Five Minutes with Celent's Monica Summerville

Monica Summerville is a seasoned expert in capital markets technology, bridging the gap between Wall Street and Silicon Valley. With over two decades of experience, she has been at the forefront of financial technology evolution, having managed front-office trading systems and market data distribution for major banks.

Now heading Celent's Capital Markets practice in London, she helps financial institutions navigate emerging technologies, with expertise in AI, cloud computing, blockchain, and data-related technologies.

Monica's career spans roles at ABN Amro, Price Waterhouse, and TABB Group, among others. She has also worked as a U.S. retail broker, holding Series 7 & 63 registrations. Monica holds dual US/UK citizenship.

In this Five Minutes Q&A, Monica shares her candid thoughts on the state of AI in capital markets, its current capabilities, challenges, and future potential.

This interview has been edited for clarity and length.

What's your perspective on the current state of generative AI in capital markets?

There's a lot of excitement, but implementations are often more challenging than people realize. Capital markets have been using AI for a long time, so many professionals understand what can be done with AI versus traditional rule-based or mathematical systems.

Generative AI is fun to play with, but there are still issues with context and understanding. We're seeing challenges with data bias and model construction for those interested in building their own LLM. It's a tool that has to be used carefully.

How do you define generative AI versus traditional AI?

From a technical standpoint, generative AI are AI systems that generate new content based on patterns learned from existing data. It's more than just machine learning - it has to genuinely create, not just give pre-programmed responses.

Generative AI tends to be better at understanding context than traditional AI. But to truly be generative, it needs that creative element - whether that's generating text, images, or other outputs.

Has generative AI lived up to the initial hype in capital markets?

There was a lot of hype initially, but we've quickly reached a point of some disillusionment. Some tech leaders have done proofs of concept showing cost savings, but it's mainly an efficiency play. That's good, but not as exciting as hoped – for now.

Banks have limited bandwidth for innovative tech projects – most IT budgets go to keeping the lights on and regulatory compliance. The feedback from some banks is that if they're going to invest in this technology, they want big upsides like new revenue streams, not just cost savings.

What are some key concerns around generative AI in finance?

There are big questions around data sources, ownership, legality, and reputability. The industry has dealt with similar issues before with alternative data. There are also concerns about system safety and accuracy. I've heard senior people say even 99.99% accuracy isn't good enough for some use cases.

Data privacy is a major issue, especially with client data. Most firms are looking at internal use cases first before client-facing applications because they're nervous about exposing their data to external models.

Where are you seeing generative AI actually being implemented?

Two main areas right now:

1. Code development - It's changing the programming landscape. A lot of coding is reusing and refactoring. Gen AI also helps with documenting old systems and frees up developers to focus on more valuable work.

2. Research - It's expanding analysts' capabilities. An analyst might typically cover 20-30 companies, but gen AI allows them to analyze a much wider universe. It helps summarize information, find correlations, and supercharge their ability to monitor industries.

It's also helping connect research to execution - tracking when portfolio managers act on analyst recommendations and how it impacts performance. This has been difficult before due to disconnected systems.

How are financial firms using AI models?

Most aren't building their own foundational models - that's extremely data and time intensive. The thinking now is to use existing models like OpenAI’s GPT-4 or open-source options like Meta's LLaMA, and then build on top of those.

Many are using private cloud setups where they can use their own data securely. Some are connecting to services like AWS Bedrock that provide access to various AI models. Very few are using fully outsourced solutions at this point.

What are your thoughts on the "hallucination" problem with generative AI?

In capital markets, many early implementations actually dial down the generative capabilities to zero. They're using the natural language understanding to parse queries, but then using traditional database lookups on the backend to retrieve information.

Some are using techniques like RAG (retrieval-augmented generation) and carefully defining prompts to constrain the GenAI. But you still can't fully trust the outputs without verification. For finance use cases where accuracy is critical, this remains a major hurdle.

What are your thoughts on the potential of AI to revolutionize areas like unstructured data analysis?

This is a big deal, but context is still a challenge. Take sentiment analysis - older systems would just look at the words used, but might miss negative news couched in positive language. Newer AI is better at understanding context by comparing to a broader set of information.

It's also getting better at understanding subjectivity - the same news could be good or bad depending on the investor's position. And we're seeing more data points incorporated, like voice stress analysis on earnings calls.

The ability to process more content is a key advantage. Cloud computing allows access to much more processing power. So while context is still a challenge, the sheer scale of data AI can analyze is leading to improvements.

Ultimately, I think we'll see hybrid approaches - using generative AI to structure unstructured data, then applying more conventional AI techniques to the structured data.

What do you think is underhyped about AI and emerging tech?

The convergence of different technologies is underhyped and really powerful. Things like AI and blockchain together, or quantum computing in the mix. Blockchain provides data security and decentralization, while AI is great at processing information and finding patterns.

Cloud has been a catalyst for a lot of this. The ability to access massive computing power is enabling new applications we couldn't do before. I think there are probably business use cases combining these technologies that we haven't even thought of yet, but could take off quickly once discovered.

Thanks for reading!

Interested in AI’s impact on finance? Every Thursday, I share curated news, expert analysis, and insider interviews. Sign up here:

Drop me a line if you have story ideas, research, or upcoming conferences to share. [email protected]