Community
The EU economy is facing long-term troubles, a consensus recently echoed by the media and European politicians.
Following GDP growth of just 0.3% in 2023, the EU’s annual GDP growth for 2024 continues to struggle between 0.2% and 0.4%, on the edge of a recession. Compared to the 2.5% GDP growth in the U.S. and 1.9% in Japan in 2023, the EU’s economic performance appears weak.
And the bad news doesn’t stop there.
The EU’s two major economic engines, Germany and France, have both lost momentum, with Germany particularly struggling. With minimal growth of just 0.1% in the third quarter of 2024, the German economy narrowly avoided a recession. Mass layoffs in the automotive industry and other economic challenges have also contributed to political instability in Germany.
Economists, entrepreneurs, and politicians generally blame overregulation for this predicament. French President Emmanuel Macron has warned: “The European Union could die because of overregulation and underinvestment.”
Ericsson CEO Börje Ekholm similarly stated that overregulation is “driving Europe to irrelevance.”
European businesses are forced to devote significant manpower and financial resources to studying and implementing the vast number of regulations introduced by the EU. This not only increases operating costs but also stifles innovation.
According to statistics from the Financial Times, since 2019, the EU has enacted approximately 13,000 pieces of legislation, covering sectors including financial services. In comparison, the U.S. passed only about 3,500 laws and 2,000 resolutions during the same period — less than half the EU’s legislative volume.
To comply with such an overwhelming number of regulations, European businesses have had to spend heavily on compliance management.
Take the European financial industry as an example:
According to a study by the European Banking Authority (EBA), compliance costs for European financial institutions in 2017 already exceeded 4% of their total costs annually. With the subsequent introduction of major regulatory measures such as MiCA, MiFIR, SFDR, and the upcoming DORA, it is estimated that the compliance costs for the European financial sector have likely surpassed 6% and could exceed 7% in the future.
By contrast, compliance costs for U.S. financial firms are estimated to be less than half of those in Europe.
Strict regulation has had a clear negative impact on European banks:
According to a report from the European Central Bank:
These figures highlight that, despite the larger scale of the EU banking system, it lags behind the U.S. in terms of efficiency and profitability.
The elected U.S. president, Donald Trump, has announced plans to significantly deregulate the U.S. financial industry, especially banking, upon taking office. This move will undoubtedly widen the gap in compliance costs between Europe and the U.S.
Although EU politicians like Mario Draghi and Ursula von der Leyen have pledged to reduce regulation, the EU’s complex legislative processes mean that substantive changes are unlikely for years to come.
To reduce compliance costs more quickly, many bankers are turning to compliance automation — using AI or IT tools to streamline compliance processes.
According to The True Cost of Financial Crime Compliance Study published by LexisNexis Risk Solutions, 80% of compliance departments reported that they are seeking opportunities to automate some of their activities in the next three years to regain competitive advantages against other banks.
The advent of large language models (LLMs) has significantly accelerated the push toward automating compliance tasks. The advantages of LLMs in processing natural language have sparked hopes of automating compliance work. Numerous RegTech startups have emerged, aiming to use LLM AI for contract review, KYC process handling, and financial regulation summarization.
However, can LLM AI truly achieve these goals?
In practice, many corporate clients have expressed dissatisfaction with the performance of LLM AI solutions from RegTech providers. This often results in low retention rates, with clients discontinuing use after a trial period.
Customer dissatisfaction primarily revolves around the fact that contracts and content generated by LLM AI consistently carry a small probability of hallucination errors. (AI models generate plausible-sounding but incorrect or nonsensical information.)
Since accuracy and reliability are paramount in legal compliance, such errors caused by hallucinations could become a fatal flaw for LLM AI in practical applications. However, the occurrence of hallucinations is inherent to the way LLM AI generates content. Current technologies, such as RAG, can reduce hallucinations but cannot completely eliminate them. Even OpenAI’s GPT-4 maintains a stable hallucination rate of 1.5%.
So, does this mean LLM AI cannot be applied to compliance automation?
Not at all. The key lies in correctly positioning LLM AI: while it cannot fully replace humans, it remains a powerful intelligent tool.
In general, AI should be regarded as a transformational software component that requires human oversight, allowing its immense potential to be better leveraged.
Thus, do not expect LLM AI to directly produce ready-to-use compliance outputs. Instead, use it to enhance efficiency within existing workflows.
Our team is also applying this principle in our work, using AI to optimize workflows in traditional GRC (Governance, Risk, and Compliance) software, targeting heavily regulated industries like finance.
We’ve introduced the following features:
Throughout this process, we do not use AI to generate any content; AI is only an auxiliary tool for human work. It acts as a “content courier,” performing precise searching, updating, and summarization of regulations while reminding users of necessary follow-ups.
It also allows humans to trace and verify the source provision for every piece of information provided by the AI, effectively avoiding potential hallucination errors at the source. We’ve also incorporated retrieval-augmented generation (RAG) to further reduce hallucinations.
We have integrated LLM AI into various modules of traditional GRC financial software (e.g., risk, policy). At every stage, AI simplifies workflows, significantly improving overall efficiency.
In this model, the flaws of LLM AI are mitigated, while its strengths are maximally leveraged.
This content is provided by an external author without editing by Finextra. It expresses the views and opinions of the author.
Amr Adawi Co-Founder and Co-CEO at MetaWealth
25 November
Kathiravan Rajendran Associate Director of Marketing Operations at Macro Global
Vitaliy Shtyrkin Chief Product Officer at B2BINPAY
22 November
Kunal Jhunjhunwala Founder at airpay payment services
Welcome to Finextra. We use cookies to help us to deliver our services. You may change your preferences at our Cookie Centre.
Please read our Privacy Policy.