Alan Turing’s work with machine learning in the 1950s laid the foundation for deep learning in the 1980s and 1990s. The financial sector has always been at the forefront of implementing these technologies. But there is no denying that generative AI’s release to the public marked a paradigm shift.
Just over two years later, AI is accelerating innovation across every sector of the economy. Once again, the financial sector is well-placed to reap the rewards. However, the industry dances on a double-edged sword. While AI promises to enhance everything from efficiency to research and potentially liberate new capabilities, it also increases complexity and risk.
There are many reasons, not least of which is the lack of an accepted, well-understood and widely recognised definition.
Amidst the ever-shifting regulatory landscape, the definition of ‘AI system’ is broad, if not fluid. The EU Artificial Intelligence Act gave us a working definition. However, the Commission’s recently published guidelines on AI system definition make clear that a) they are not binding and b) authoritative interpretation may only be given by the Court of Justice of the European Union.
“[M]y conclusion after reading the 12 pages is that the guide leaves more questions than answers. However, it is indisputable that the Commission is proposing a broad interpretation…” — Jesper Løffler Nielsen, Ph.D., Cert. IT Attorney, Tech Law Team Head at Focus Advokater
Clear definition or not, AI and related technologies are here to stay. When ChatGPT stoked our collective imaginations in November 2022, the big question was: How can we use this to create a better, stronger, more resilient, business?
A model suggested by AI thought leader and former Harvard Business School professor John Sviokla, with colleagues Paul Baier and Jimmy Hexter, describes which industries are best placed to benefit.
Industries relying on the interpretation and manipulation of Words, Images, Numbers and Sounds (WINS), have the most to gain — and the most to lose. According to Sviokla “industries with a high percentage of cost in WINS work and that are highly digitised are ‘In the Crucible’”. He suggested that if a business ‘in the crucible’ did not embrace GenAI — and fast — it risked falling behind competitors.
Financial services firms sit clearly ‘In the Crucible’. It is no surprise, then, that the industry spearheads AI investment. In 2023, that amounted to roughly $35 billion worldwide. AI investment in the financial sector is projected to grow to $97 billion by 2027. As a percentage of revenue, that trails only media, entertainment and sport.
It is not a question of whether big data, AI and (eventually) quantum computing will affect financial services firms. The question is how can you leverage innovative tech’s potential, at speed, without creating chaos?
For more than a decade we’ve been told that “data is the new currency.” And we know that finance firms utilise vast amounts of data, from financial modelling and customer analytics to risk management and fraud detection. Put simply, data drives decision making.
But today there is an almost absurd abundance of data. According to the latest estimates, 181 zettabytes of data will be generated in 2025. For perspective, that is enough data to fill 45.25 trillion DVDs. Stacked on top of each other, those DVDs would stand 54,300,000 km high, or 135 round trips to the moon.
There is good news, however. Data integration firm Qlik says that “technically we’re reaching an inflection point where ingestion and transformation of data can be done in real time and hybrid transactional and analytical data can be stored and processed in the same place.” This immediacy will affect everything from operational efficiency to risk mitigation and decision making.
In the near term, two promising areas are retrieval-augmented generation (RAG) and small language models (SLMs).
We’ve all heard about GenAI hallucinations. Whilst it might be fun to prompt your favourite LLM into wrong answers, it would be unacceptable to deploy delusional customer-facing AI agents. By augmenting the LLM with a relevant vector database, the problem is mitigated. Implementing RAG requires additional steps and data sources need to be regularly updated. But the output is far better than using an LLM or retrieval-only system alone.
As the name suggests, SLMs get trained on a smaller, curated dataset when compared to LLMs. SLMs can be used in a niche solution such as an AI agent. For example, a firm might deploy an AI agent trained only to answer customer questions about insurance products, or personal loans.
Three advantages to using SLM AI agents are:
Ultimately, firms may need to develop hybrid systems that include data collection, vector search capability, RAGs and integrated SLM agents.
Automation is one of the top use-cases for AI in finance. Three of the top four industries with high potenti AI automation/augmentation are banks, insurance and capital markets. In 2023, Accenture concluded that 54% of work done in banking has a high potential for automation. A further 12% has a high potential for AI-driven augmentation.
Today, most banks in developed economies use, or plan to use, AI automation to drive efficiency and cut costs. A 2024 survey by the Bank of England and the FCA found that 75% of UK financial firms are already using AI and that 55% of all AI use-cases have some degree of automated decision-making.
