By John Gist, Global Head of Innovation Labs at Fidelity International
Would you take financial advice from a virtual avatar?
Five years ago, UBS did something rather incredible: they digitally cloned their chief economist. With his likeness captured by more than 120 high-definition cameras then rendered in 3D, Daniel Kalt sat down and diligently trained his AI virtual double in the art of financial analysis and advice. And just like that, the multinational investment bank’s most senior economist had the power to “meet” with multiple clients simultaneously via interactive video chat.
At the time, UBS’s Companion program was seen as a technological miracle — and now, with the proliferation of generative AI, human-like digital assistants are set to shake up the financial services world once again.
A human experience
UBS’s digitization of Kalt gave us a compelling glimpse at the possibilities that arise when virtual avatars and AI technologies are combined. By adding a human appearance to a conversational computer program, financial institutions are empowered not only to address the need for round-the-clock customer support, but also overcome what is widely seen as the greatest limitation of conventional chatbot-style digital assistants — the absence of a ‘face-to-face’ experience.
According to research from the Harvard Business Review, “the most effective way to maximize customer value is to move beyond mere customer satisfaction and connect with customers at an emotional level”. Achieving this sort of connection with a text-based platform is almost impossible, no matter how intuitively it can respond. With a “digital human”, however, the essential elements of a deeper and more personal bond — body language, facial expressions, and other nonverbal cues — are all within reach.
Whether it’s conveying warmth with a smile or confusion with a furrowed brow, AI-infused digital humans can speak with real people in a remarkably “real” way. In a banking context, this promises to be a game-changer for making online services more accessible and cost effective, allowing institutions to split the difference between scalable, low-cost digital channels and highly engaging human customer service.
AI’s quantum leap
Like almost all forms of technology, digital humans are an evolution, building on the capabilities of an earlier, more rudimentary system. In this case, the humble chatbot.
Simple, text- or graphics-based advisors have a lot of business benefits. They’re scalable, inexpensive, and serve as a functional frontline for client queries. Customer service bots are adept at narrowing down initial enquiries, while robo-advisors can ask introductory questions on key topics, like risk tolerance, time horizons, and investment values. Based on the customer’s answers, they can make uncomplicated suggestions, such as which portfolio to pick from a pre-compiled set. In recent years, the rising popularity of direct indexing and fractional shares, combined with advancements in computing power, has enhanced the capabilities of chatbots in providing more tailored market advice — but fundamentally, they are still simply sophisticated spreadsheets selecting relevant responses from a predetermined list.
The quantum leap we’re currently witnessing in Generative AI is changing that. Large language models (LLMs) like ChatGPT and Google Bard are trained on massive datasets comprising billions of words — a meticulous process that enables them to form natural responses to user prompts, going beyond pre-set selections and comprehending contextual cues. The result is rather remarkable (particularly for first-time users): a fluid back-and-forth conversation with a computer.
LLMs still need to be trained on data, however, so their knowledge is limited to the information they’ve been exposed to. Put simply, AIs can’t yet generate entirely new ideas of their own.
What does this mean in the context of finance and banking? Currently, AI-powered digital advisors and customer service agents can engage in structured conversations with clients, providing general advice on market conditions and sharing the latest thought leadership articles and marketing materials. However, without access to specific customer data, it cannot address individual circumstances. If someone has a more intricate financial query — one pertaining to complex mortgage arrangement, or exposure to certain industries and sectors, for instance — a digital human will be able to gather far more information that a basic chatbot, but ultimately, lacking bespoke knowledge of the specific issue, it will need to transfer the case to a living breathing colleague.
A huge undertaking
This hybrid computer-and-human scenario is no bad thing, offering financial clients the convenience of instant digital advice backed by the input of an experienced member of staff. But building an effective AI virtual advisor or customer service agent requires more than just pairing an existing LLM with a 3D avatar.
ChatGPT-style platforms are generic, knowing no more about financial matters than a standard internet search engine. To create a truly revolutionary digital consultant — one that can match the performance of human counterparts — it would have to be built on the entirety of an organization’s digital archive, including all client data.
Getting this right requires a rigorous, multi-layered strategy, starting with the cleaning and centralization of data. With digital information consolidated and readily accessible, an open-source AI can be trained to incorporate mission-critical specifics.This is harder than it sounds, demanding a colossal input of data science expertise, computing power, and financial resources. Take JPMorgan Chase’s recently revealed ‘IndexGPT’ project, which is being worked on by some 1,500 data scientists. And then there is ‘BloombergGPT’. The market data giant’s recently revealed finance-focused LLM has been trained on more than 700 billion tokens, or word fragments, and cost an estimated $10 million to complete.
Facing such prohibitively high costs, AI-curious financial firms of more modest means should start by experimenting with freely available LLMs. They’re generic, true, but even with a relatively small amount of in-house content incorporated, they can bring a serviceable virtual assistant to life.
The secret to long-term success
We’re at the foothills of a massive revolution in the banking and finance world. The era of chatbots is fading, making way for the rise of digital humans.
How institutions approach this upheaval will have far-reaching consequences for their future success — fall behind the technological bandwagon, history teaches us, and catching up can be incredibly difficult. As clients rapidly acclimate to the deep engagement provided by virtual avatars and the convenience of round-the-clock availability, organizations that fail to embrace innovation risk fading into obscurity.
That’s why, whether building a bespoke model or exploring what’s already on offer, financial institutions should be positive — though not foolhardy — in their approach to AI. By doing so, they can help shape an industry that revolves even more around the customer, empowering individual and corporate clients with personalized services, and securing their own long-term success in the process.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.