Don’t let the Chat GPT hype put you off: large language models (LLMs) and generative artificial intelligence (AI) are the secret weapons banking needs to face down Big Tech’s threat.
That’s the view of fintech entrepreneur Leon Gauhman, co-founder and CPO/CSO at digital product consultancy Elsewhen, which boasts clients including Spotify, Google, Microsoft and Mastercard.
With neobanks apps outperforming those of legacy banks and Apple and Google pulling ahead in the race for digital wallet dominance, Gauham argues that embracing LLM will allow banking incumbents to reset the scales in their favour.
If 2022’s over-hyped topic of tech conversation was the metaverse, this year’s equivalent is OpenAI’s ChatGPT. At the height of the hysteria, it seemed possible that the all-powerful chatbot would outrun the roles of academics, journalists and even lawyers.
However, the more we discover about the limitations of ChatGPT, it’s obvious there is still some way to go to iron out errors, before these technologies can deliver on their revolutionary hype.
Incumbents and neobanks should still pay close attention because the key takeaway from the evolution of ChatGPT and similar generative AI tools lies within the capabilities of large language models (LLMs) – the core tech that powers them. Microsoft and Google have both just launched LLM powered workplace collaboration tools with Microsoft also investing $10billion in OpenAI, the company behind ChatGPT.
The speed at which both companies are integrating LLMs into their businesses underlines the importance that big tech – an emerging competitor for incumbents and neobanks – attaches to this technology.
With Apple and Google forging ahead in the race for digital wallet dominance, new developments in AI-driven deep learning could allow incumbents and neobanks to reset the scales in their favour, in the following four areas:
1) Customers could have their own AI personal banker
large language models play into one of banking’s biggest advantages: its proprietary datasets. Using vast pools of customer data and insights, banks can deploy deep learning and natural language processing tools to create their own valuable IP; rather than trying to replicate their rivals with piggyback products.
For example, a well-trained version of ChatGPT provides scope to create a personalised AI banker that gives customers real-time recommendations perfectly tailored to their individual circumstances and needs. At a stroke, this could reinvent banking’s reputation for online user experiences, carving out space for genuinely innovative, responsive products that align with user expectations.
2) AI can supercharge banking productivity
Large language models aren’t solely about customer experience: the technology can also help employees by amplifying the resources they have access to. For example, Microsoft claims its Copilot tool will help Office users create presentations and prepare for meetings by providing relevant updates. Meanwhile, Google describes its AI tool as a “collaborative partner” that can suggest, summarise, and provide insights.
More broadly, these technologies can potentially revolutionise labour-intensive workflows around processes such as KYC, compliance and AML. With these core operations accounting for 15-20% of banks’ budgets, the implications for increased productivity, freed-up time, and budget are huge.
3. Large language models are the keys to top-gear digital transformation
LLMs have the potential to energise the entire financial services stack in support of digital banking, a reality that legacy players have famously grappled with to date. With new providers set to build large-scale open-source models, banks will have a direct channel to train LLMs using their data. This development, in turn, will allow banks to add a generative AI layer across a wide subsection of digital processes, from product design to mobile banking and cybersecurity to staff onboarding.
For example, Swedbank already uses generative adversarial networks to identify fraudulent transactions – drawing from synthetic modelling to understand and predict under-the-radar anomalies. In the race for digital banking transformation, this capacity becomes a superpower, enabling banks to leapfrog more nimble competitors.
4) Large language models herald a new era of proprietary modelling
LLMs allow banks to take proprietary data, extract valuable insights from it and use resulting action points to develop new personal banking/ wealth management strategies or a conversational interface.
LLMs could, for example, analyse ten years of banking data around mortgage defaults and use the findings to create a new, constantly learning underwriting framework for making better lending decisions.
In these times of hyper-innovation, banks are often criticised for being too cautious and slow to adapt. In the case of LLMs, a careful approach is best. As we have seen with Bing ChatGPT powered search and Google’s AI chatbot Bard, AI models tend to ‘hallucinate’, generating biases or inaccuracies. An AI chatbot hallucinating a wrong response about NASA’s James Webb Space Telescope is relatively benign.
By contrast, a bank brand’s AI chatbot hallucinating inaccurate financial advice would be seriously bad news. This potential risk makes rigorous human input and oversight in data training stages a necessity not a choice.
But caution is no excuse for inaction. If banks want to future-proof their businesses against the threat of big tech, they must push ahead on understanding, experimenting with and training LLMs. The biggest risk for banks, evident with other recent tech pivots, is to do nothing. Now is the moment to regain ground, using LLM as an agent of transformation to make customer-centricity a core business metric.