Imagine creating a single, high-quality piece of content and then handing it off to an AI with the instruction to make it useful for any medium or platform whilst meeting all compliance standards. We’re not quite there yet, but AI-enhanced COPE is already in use.
For example, your research team might create an in-depth report intended for your top clients. Using a COPE strategy, the team would re-purpose the report into a series of blog posts. Next it would create infographics for social media, a video script and a podcast episode. This keeps the core message consistent, while tailoring the delivery to the specific audience. This is the power of COPE.
“[It] is around personalisation of content — not so much personalising content, but distributing content in a personalised manner, so that it lands with the right people, they can utilise it, they can act on it […] with the highest effect." — Gyula Szathmary Head of Partner Content & Activation, Director, Saxo Institutional
Because finance is a regulated industry, AI is not the “everything solving rocket ship” some would have you believe. But it does make accomplishing these tasks faster. Since timely, accurate information is crucial, this approach can save resources and increase efficiency.
Another use-case for AI-enhanced COPE methodology is regulatory updates across all platforms and touch points. Global finance companies already face the challenge of keeping abreast of different regulators. Using a tech-based approach, e.g., application programming interface (API) along with COPE, can streamline the process.
Infrastructure is a potential choke point for implementing AI and related technologies. According to a recent survey conducted by Nvidia:
“Forty percent more [financial services] companies reported increased AI infrastructure spending compared with the previous years and 98% of management said they’ll further increase AI infrastructure spending in 2025.”
Where is the spend going?
Financial firms are building "AI factories" with full stack accelerated computing. This signals a shift from exploration to deployment. It also positions companies to take advantage of agentic AI (see SLMs above). These systems leverage data and use sophisticated reasoning to solve complex problems. Applications include personalised financial advice, automated risk assessment and real-time fraud detection.
Another important consideration for cloud-based solutions is energy-efficiency. According to Nvidia’s survey, the two biggest challenges are the cost of transitioning to energy-efficient computing and tracking energy consumption on a per-workload basis. GPU-enhanced cloud services can offer improved power usage effectiveness (PUE).
The Quantum Question
If the idea of computation speeds exponentially faster than Elon Musk’s Colossus, or the Lawrence Livermore National Laboratory’s El Capitan makes you salivate, take a deep breath.
On December 9, 2024, Google announced that its Willow chip “performed a standard benchmark computation in under five minutes that would take one of today’s fastest supercomputers 10 or 10 septillion years. […] It lends credence to the notion that quantum computation occurs in many parallel universes, in line with the idea that we live in a multiverse…”
Not to be outdone, on February 19, Microsoft unveiled its Majorana 1 chip, touting a “new state of matter” in the process. While the physics behind Microsoft’s topological qubits is still in question, both companies’ research shows promise.
Whether commercially relevant quantum computing is months, or years, away is beyond the scope of this paper. However, as witnessed with generative AI, speed of implementation will be key to gaining — and maintaining — a competitive edge. In other words, watch this space.
In addition to real-time data processing there are several ways AI can enhance competitive intelligence. For example, using natural language processing (NLP), AI can comb through competitors’ earnings call transcripts looking for strategic shifts. Unlike in the past, where analysts might need hours or days to read these transcripts, AI can do it in a few seconds. AI can also use pattern recognition to model and forecast future scenarios. High-frequency trading firms have been using machine learning for years to analyse market data, identify opportunities and execute trades in fractions of a second.
Artificial intelligence, machine learning, deep learning, big data, SLMs, agentic AI, cloud computing, quantum computing…
We can’t ‘not’ talk about them because there is so much potential. They have, for lack of a better phrase, captured the zeitgeist.
However, there is another side that must be considered…
"My freedom will be so much the greater and more meaningful the more narrowly I limit my field of action and the more I surround myself with obstacles. Whatever diminishes constraint diminishes strength. The more constraints one imposes, the more one frees one’s self of the chains that shackle the spirit.” — Stravinsky
The intersection of innovation and regulation in finance is somewhat like antecedent and consequent in musical composition. Innovations like disruptive fintech, blockchain, or AI introduce new risks for regulators to address. Regulators, in turn, introduce new laws. The new laws increase compliance obligations, which creates an opportunity for innovative technology to meet those obligations. Call and response, push-and-pull.
Innovative technology has aided market participants and regulators alike. The use of technology for compliance and supervisory monitoring predates the GFC. However, the ensuing regulatory requirements and increasing complexity in financial markets was “a catalyst for greater use of technology.”
