I was recently asked to speak on a panel about the impact of AI on global capital markets at the International Capital Market Association (ICMA) FinTech and Digitalisation Forum. With the conference now over, it felt like a useful moment to step back and share a few thoughts on where the industry actually is, and where the real challenges still sit.

AI has been part of capital markets conversations for years, but over the past 12–18 months that conversation has shifted significantly. There’s now broad agreement across the industry that AI delivers real value, with the questions shifting from “can we use it?” to the much harder one:

“How do we deploy it safely, at scale, and in ways the market can trust?”

This write-up offers a brief answer to that question by outlining what works, and what doesn’t, based on a decade of experience applying AI to debt market data.

The bottleneck is not the model

As the process of adoption plays out across the debt markets, it’s becoming increasingly obvious that the limiting factor isn’t model capability. Large language models (LLMs) are powerful, but they are fundamentally dependent on the information they are given. In capital markets, that information is:

• Slow
• Highly unstructured
• Inconsistent and fragmented
• Difficult to validate

These constraints aren’t things that can be fixed by improved prompts or models. Without strong data foundations, even impressive systems remain fragile once they’re exposed to real-world scenarios.

Why so many systems fail in the real-world

This is where industry progress diverges.

While access to powerful models is accelerating, investment in data infrastructure, evaluation, and governance is not. In the past two years, many new AI startups have emerged, layering GenAI models on top of existing data estates. This often leads to impressive demos, but very few systems that are safe to deploy in real-world scenarios.

Why? Because the data layer and workflow integration underneath it can’t support safe, repeatable use.

Therein lies the real challenge: not just whether AI works — it’s whether it works reliably, repeatedly, and under pressure. Real-world systems need to operate on messy, incomplete, time-sensitive information, and still be trusted for high-stakes decisions.

Why data readiness is a moat

This is where 9fin’s approach has always been different. We’ve been using AI, data science and machine learning to solve some of the hardest problems in capital markets for almost a decade. We’ve spent years acquiring hard to access data, parsing and structuring unstructured disclosures, standardising financial concepts across issuers and deals, maintaining lineage, audit trails, and versioning, monitoring data quality and model behaviour over time.

The result: a platform designed around AI-ready data foundations, not just data consumption.

Structuring unstructured financial data is expensive and technically hard. The work compounds over time. Models can be swapped; data foundations cannot be recreated overnight.

Trust is built from data

In the next phase of AI adoption, the firms that succeed will be the ones whose systems hold up when real decisions depend on them.

At 9fin, we’ve always believed that trust is built from the ground up. By investing early in data foundations — structuring unstructured disclosures, maintaining auditability, and designing systems for real-world workflows — we’ve been preparing for this moment long before GenAI entered the mainstream conversation.

As the industry moves forward, the question won’t be “who is using AI?” It will be “whose AI do you trust?” And in capital markets, that trust will always be earned through data.

About the author

Hi — I’m Sean Löfgren, 9fin’s VP of Engineering. I joined in 2017 to build 9fin’s earliest data and AI infrastructure, alongside our co-founder and CTO, Huss. Now I’m responsible for building 9fin’s Engineering teams and leading the development of scalable, AI-driven intelligence products for the debt capital markets. I studied Computer Science and earned an MSc in Machine Learning from Imperial College London, and I previously worked in data at M&G.

If applying AI to one of the world’s largest and most important financial markets sounds like an interesting challenge to you, why not check out our open roles? The 9fin data & AI team is hiring.