Voices of compliance: How AI is transforming financial regulation

Financial firms have an AI problem. Not that they aren't adopting it—quite the opposite.

Take the risk and compliance sector, where stakes are exceptionally high and regulatory landscapes shift at dizzying speeds. A full 68% of firms are rushing to implement AI in these critical functions, yet only 32% feel ready to scale these systems effectively.

This disparity between ambition and readiness represents a fascinating challenge in financial compliance. Implementing AI in financial compliance isn't about showcasing innovation in annual reports – it's about building systems worthy of trust, capable of transparency, and designed to perform correctly from the outset.

While this challenge isn’t unique to financial services, the industry’s regulatory pressures make it a compelling place to start our new interview series – where we speak with innovators across sectors who have moved beyond interest and hype to build AI systems that are ethical, explainable, and solving real-world challenges.

Beyond the hype cycle: Building trustworthy systems in high-stakes environments

In the first of our interviews, we speak to Peter Kenny, Managing Director at ACA—a leading governance, risk, and compliance advisory firm serving the global financial services industry.

From mastering multilingual compliance to navigating the bumpy road toward automation, Peter shares how ACA is integrating trustworthy AI into real compliance workflows—offering hard-won insights into what it takes to scale responsibly and where the real opportunities lie.

Whether you're a FinTech founder, compliance lead, or AI vendor looking to break into this space, these conversations aim to spotlight what's working, what to watch out for, and what it takes to lead in a sector that demands rigor as much as speed.

Read on for Peter's insights on how AI-powered transcription is reshaping financial compliance, despite the skepticism of risk-averse compliance teams.

From lexicons to LLMs: The hidden history of AI in finance

“The current fashion would have us think that AI didn't exist before the invention of large language models. Of course it did. Vector analysis, natural language processing, voice recognition – these are all branches of the AI tree and critical components of modern compliance systems.

I can actually tell you what it was like before email was a commonly used business tool. Email started really coming onto the scene in the late 90s and it changed everything, as you can imagine.

We were scratching our heads trying to think, “Well, how are we supposed to treat this?” The closest thing that we could come up with was going back to the Securities Act of 1934 and a branch office manager's requirement to approve any communication that went from a registered person to a client in advance.

We thought, “Hey, why don't we just apply that standard?” We figured somebody would review any email that would go externally before it got sent out – and really all we managed to do with that was to crash the email system across the entire bank.

That was probably the longest 2 days of my life, professionally speaking.

From random sampling to AI-powered analysis

Early guidance suggested you could just look at 2% of your communications at random and figure out a better way to do it. The better way going into the early 2000s was what's known as a lexicon-based surveillance system.

A lexicon is a fancy word for dictionary, which is to say these are the particular words or phrases that I'm interested in from a compliance perspective.

What we quickly found was that we had to be mindful of the offsets – in other words, the sense that the word isn't – which was more challenging than finding the word itself. Take the word "ticket" – you would be interested in ticket from a trading perspective or from a gifts and entertainment point of view, like if a registered person was being offered Super Bowl tickets exceeding their annual limit.

But I can rattle off a dozen different examples of good uses of the word ticket that have nothing to do with the business.

As channels continue to proliferate and volumes have grown like a hockey stick, that offset becomes ever more important. This is where increasingly, artificial intelligence comes into the mix – first as a methodology to reduce that false positive rate, and then as a new means to generate those items of interest more rapidly.

The 90/10 problem: Why compliance officers remain skeptical

The kinds of transcription rates we're seeing now – I'll just take one example, Teams out-of-the-box, offers live transcription of any call in English with seemingly anywhere from 85 to 90% accuracy. If you told me that 10 years ago, I would have said no way, not possible. It seems miraculous.

But human psychology is such – and in particular compliance officers' psychology is such – that they're not impressed with the 90% you can get. They're worried about the 10% that you can't.

And that 10%, there's really diminishing returns in terms of your ability to improve the models, especially when you're talking about really challenging examples like where you've got a lot of crosstalk, emotion, or code switching – which is to say I started this conversation in English, but then I happen to switch over into Spanish.

A large language model may be able to handle English or Spanish if it recognizes that as such, but the key is in the acronym itself – it's a large language singular model. What we need are humungous multilingual models, and that's more complicated because you're talking about bringing together all sorts of different combinations.

Curious how we’re tackling this at Speechmatics? Learn more about our multilingual speech recognition.

Power tools for the mind: Reframing AI's role

The way that I often explain to folks when they get a little overwhelmed by AI is to think of it like power tools for the mind. Carpentry did not go away with the invention of electric power tools. It did, however, change dramatically. In a very similar way, we are now seeing these different kinds of power tools for the mind being developed. That does not mean the jobs go away. What it does mean is that the jobs will change.

It is incumbent upon each of us to understand how these new tools can be brought to bear, but also to recognize that again, there are limits to each of them. Knowing where those limits might lie is critically important in their judicious application.

The compliance landscape will continue evolving, but I've said this before and I'll say it again – our North Star is reliability. As long as we apply that lens to innovative technologies, we'll continue making progress managing regulatory and reputational risks more effectively.”

Peter Kenny is Managing Director at ACA, where he oversees the firm's eComms Surveillance product. In our next conversation, we'll explore how global financial firms navigate the Babel of multilingual compliance.

Kenny shares a brilliant gambling metaphor that perfectly captures the trade-offs: "It's like a roulette wheel. Do you put all your chips on one number or spread across the board?" When dozens of languages, dialects, and mid-sentence codeswitching enter the equation, perfect accuracy becomes a game of diminishing returns. Don't miss it.