When Probabilistic AI Meets Financial Markets

Why Trusted AI Will Define the Future of Financial Services

Artificial intelligence is evolving at an extraordinary pace, but one question continues to dominate conversations across financial services: can it be trusted? While AI has become one of the biggest topics in technology, organisations operating in highly regulated environments cannot afford to rely on systems that are only probably correct. For banks, asset managers, investment firms and other financial institutions, accuracy, transparency and auditability remain non-negotiable.

In this episode of FinTech Focus TV, host Toby Babb is joined by Rowland Park and Simon Gregory from Limeglass to discuss one of the most important conversations happening in financial technology today. Drawing on decades of experience developing technology businesses and solving complex information challenges for the financial markets, they explain why the future of AI is not simply about building larger language models. Instead, success will depend on ensuring AI has access to the right information, can demonstrate where its answers come from and can be trusted by organisations whose decisions have real financial consequences.

For businesses investing in AI in financial services, and for professionals building careers in financial technology, the episode offers valuable insight into where the industry is heading and why trusted AI could become one of the defining competitive advantages over the coming years.

AI in Financial Services Has Changed Faster Than Anyone Expected

Although artificial intelligence feels like a recent revolution, Limeglass has been tackling many of today's challenges for more than a decade. As Rowland explains, the business was founded almost ten years ago after he and Simon identified a growing problem while working within financial market research. The amount of information being produced across financial markets was increasing rapidly, yet professionals still struggled to find the most relevant insights when they needed them.

Their original goal was not to build generative AI. Instead, it was to help financial professionals understand vast quantities of information by identifying the meaning behind paragraphs and documents rather than relying solely on keywords.

Ironically, that early work positioned Limeglass perfectly for today's AI revolution.

While generative AI has transformed how organisations consume information over the past two years, Rowland explains that the underlying challenge has remained exactly the same. Financial institutions still need to identify the highest-quality information before any AI system can produce useful answers. Rather than making their original work obsolete, the rise of AI has reinforced just how important that foundation really is.

Today, banks are using Limeglass technology within client-facing solutions, including white-labelled AI assistants and internally developed AI products. Rather than replacing their clients' AI initiatives, the company's technology enhances them by improving the information those systems receive before any response is generated.

Why Probabilistic AI Creates Challenges for Capital Markets

One of the central themes throughout the conversation focuses on probabilistic AI.

Toby references discussions taking place across the industry about whether probabilistic models are suitable for regulated financial markets, particularly when decisions must be accurate and defensible. Unlike many consumer applications, financial services cannot simply accept answers that are "close enough."

Rowland's response perfectly summarises Limeglass' position.

"Probabilistic means problematic."

While the phrase is deliberately memorable, it reflects a much broader point about enterprise AI.

Traditional software behaves deterministically. The same input produces the same output every time. Large language models operate differently. They generate responses based on probability, meaning identical questions can produce different answers on different occasions.

Simon explains that this creates an entirely new way of thinking about enterprise technology. AI will always provide an answer, regardless of whether it has complete information available. More importantly, the response is often presented confidently and coherently, making it difficult for users to recognise when something is inaccurate.

For financial institutions, where regulatory compliance, investment decisions and market analysis depend upon factual information, that uncertainty introduces significant risk.

Rather than criticising AI itself, the discussion focuses on solving this problem by improving the quality of the information available before the model begins generating answers.

Better Information Creates Better AI

A recurring message throughout the episode is that AI is only as effective as the information it receives.

Simon explains that many people mistakenly think large language models function like sophisticated search engines. In reality, they still require accurate, relevant information to generate trustworthy outputs.

If the correct facts are unavailable, AI cannot magically create them.

This shifts attention away from prompt engineering alone and towards retrieval, search and knowledge management.

Limeglass approaches the problem by building an extensive knowledge graph that maps financial concepts and relationships across markets. Over many years, the company has created an evolving structure containing hundreds of thousands of interconnected financial topics, with new concepts continually being added as markets develop.

