AI & digital transformation in the sell-side

Bloomberg Professional Services

The meteoric rise in artificial intelligence promises to change how the sell-side and the entire financial industry do business. Many firms are already using the technology to tackle day-to-day tasks such as sifting through mountains of data or extracting insights from earnings calls. 

However, AI is not a magic bullet to solve every problem, and many organizations will face some hurdles in harnessing this new technology’s potential.

The realities of artificial intelligence

These issues were front and center at this year’s Sell-Side Leaders Forum hosted by Bloomberg in New York. During a panel on AI and machine learning (ML), moderated by Tiffany Wong, Americas Head of Digital Strategy Solutions at Bloomberg, industry leaders Shawn Edwards, CTO at Bloomberg, and Dan Bosman, CIO at TD Securities, dug into some of the realities of integrating AI into a business. 

Key challenges

Kicking off the discussion, Dan Bosman identified three critical challenges for firms looking to adopt AI:: 

• Privacy and intellectual property

First and foremost is the murkiness around data. AI models are built on volumes and volumes of data. It begs the question: who owns all the data these models are being trained on, how confident are you in your data, and importantly, who owns and is accountable for what the model spits out? The regulatory landscape is still evolving, and companies will need to learn and listen before diving head-first into AI. 

• Risk and governance

This isn’t a new challenge but an evolving one. Most on the sell-side have already used artificial intelligence and machine learning models, like those at Bloomberg, to process volumes of data and generate insights. So, there’s a precedent for using these models in tandem with risk management and governance. But the reality is today’s AI models have introduced new complexities. These new models aren’t necessarily as transparent, so it’s a challenge to get governance teams to understand how the model is trained and how it works.

• Economic viability

Companies must focus on finding cost-effective solutions to the right problems. Drawing a parallel to the early days of cloud computing, Bosman noted that over-provisioning and choosing the largest models might not always be the best approach. Companies need to balance performance with cost efficiency.

Moving beyond chatbots: A holistic approach

Shawn Edwards echoed the challenges pointed out by Bosman. Additionally, Edwards highlighted that AI has already helped research analysis, investment banking, and sales and trading. The challenge now is not just about enhancing employee productivity; it’s about figuring out how clients can interact and trust AI-generated information.

And this concern goes beyond just relying on a chatbot. After all, chatbots and other AI tools that are powered by general large language models (LLM) don’t always understand the nuances of an industry — making them prone to generating inaccurate information.

Bloomberg uses AI as a complement, not a replacement, to existing workflows. For example, the firm’s earnings call transcript tools combine human expertise with AI capabilities to generate more accurate and relevant summaries.

The idea is not about feeding strings of text into a LLM and hoping for a contextually accurate answer. Instead, Bloomberg collaborates with internal sell-side analysts to craft questions that an AI can use to analyze an earnings call. These may be questions about forward guidance or potential macroeconomic effects. 

Getting there requires more than an off-the-shelf AI model. Bloomberg first uses technologies such as dense vector databases and retrieval augmented generation (RAG) to understand the essence of every paragraph in a transcript. Then, a large language model will summarize the most relevant information in a few bullet points. 

The goal is not just to answer questions but to help users understand the why behind the answer. At any point, users can click on the summary to see where the model sourced the information. That way, users are always in control and never have to second-guess whether the information is accurate. 

Looking ahead

What’s next for AI is a lot of investment. Investments in robust platforms that ensure data is accessible, organized, and trustworthy. It’s not enough to feed endless amounts of data into an AI model, The data must be carefully curated and the system must be trained to understand the nuances of the financial industry.  

That starts and ends with the people behind the screens. It is essential to have skilled professionals who can bridge the gap between technology and business needs, ensuring that AI tools are used to their fullest potential.

Bloomberg has been doing just that for decades, investing steadily in the data, infrastructure, and people that make AI a value-add to any business.