.

Become a bobsguide member to access the following

1. Unrestricted access to bobsguide
2. Send a proposal request
3. Insights delivered daily to your inbox
4. Career development