This enables financial content to be tagged with significantly greater precision than traditional search technologies.

Instead of simply searching documents using keywords, the platform identifies the actual meaning contained within paragraphs, allowing AI systems to retrieve highly relevant information while filtering out unnecessary content.

According to Simon, improving retrieval solves multiple problems simultaneously. More accurate information leads to more reliable AI outputs while also reducing computational costs because the language model processes less irrelevant information.

In other words, better search creates both better answers and greater efficiency.

Trust Will Become AI's Greatest Competitive Advantage

Perhaps the strongest message throughout the discussion is that the next phase of AI adoption will not be won by whoever builds the biggest model. Instead, organisations will differentiate themselves by building systems people genuinely trust.

For financial services, trust extends far beyond accuracy alone.

Simon introduces what he describes as the key principles underpinning trusted AI, including auditability, authority, provenance and ensuring that evidence can always be traced back to its original source.

Rather than expecting users to blindly accept AI-generated responses, Limeglass enables financial professionals to compare AI outputs directly against the original analyst research and supporting documentation. This allows users to understand not only the conclusion generated by AI but also the evidence behind it.

For regulated industries, this transparency becomes incredibly important.

Investment professionals, analysts and compliance teams need confidence that recommendations can be verified, audited and explained. AI therefore becomes an assistant supported by evidence rather than an opaque system asking users simply to trust its conclusions.

As Toby highlights during the conversation, financial services and healthcare remain two of the industries where trust matters most. The opportunity for AI is enormous, but adoption will ultimately depend upon whether organisations can demonstrate reliability alongside innovation.

For firms investing in enterprise AI, and for professionals building careers across financial technology, data engineering, AI infrastructure and capital markets technology, this shift represents an exciting evolution. As AI continues to reshape financial services, demand is likely to grow for specialists capable of combining machine learning with governance, search technologies, information architecture and trusted enterprise systems, areas that are becoming increasingly important across the financial technology recruitment market.

Enterprise AI Must Move Beyond Proofs of Concept

Another fascinating part of the discussion centres on the difference between impressive demonstrations and genuine enterprise deployment. While many organisations have successfully built AI proofs of concept, moving those solutions into production presents a completely different challenge.

Rowland explains that many financial institutions initially believed they could build every aspect of enterprise AI internally. Like many businesses across financial services, they recognised the potential of generative AI and invested heavily in developing their own capabilities. However, as projects matured, many began encountering the practical limitations that only emerge when AI is deployed at scale.

The issue is not whether AI works. It clearly does. The challenge is whether organisations can consistently trust its outputs when clients, regulators and investment professionals depend on those answers every day.

Simon reinforces this point by explaining that many of the biggest challenges sit beneath the surface. Creating a chatbot or demonstrating an AI application has become relatively straightforward. Modern large language models make it possible to develop impressive interfaces quickly. The difficult part comes afterwards, when organisations need to ensure those systems remain accurate, repeatable, auditable and cost-effective over months and years of production use.

That distinction between experimentation and operational deployment becomes one of the defining messages of the episode.

As Toby remarks, productivity has become one of the defining business challenges across almost every industry. AI offers enormous potential to improve efficiency throughout financial services, but only if organisations can rely on the technology consistently enough to embed it within day-to-day operations.

This is where enterprise AI differs from consumer AI.

Rather than seeking interesting responses, financial institutions need dependable ones.

Why Search and Retrieval Remain Fundamental to AI

Throughout the conversation, both guests repeatedly return to a surprisingly simple idea.

Search still matters.

Although generative AI has changed how users interact with information, the underlying challenge of finding the right information has not disappeared. Instead, it has become even more important.

Simon explains that AI cannot answer questions accurately unless it first receives the correct evidence. If the retrieval stage is weak, even the most advanced language model cannot compensate for missing or poor-quality information.

Limeglass therefore focuses heavily on enriching information before AI ever processes it.

