In 2024, many leading financial institutions have adopted AI-powered deep learning solutions for their cybersecurity and anti-fraud systems. With generative AI rapidly gaining traction across all sectors, a new era is emerging in customer and data services. Ahead of FStech’s inaugural The Future of AI In Financial Services conference, Silvia Iacovcich investigates trends, risks and expectations in an ever-changing landscape.

Artificial intelligence (AI) is being adopted by financial institutions across a broad range of operations and those who are not are racing to deploy AI, driven by technological advancements and the competitive landscape of the financial services industry.

Demonstrating the expected transformative effect of AI, a report from the McKinsey Global Institute (MGI) projects that the sector’s AI spend will increase from $35 billion in 2023 to $97 billion in 2027.

MGI estimates that across the global banking sector, generative AI alone could add between $200 billion and $340 billion in value annually, or 2.8 to 4.7 percent of total industry revenues, largely through increased productivity.

Over the past three years, traditional machine learning (ML) models have been widely used to tackle issues and improve services, including money laundering detection, anti-fraud monitoring, Know Your Customer (KYC) and biometric authentication.

With the advent of generative AI, advanced natural language processing (NLP) and large-scale language models (LLM), a range of impactful AI trends are now set to further reshape the financial sector.

News analytics – boosting investment decisions

Rajnish Kumar, head of investment technology and AI investment platform at AllianzGI, believes these new models are revolutionising the way AI handles unstructured data, as he tells FStech how his team is using them to significantly improve and speed up processing and analysis to deliver value.

Kumar highlights how firms are starting to leverage AI and generative AI models for sentiment analysis and data extraction to boost their investment strategies. Using AI is leading to more accurate reports with fewer errors, expediting investment analysis, he says.

“We now have better sense of sentiment tone using the new foundational model, as it is trained over larger data sets with billions of parameters, providing more variance. The bigger change is that we will see these AI versions from the past continuing to get better over time with the new inputs in each type of report,” he explains.

He notes that AI models can automatically review and analyse vast amounts of data, reducing the need for investment teams to sift through time-consuming reports, press releases, published news, and third-party analyses.

“Generative AI can transform tabular documents into more readable formats and assist in decision making by extracting key information, identifying risks, and generally facilitating better informed decisions; in investment use cases, earning call transcripts has also been very well-received by our investors, as it saves them from having to read through lengthy documents," Kumar says.

The investment professional also suggests that the integration of advanced AI and ML technologies into news analytics will transform the financial sector, as it provides real-time insight, improved market intelligence and trading strategies which are crucial for portfolio optimisation and strategic investment decisions to reshape conversations between players.

“This year, new analytics have seen a significant change, by leveraging AI to analyse millions of news articles, we can tag relevant news to specific companies and identify key issues and keywords that matter most to investors,” Kumar adds.

Hans Tesselaar, executive director at Banking Industry Architecture Network (BIAN), similarly emphasises that news analytics could be highly effective if banks and financial institutions correctly train their engines and filter the information they want to include.

“Filtering the pool of data by choosing it from reputable sources such as the United Press rather than from social media is crucial for this institution to take better informed decisions,” he notes. Because of this, the BIAN executive is also predicting a shift towards an alternative revenue model for news agencies, where specific data could be provided at a price.

Challenging accuracy

Despite enthusiasm for AI tech, generative models’ ability to produce creative data is flawed, and can provide inaccurate or misleading information that could – in a severe case – potentially damage a financial institution’s reputation.

While the current scenario seems to be positively improving, Guy Ward Thomas, partner at early stage and venture capital tech focused fund DN Capital, is calling for a more cautious approach, noting that generative models might not yet be suitable for many workflows within banks and other financial institutions.

The investor cites as an example its tech portfolio company HawkAI, which focuses on anti-money laundering (AML) solutions. According to Ward Thomas, the business believes it is still premature to implement generative AI in processes aimed at detecting complex issues such as fraud.

“If trained on the right data, generative AI could be used to write reports once money-laundering is detected, and to create more compelling case reports,” Ward Thomas explains.  “However, for now LLMs are still too hyperbolic for use in a regulated industry.”

“I think this will change once we've developed models trained specifically on much narrower models trained on data that is used for reporting financial crime to produce sober and more careful response,” he adds.

Ward Thomas suggests that banks might be hesitant to deploy generative AI models because of their training on literary documents and use of often sensationalistic language, which may be inappropriate for more specific and sophisticated case reports.

“At this early stage, errors around data will be unaffordable for financial institutions, as they will need to be 100 per cent sure of their data accuracy, and nobody wants to see their names on newspapers and have their reputations ruined,” notes Tesselaar.

Ward Thomas adds: “We expect that in a few years, LLMs will produce more financial contract reports and analysis as models and personnel are trained on the specific datasets, and these models are approved by regulators and compliance teams to mitigate risks.”

Financial institutions need to ensure that any generative AI solutions are thoroughly validated and tested before deployment is critical to avoid issues related to accuracy and reliability.

But with questions still being asked over its reliability surrounding critical tasks and data, the tech is thriving in non-sensitive activities such as customer service, marketing and sales, spawning new trends that are set to transform the world of client assistance.

More chatbots, please

DN Capital’s Ward Thomas believes AI in customer service has reached ‘the crest of the wave’ and predicts an increase in solutions designed to personalise customer service and provide valuable insights into customer satisfaction.

“We recently invested in a business offering a range of tailored solutions for financial firms including a contact centre, conversational automation platform, voice chat bots for large enterprises, focusing on lower sensitivity tasks of interacting and making customer service interactions more efficient and more effective,” Ward Thomas explains.