If the core purpose of regulation is to ensure stability, prevent systemic risk and protect consumers, then risk from the use of technology applies to both sides. As the OECD reminds us:
“Advances in technology do not render existing safety and soundness standards and compliance requirements inapplicable. Many of the risks related to AI are not necessarily new or unique to AI innovation but rather exacerbated and amplified by the use of such innovation…”
The OECD also noted that in the absence of explicit regulation regarding the use of AI, “existing rules or guidance should generally apply regardless of whether the decision came from AI (with or without human intervention)…” This includes guidance on third-party risk management, cyber-security, operational resilience regulations and consumer protection laws.
In the coming 12 months technology will continue to impact the financial sector. That includes increased use of regulatory technology (RegTech) and supervisory technology (SupTech).
Will AI and automation fully take over regulatory compliance functions? Even with recent improvements in AI and ML, there are challenges.
Examples include data privacy, cybersecurity threats and integration/interoperability with legacy systems. Another consideration is adaptability. As RegTech evolves, regulations will inevitably change to keep up.
RegTech systems still need human oversight to update compliance frameworks. Compliance often requires nuanced interpretation, which means human judgement. Therefore, a more likely scenario is a hybrid approach where humans handle decision-making and policy updates and RegTech automates routine compliance.
The same technologies that launched the RegTech industry propelled a similar rise in SupTech. According to the Cambridge SupTech Lab, “Over half the world embraces technologies and data science tools to enhance financial supervision. A total of 164 financial authorities in 105 countries have live SupTech implementations.”
There are obstacles to SupTech adoption. A lack of funding is a persistent challenge, but limitations in data analytics and technical IT skills top the list. As with RegTech, legacy systems were not designed to integrate seamlessly with new SupTech solutions. Furthermore, some emerging economies are still in the digital transformation developmental stage.
Closer to home, EU supervisory authorities are using SupTech. For example, the ECB developed tools such as Athena, which uses natural language processing and automated translation capabilities to help supervisors “find, extract and compare information from a corpus of millions of articles, supervisory assessments and bank documents.”
The European Banking Authority views Digital Transformation, RegTech and SupTech as a triangle, with “RegTech and SupTech solutions […] ready to become key for financial market participants and regulators to ensure an effective, safe and sustainable market.”
SupTech use continues to expand across supervisory areas, including anti-money laundering (AML), countering the financing of terrorism (CFT) and counter proliferation financing (CPF). Worldwide, SupTech adoption is likely to increase in the coming 12 months.
While the use of RegTech and SupTech has improved compliance and efficiency, it also introduces potential systemic risks. For example, expanded use of cloud-based platforms increases attack vectors for hackers. By targeting regulatory infrastructure, hackers could disrupt financial stability or manipulate data, causing a ripple effect across the sector.
Another potential risk is over-reliance on a small number of RegTech or SupTech providers. If multiple large financial institutions rely on one provider, it could become a single point of failure in the case of cyberattack.
The EU has been proactive in addressing potential systemic risks. For example, consider the Digital Operational Resilience Act (DORA). As the financial sector becomes increasingly dependent on information and communication technology (ICT), it exposes entities to third-party risk. DORA was created to ensure financial sector “stay[s] resilient in the event of a severe operational digital disruption.”
From a compliance or regulatory perspective, it may be prudent to “think first, act later.” However, that could prove costly. Again we look to AI thought leader John Sviokla for insight:
“Firms with heavy reliance on WINS work need to act today to fend off stiffer competition and to overcome disruptive competitors within 36 to 60 months. Don’t be caught with high costs, old processes, a data disadvantage, fleeing talent and expensive capital.” — John Sviokla
Is there a way to achieve balance — an ecosystem that fosters innovation whilst also maintaining stability, security and consumer protection? In short, yes. Using a regulatory sandbox approach acts as a bridge between innovation and regulation. In the EU, “currently operational sandboxes have adopted a model based on customised application of existing rules, using the available tools for supervisory discretion.”
In brief, the intersection between innovation and regulation requires a balanced approach from all stakeholders. The financial sector is in a unique position to take a leading role, to leverage and highlight the power of collaboration from the top down. In practical terms, this includes the board of directors, executives, information technology, risk management, analysts, sales and marketing. To maximise return on investment, every department can — and should — play a role. Achieving this level of collaboration may require a cultural shift toward continuous learning. Compliance may need to change from ‘burden’ to competitive advantage. And in keeping with the overall idea of collaboration, firms may need to establish cross-functional teams or working groups.
Finally, as we move at speed into the digital future, keep in mind these words attributed to Heraclitus: “The only constant in life is change.”