Rather than simply indexing documents, the platform identifies meaning, relationships and context across financial content. This enables AI systems to retrieve concise, highly relevant evidence while avoiding unnecessary material that can introduce noise or increase the likelihood of hallucinations.

The guests also discuss the importance of timeliness.

Financial markets change constantly. Company announcements, earnings releases, macroeconomic events and market-moving news can alter investment decisions within minutes. Traditional search engines often prioritise relevance based on historical popularity rather than recency, but financial professionals frequently require the newest information available.

Ensuring AI retrieves both relevant and current information therefore becomes essential for producing reliable outputs.

The discussion highlights that improving retrieval not only increases accuracy but also reduces computational costs by limiting the number of tokens consumed by large language models.

For financial institutions investing in enterprise AI, improving retrieval may therefore deliver better performance and lower operating costs simultaneously.

Building Trust Through Transparency

One of the strongest themes running throughout the episode is transparency.

Simon explains that one of the biggest mistakes organisations can make is attempting to validate AI using another AI system. If users cannot trust one model completely, relying on a second model to verify the first simply creates another layer of uncertainty.

Instead, answers should always be traceable back to their original evidence.

Limeglass achieves this by presenting AI-generated summaries alongside the original analyst research or source material from which the information was derived. Users can compare both versions side by side, preserving the nuance and context that often disappears during summarisation.

For regulated industries, that distinction is significant.

Auditability, provenance and explainability are becoming increasingly important considerations as financial institutions adopt artificial intelligence across research, investment, compliance and client-facing services.

Rather than replacing human judgement, trusted AI enhances it by making high-quality information easier to access while allowing professionals to validate every important conclusion.

As financial services continue embracing AI, this balance between automation and accountability is likely to become one of the defining characteristics of successful enterprise implementations.

AI Hype Versus Production Reality

Towards the end of the conversation, Toby raises another timely topic.

Artificial intelligence has undoubtedly created enormous excitement, but with that excitement comes hype.

Rowland and Simon acknowledge that AI has become one of the biggest talking points across LinkedIn and the wider technology industry. New experts appear almost daily, countless products promise revolutionary capabilities, and businesses frequently showcase ambitious visions for the future.

However, delivering production-ready AI is considerably harder than presenting ambitious ideas.

Simon explains that Limeglass spent around twelve months undergoing rigorous adversarial evaluation with a major financial institution before proving the value of its technology. Those extensive assessments demonstrated that improving retrieval and information quality produced measurable benefits within enterprise environments.

That real-world validation distinguishes practical AI solutions from marketing claims.

The guests are careful not to criticise the rapid pace of innovation. Instead, they recognise that large language models will continue improving. Their point is that the underlying challenges surrounding trust, information quality and governance will remain important regardless of how sophisticated AI becomes.

As Rowland explains, even as AI improves, organisations still need mechanisms that reduce uncertainty before information reaches the model.

Garbage in will always produce garbage out.

That simple principle remains just as relevant today as it was before generative AI transformed the technology landscape.

What This Means for the Future of Financial Technology

The conversation concludes by looking ahead.

Having established a strong presence across the sell side, Limeglass is increasingly exploring opportunities across the buy side, applying its technology to financial research, earnings calls, regulatory documentation, market news and even internal communications.

Because its knowledge graph spans hundreds of thousands of interconnected financial concepts, the platform can connect structured and unstructured information across multiple data sources, creating richer relationships between research, market events and financial data.

For investment professionals, that opens up entirely new ways of discovering insights.

For financial institutions, it creates opportunities to build enterprise AI solutions that are more accurate, more transparent and ultimately more trustworthy.

For the wider financial technology industry, the episode also highlights an important trend that extends beyond software alone.

As enterprise AI becomes embedded across financial services, organisations will increasingly require professionals who understand data engineering, AI governance, information architecture, enterprise search, machine learning infrastructure and financial technology platforms. Demand for these specialist skills continues to grow across investment banks, asset managers, capital markets firms and FinTech companies alike, making this an exciting period for both employers and professionals working within financial technology recruitment.