The past, present, and future of AI in financial services

As the use cases for AI in financial services continue to grow and deliver value for organizations and customers alike, I’d like to provide some insight on where I think the technology is delivering most value at the moment, and also where I think we are headed. Firstly, though, a little on how far we’ve come.

Many people don’t realize that AI has been around since the 1950s. Models like linear regression, support vector machines, and many more have been used for decades. The application of traditional and novel algorithmic design choices continues to unlock real value in financial services.

Deep learning has also been around for a long time, but use cases only gained traction in the mid-2000s as datasets and computational power expanded enough to showcase its true potential. As financial services use cases evolved, deep learning became a key tool to solving problems we otherwise could not accomplish with more rudimentary machine learning models. 

Today, we are seeing a lot of investment in neural network-based models, totaling billions of parameters, and being trained on multi-billion point datasets. Compared to first-generation models, these models are computationally expensive and highly complex. The ability to train models of this size, with increasing ease, shows how far technology has come.

Competition, collaboration and customer experience

The computational hardware and the advent of increasingly powerful GPUs is expanding the boundary for larger neural networks being trained on massive datasets. All of this advancement is paying off for banks, consumers and the fintech ecosystem at large. Creative solutions from third parties, fintechs, and challenger banks are solving tough problems in the financial services sector, which is pushing incumbent banks to challenge the challengers and harness the power of AI in the products and services they offer to their customers. One thing incumbents have over their agile rivals, however, is troves of data, which is the lifeblood of AI. Knowing how to harness this data is perhaps the key challenge incumbents face, which is naturally leading to increased collaboration with fintechs and third-party data specialists.

Customers are greatly benefitting from the current competitive environment on both the corporate and retail banking sides. Personal finance is a great example of the latter, as AI now allows consumers to have a personal assistant for their own finances, democratizing access to advisory services.

While chatbots have existed for a number of years, it is only recent advancements that have enabled more compelling use cases. From traditional rule-based bots to research in deep learning based generative model based bots, we have seen tremendous advancement in chat-bot quality. Neural network-based chatbots, for example, can provide an easy interface for users to get spending advice, understand their balance and spending, and get insight into transaction details.

With the advent of multi-billion parameter knowledge models, finetuned on personal finance data, the performance and usability of chatbots is better than ever, with most capable of delivering detailed account insights. The increase in the move towards digital channels brought about by the pandemic has also created a wealth of data on which models can be trained, further improving personal finance products and services.

These use cases have tangible and immediate benefits for both banks and consumers. Customers no longer have to spend time waiting in line to speak to customer services representatives when they know exactly the information they need and how to access it via automated channels. Meanwhile, banks can improve customer experience and reduce overheads tied to their customer support cost-centers.

Looking to the future

Over the next few years, I believe there will be an increase in data-sharing between banks and fintechs. The data banks hold is sensitive and highly safeguarded, but improvements in federated learning and synthetic data generation methods will allow partnerships between those developing models, and those holding the data, to flourish.

The next area is natural language processing (NLP). As mentioned above, there have been huge advances in massive billion parameter neural network-based architectures trained on multi-billion point datasets. From these, there are countless possibilities for transfer learning and knowledge distillation on more specific tasks. One only needs to look at the incredible use cases enabled by GPT-3 to understand the potential of such models.

Another area where I believe data and AI will create opportunities is in providing access to credit through leveraging alternative data. Using such data, banks can begin to provide services to the “credit invisible,” unlocking financial support for those without traditional credit histories, providing a fairer environment for consumers to gain access to capital.

In a similar vein, an area that I believe will see significant investment is algorithmic fairness and the push for elimination of algorithmic bias in predictive and decision-making models. Increasingly, banks will be required to understand and explain how and why decision-making models arrive at certain outcomes, particularly if they negatively affect certain groups or individuals.

AI can often be maligned by those who believe every advancement will move us closer to the decimation of the job market or even a more outlandish science fiction scenario. But AI is increasingly being used for good across industries, from healthcare and environmental modeling to financial services. Data is the most effective asset we have in the fight against inefficiency, inequality and injustice and AI is the means by which we will unlock its true potential.