The AI paradox in RegTech: Industrial revolution or recession catalyst?

Ani Petrova
2 days ago
5 min read

Updated: 6 hours ago

Over the past couple of months, Sebastian Siemiatkowski, CEO of Klarna, made headlines for all the wrong reasons — at least if you’re a compliance analyst, customer service agent, or anyone working in a traditionally “safe” white-collar job. On a recent episode of The Times Tech Podcast, Siemiatkowski didn’t mince words: “Unfortunately, I don’t see how we could avoid [a recession], with what’s happening from a technology perspective.” His comment referred to the rapid adoption of AI systems across every layer of modern finance — systems that, in Klarna’s case, were already handling workloads equivalent to 700 full-time employees.

It was a sobering prediction, particularly for those in compliance operations, where AI has surged from experiment to expectation. And yet, just weeks later, a different message emerged — this time from Threadneedle Street.

Project Hertha, a collaborative experiment between the Bank of England and the Bank for International Settlements, revealed that AI could detect 12% more illicit accounts and uncover 26% more previously unseen criminal typologies. Perhaps more importantly, it achieved these gains using entirely synthetic data, side-stepping the privacy minefield that has long stalked machine learning in financial services.

Two perspectives. One a warning, the other a glimpse of promise. Together, they illuminate a pivotal question: Are we standing at the edge of a compliance renaissance or on the brink of a white-collar recession?

A decade of compliance, transformed in 24 months: The AI paradox in RegTech

The speed of change is staggering. In just one year, AI adoption in UK financial services increased from 58% to 75%, according to a joint survey by the BoE and FCA. Foundation models — the kind underpinning LLMs like GPT or Claude — now account for nearly one-fifth of all enterprise AI use cases. HSBC alone is using machine learning to screen 900 million transactions each month across 40 million accounts.

The AI-in-RegTech market, once a subdomain of innovation budgets, is now a growth engine. It surged from $1.37 billion in 2023 to $1.89 billion in 2024, with projections exceeding $6.6 billion by 2028. That’s a 37% compound annual growth rate — not typical of niche software, but of a technological revolution in motion.

Yet, as Klarna’s experience shows, acceleration doesn’t come without friction. While the company’s AI assistant could handle most customer conversations, its quality was “lower than human workers,” forcing leadership to reconsider some layoffs. The human-AI gap still matters, especially where judgment, context, and nuance are required.

Which brings us to compliance.

Why AI is rewriting the DNA of financial crime operations

For decades, financial institutions have relied on rules-based engines to flag suspicious transactions, PEP hits, or KYC inconsistencies. These systems, while helpful, created a new problem: too many alerts, too little clarity. False positives frequently exceed 90%, leading to bloated teams, operational bottlenecks, and an epidemic of investigation fatigue.

AI offers a new path — not just faster flagging, but smarter triage.

Innovators like Hawk:AI promise near-human accuracy in alert generation while drastically reducing noise. Leo RegTech’s conversational agent, “Eva,” is tailored for UK fund regulation and speaks the language of compliance analysts. SAIFR, recently lauded at the RegTech Insight Awards, is trained specifically on enforcement actions — a critical advantage over generic NLP.

And then there’s the evolution within regulators themselves. In the US, the Financial Accounting Standards Board introduced AI tools into the GAAP taxonomy validation process. Compliance isn’t just watching AI from the sidelines — it’s embedding it into its foundational infrastructure.

But for all the headlines about better detection, the real frontier lies elsewhere.

Solving the 95% problem: After the alert, before the resolution

Detection is only the beginning. What happens after the alert — the human-involved investigations, the document hunts, the back-and-forth RFIs with customers — is where compliance teams spend the bulk of their time. This is the 95% problem: the post-alert operational workload that defines financial crime fighting today.

This is the space ComplyStream was built to transform.

When an alert is triggered, from a failed eKYC check, a flagged transaction, or a sanctions screening anomaly, the current process is slow, manual, and fragmented. Analysts jump between platforms, compose emails, chase documents, and rekey data. It’s compliance by copy-paste.

ComplyStream reimagines that process. It sits at the intersection of data, decision, and dialogue. Alerts from existing systems flow into a single task resolution layer, where data is enriched, context preserved, and communication automated. Need to request documents from a customer? A template-driven RFI goes out instantly. Need to track that customer’s response, verify the attachments, and ensure the analyst signs off? It’s all handled within one AI-native platform.

Crucially, ComplyStream doesn’t remove the analyst. It removes the overhead. The repetitive, administrative drag is automated, leaving investigators to do what machines still can’t: exercise judgment.

The compliance jobs debate: Displacement or elevation?

Siemiatkowski’s honesty about job displacement is rare. Silicon Valley still clings to the notion that AI will “create more jobs than it destroys,” even as private conversations suggest otherwise. Anthropic CEO Dario Amodei warned recently that up to 50% of entry-level white-collar roles could disappear within five years.

But history provides useful precedent. The introduction of Excel didn’t lead to fewer accountants — it led to more strategic ones. The printing press didn’t destroy writing — it scaled it.

In RegTech, we’re not seeing a wholesale replacement of compliance teams. We’re seeing a reshaping. New roles are emerging:

Prompt engineers who can extract meaningful insights from LLMs
Synthetic data specialists who generate safe training environments
Regulatory translators who convert policy into executable logic

Above all, the most valuable professionals will be those who understand how to orchestrate AI, where to let it run, where to reel it in, and how to align its outputs with risk appetite and regulatory expectation.

The regulatory crossroads

Far from resisting AI, regulators are adapting. The Bank of England, FCA, and PRA have all committed to a “pro-innovation, pro-safety” agenda, opening doors to controlled experimentation via sandboxes and innovation hubs. The balance is delicate. Explainability, auditability, and systemic risk remain front of mind. But there is now a consensus: AI isn’t optional. It’s a new layer in the regulatory fabric.

The question is how institutions deploy it.

Building the AI-native back office, human-in-the-loop

ComplyStream’s bet is simple but profound: The future of compliance isn’t human or AI — it’s both.

Our platform is designed from the ground up to be AI-native, yet human-centred. From automated case triage to integrated communication with clients via existing portals (Salesforce, Dynamics, etc.), every function is designed to elevate, not erase, the analyst.

It’s back-office automation not as a cost-cutting weapon, but as a creative tool. A force multiplier for judgment, not a substitute for it.

This approach — blending high-context workflow automation with decision transparency — is already proving valuable. In projects with clients like Universal Partners, ComplyStream has shown how intelligently applied AI can reduce time-to-decision, improve RFI accuracy, and bring real visibility to compliance KPIs. And it’s doing so without sidelining the people who know the risks best.

Through the valley: Why we’re optimistic

Every industrial shift begins with disruption. But history’s lesson is clear: long-term prosperity often follows short-term uncertainty.

We believe the same is true in RegTech.

Yes, roles will change. Yes, operational models will shift. But the mission — to protect the integrity of financial systems — remains constant. And AI, applied responsibly, can help us do that with greater precision, speed, and confidence than ever before.

In the paradox lies the opportunity. And in the post-alert world, ComplyStream is building the architecture to make it work for people, for regulators, and for the next generation of financial institutions.