Thanks to AI, general customer service is turning into a much smoother experience before the client is passed to the human operators – and now many financial firms are starting to build in-house dedicated foundation models that can incorporate generative AI solutions.

Amsterdam-based multinational banking and financial services firm ING for example is expanding its offering with contact centres that offer a customer-facing chatbot, enriching its current chatbot with generative AI features to improve the customer satisfaction score and deflection rate, at ING spokesperson Marc Smulders tells FStech.

ING’s chatbot is currently able to handle 200 of 5,000 customer inquiries daily in the Netherlands and has colleagues reviewing every conversation to make sure that the system doesn’t use discriminatory or harmful language or hallucinate. The firm has also begun incorporating software engineering to accelerate daily coding activities and integrate ‘copilot’ risk and security aspects by leveraging GitHub copilot.

Incorporating generative AI in the client facing landscape encompasses different layers of customer service, including the sales and marketing area, where generative AI is being developed to personalise and tailor experiences.

ING aims to increase customer acquisition by using generative AI-powered personalised marketing for 3 specific segments – expats, young couples and gen z – always with consent from the customer, Mulders adds.

AllianzGI is in a similar state of development, with Kumar revealing the firm is building an internal tool, a copilot that will be trained to provide a more tailored experience for its clients. “A priority during the development includes refine its processes to ensure these services are fed with accurate data and include human training to maximise their effectiveness,” the AI specialist notes.

Tesselaar, who explains that BIAN is also implementing its own AI to facilitate conversations with banking institutions, adds: “Banks will use their data to improve customer service and generate better conversations between players, these are the low hanging fruits that banks must take advantage of, because this kind of technology is now available.” Banks own data themselves and financial institutions won’t need to be dependent on outside data to make decisions.

“To make it effective, we have trained our AI engine with in-house materials and documents. The engine will enable our members to quickly access information about our organisation and banking standards through a chatbot. I expect it will become extremely accurate, especially because no human can hold all the details on a specific product, therefore our member engagement will go sky high,” he says.

A regulatory switch

As AI continues to refine processes in financial services, experts predict that issues related to the use of personal data will persist in the coming years, despite the implementation of the European Union’s landmark Artificial Intelligence Act which entered into force at the beginning of August.

AllianzGI’s Kumar admits that data privacy and compliance poses difficulties for the German firm, adding: “At AllianzGI, we have spent months working with our legal compliance team to align with AllianzGI Tech standards. However, this is an evolving area, especially with the EU AI Act. Data privacy, security, and cybersecurity will continue to be key focus and challenges to keep up with AI development."

Despite financial organisations getting more comfortable with the regulatory steps, security and contractual arrangements with data providers will continue to transform and financial firms could struggle to keep up.

Facebook owner Meta fully embraced open source with the launch of its Llama at the end of July, a move which has opened the doors to new innovative resources for financial institutions who are increasingly exploring the use of advanced AI models to source and generate data; but the persistent risk in using these models remains to be proven.

“I think financial institutions are going to be nervous of open source and more comfortable with services that have been trained on more limited data sets, that fewer people have access to, where they can go to the provider of that model and interrogate the provider on that model, on what data has been trained,” Ward Thomas explains.

Finally, as AI tools develop to increase productivity, training models to sharpen their accuracy is crucial to avoid errors that have jeopardised giant firms’ reputation in other industries.

The case of Air Canada – which has been held liable by one of its customers for an AI chatbot giving passenger bad advice – highlights broader risks businesses must consider when adopting AI tools in all aspects.

“In the long run, it will be much worse if we create unreliable models by simply not taking time to refine them now,” Kumar says.

Understanding the sensitivity levels of different data streams and incorporating this data into a model requires corporate understanding, talent and training will be the key decision that will improve results without having to expose financial institutions to regrettable risks.

“We source data from vendors, and we want to make sure we are transparent with them, but I believe we are still in the early stages as regulations will continue to evolve in these areas,' Kumar highlights.

And even when a model is ready, regulatory approval can be a lengthy process, and ensuring that these models are safe, ethical, while complying with various standards and laws, is a significant task – and that is without even considering the larger existential questions of bias, the potential impact on society and the ever-present fear that low-level workers could be squeezed out of the market by powerful automation.

"The biggest obstacle is likely the need for effective change management and skilled personnel to train these systems. The industry is facing a talent shortage, which slows the pace of innovation. Instead of expecting rapid change, transformation often happens through the same individuals managing day-to-day tasks, complemented by input from additional talents. Although this process takes time, it is the most effective approach," says Kumar.

“There will be a lot of partnerships between banks and tech firms, but it will be a while before these will be rolled out to a wider audience,” concludes Tesselaar.

AI is transforming the financial services industry at a rapid pace, revolutionising everything from cybersecurity to customer service. With AI spending expected to nearly triple in the next three years, its impact is set to be enormous. Generative AI, which powers advanced natural language processing and large-scale language models, is particularly promising.

However, this rapid adoption comes with challenges. Ensuring the accuracy and reliability of AI models in critical areas such as fraud detection and compliance is crucial. The evolving regulatory landscape, highlighted by the EU's new Artificial Intelligence Act, adds another layer of complexity. Financial institutions must rigorously test and validate their AI solutions to avoid costly errors.

Despite these hurdles, the benefits of AI are too significant to ignore. By harnessing AI's potential, financial firms can gain a competitive edge, offering personalised services and making data-driven decisions.

The future of financial services is undeniably tied to AI, and those who navigate its challenges successfully will lead the industry into a new era of innovation and efficiency.

FStech’s The Future of AI in Financial Services takes place on 26 September in London.

---