Practicalities of Implementing GenAI in Capital Markets

Following the opening keynote of A-Team Group’s AI in Capital Markets Summit (AICMS), a panel of expert speakers focused on the practicalities of implementing GenAI. The panel agreed that industry hype is waning and there is enthusiasm for GenAI with firms beginning to develop use cases, although one speaker noted: “People understand the risks and costs involved, but they were initially underestimated, I would say dramatically in some cases.”

Considering the use of GenAI, an early audience poll question asked to what extent organisations are committed to GenAI applications. Some 46% said they are testing GenAI apps, 24% are using one or two apps, and 20% are using a number of apps. Nine percent are researching GenAI and 2% say there is nothing in the technology for them.

Value of GenAI applications

A second poll questioned which GenAI applications would be of most value to a delegate’s organisation. In this case, 53% of respondents cited predictive analytics, 39% risk assessment, 39% KYC automation, 28% fraud detection and 19% portfolio management.

The panel shared their own use cases, with one member experimenting with GenAI to produce programming code and creating an internal chat box for data migration, as well as scanning data to surface information that can be categorised, sorted, filtered and summarised to create ‘kind of conversational extracts that can be used.’

All agreed that GenAI produces some low hanging fruit, particularly in operational activities such as KYC automation, but that the technology is too young for many applications, leading firms to build capability internally before unleashing GenAI apps for customers as there is still work to do around issues such as risk integration and ensuring copyright and data protection are not compromised. One speaker said: “There is a lot of experimentation and some research to do before we’re confident that we can use this at scale.” Another added: “There are just not enough skilled people to allow us to push hard, even if we wanted to. There’s a real pinch point in terms of skills here.”

Risks of adopting GenAI

Turning to risk, a third audience poll asked the audience what it considered to be the biggest risk around adopting GenAI. Here data quality was a clear leader, followed by lack of explainability, hallucinations, data privacy and potential misuse. Considering these results, a speaker commented: “We’ve already got existing policies and governance frameworks to manage traditional AI. We should be using those to better effect, perhaps in response to people identifying data quality as one of the key risks.”

The benefits of AI and GenAI include personalisation that can deliver better products to consumers and improve the way in which they interact with technology. From a regulatory perspective, the technologies are focused on reducing financial crime and money laundering, and resulting enforcements against fraudulent activity.

On the downside, the challenges that come with AI technologies are many and include ethical risk and bias, which needs to be addressed and mitigated. One speaker explained: “We have a data science lifecycle. At the beginning of this we have a piece around the ethical risk of problem conception. Throughout the lifecycle stages our data scientists, machine learning engineers and future engineers have access to python libraries so that when they test models, things like bias and fairness are surfaced. We can then see and remediate any issues during the development phase so that by the time models come to validation and risk management we can demonstrate all the good stuff we’ve done.” Which leads us to the need, at least in the short term, for a human element for verification and quality assurance of GenAI models in their infancy.

Getting skills right

Skills were also discussed, with one panel member saying: “We are living in a constantly more complex world, no organisation can claim that all its workforce has the skill set necessary for AI and GenAI, but ultimately I am hopeful that we are going to create more jobs than we are going to destroy, although the shift is not going to be easy.” Another said: “In compliance, we will be able to move people away from being data and document gatherers and assessors of data in a manual way to understand risk models, have a better capability and play a more interesting part.”

Taking a step back and a final look at the potential of GenAI, a speaker concluded: “Figuring out how to make safe products that we can offer to our customers is the only way we have a chance of reaching any sort of utopian conclusion. We must chart the right course for society and for people at work, because we’re all going to be affected by generative AI.”