Artificial Intelligence (AI) is not news. Everyone’s smartphone, tablet, laptop and desktop use AI to identify images and look for patterns and connections. Generative AI goes one step further: creating text, images and videos from AI data.
A 2022 Gartner report revealed 80% of CFOs wanted to invest more in AI technologies by 2025. Steve Rodgers, Chief Technology Officer in Analytics at Beeks Group, comments: “There is no doubt about the tremendous transformative power of GenAI to accelerate many aspects of financial firms’ business. As MSPs in the financial space we are both excited about the possibilities and wary about the challenges that GenAI can bring.”
These include:
Cost
Rodgers says: “GenAI advances in the finance sector are hampered by the economics of scarcity in skilled resource.” Add into the mix the cost of GPU hardware and subsequent model training and updating, the barriers to entry are not insignificant – even more so when attempting to adopt AI solutions in colocations without the huge power of Cloud on tap.
Regulation
Commenting on the slowness of regulatory bodies to keep up with the rate of change, Beeks’ Chief Information and Security Officer Oscar Neill says: “Back in January 2024 the National Cyber Security Centre released a paper stating advanced uses of AI in cyber operations are unlikely to be realised before 2025.1 Six months later there are already Proofs of Concept up and running, but there are no traditional ISO controls yet in place to cover security concerns. With legislation and governance lagging so far behind developments there is nothing to give firms assurances about security.”
Security
Explaining further Neill says: “The connectedness between AI systems is clearly a strength but as businesses increasingly require tasks to be automated, connectedness potentially poses a huge security problem. Novel cyber-attack vectors are emerging, with a mode of attack that is two to three times cheaper than human hackers. These can strike both live and training data, poisoning models before they are even put to work,” says Neill. “They can also coax ‘safe’ AI agents into exploiting known vulnerabilities in software systems.”
Privacy
With AI listening to conversations, reading documents and social media, analysing images, processing biometrics and location data, we live in a reality where individual and corporate privacy are at risk of significant dilution. “In the financial sector this is anathema,” asserts Rodgers.
Irresponsible use of public accelerators such as MS CoPilot or Chat GPT can reveal strategic intent to a damaging degree, yet these types of online applications are being used increasingly by employees across all levels of business organisations. “Firms need to enforce clear AI adoption policies to protect their privacy and that of their employees and customers,” he says.
Accuracy
The financial industry lives or dies by the accuracy of its lightning-fast data. However, the more complicated AI models become, the more crucial frequent retraining on fresh data becomes too. Whilst this is expensive, without it models can become prone to AI hallucinations, making up patterns that are not real. This raises serious doubts in the output.
“Retrieval Augmented GenAI (RAG) is an emerging architecture that can help – enabling dynamic lookup of new data and fine-tuning of the models without expensive retraining,” comments Rodgers. “Having smaller models which can operate entirely at the Edge (which for financial firms is often a colo or Exchange data centre) will also help address latency concerns, as well as cost of use and privacy issues. Just look at what Apple, Samsung and Google are doing to optimise AI models on the upcoming generation of smartphones, and imagine the power of this applied to financial markets use cases,” he says.
Infrastructure information to assist data centres and customers
There is scope for MSPs to train AI models on infrastructure data to provide better Network Operation Centre (NOC) support. They could also allow their customers to bring their own AI to the infrastructure data more easily. Rodgers says: “Embedding AI alongside our Analytics platform will enable predictive forecasting to identify anomalous behaviour and will provide an additional steering signal for trading systems.”
Streamlining Operations for quicker Settlement Cycles
As Exchanges strive for tighter T+1 Settlement Cycles, they can be delayed by the time it takes to review and take a position on the hundreds of daily Corporate Actions they receive.
Rodgers comments: “Since these notifications do not follow a uniform or pre-defined format there is some interest in using GenAI to develop conversational summaries of Corporate Actions to help determine stock positions much more quickly and efficiently.”
New revenue streams
The lid is set to be popped off the Market Intelligence possibilities as prudent GenAI inferences enable the curation and packaging of raw market data into saleable artefacts. The ‘democratisation’ of market insights through data monetisation can drive competitive innovation and enable smaller firms to take advantage of advanced analytics.
“This is not necessarily a field that Beeks would enter into ourselves,” says Rodgers, “but it is definitely an arena that we could support with optimised infrastructure and strategic partnerships.”
Forecasting
“Being able to predict the future is gold in Capital Markets businesses,” says Rodgers. “We can see that advancements in the Large Language Models (LLMs) underpinning GenAI are augmenting traditional Machine Learning (ML) techniques to allow for accurate forecasting with reduced costs and time to market.”
Beeks Response
“Beeks have listened to capital markets participants since 2011 and have built our Infrastructure-as-a-Service (IaaS) offering entirely in response to their needs and concerns,” says Rodgers.
“We are dedicated to ensuring our customers can focus on their strengths without having to worry about the performance, reliability or security of the underpinning infrastructure. There is no reason for us to deviate from this in the context of AI.”
Beeks has gained a reputation for extreme low latency compute on automatically monitored networks, providing cost-effective, elastic access to resource, and smart, dynamic use of hyperscalers’ bandwidth, in fully secure co-located private cloud environments.
Beeks is now considering which AI features will benefit its colo customers, including low latency AI, direct access to GPU-capable servers, and API access to GPU-accelerated hosted models.
“This makes Beeks vital participants in the conversation about GenAI in the financial sector,” Rodgers says.
