Originally published in Eurofi magazine.

When ChatGPT launched last November, generative AI quickly became of huge public interest. Already, 62% of people across nine nations say they have experimented with generative AI
tools in the last three months, according to survey data from the Oliver Wyman Forum.

The excitement is warranted. Generative AI holds the potential to help financial services firms re-imagine their entire businesses around the needs and desires of their customers.

There are at least four areas where generative AI can help firms dramatically improve their operations. In customer service, conversational AI assistants can understand and speak
natural language, allowing firms to create customized mass outreach to customers. In marketing and communications, AI can design visual product and brand content for logos and packaging, create website layouts, and write blog posts, articles and social media posts. In tech and IT, AI can help with code generation, reducing the time and resources need for software development, and create synthetic data to train machine learning models or test applications. And in terms of personal productivity, generative AI can create automated notes and summaries of meetings, help people manage priorities and tasks, and assist with scheduling.

But first, firms must clear three short-term hurdles that make it difficult to embed the technology today. First, generative AI poses unique risks that traditional AI systems don’t have to contend with. One is defamation: programs inadvertently producing defamatory content. Another is hallucinations and opaque logic and processing. Generative AI also creates confidentiality concerns such as data leakage and copyright issues. To address these, financial institutions need to beef up their governance, data quality, talent functions and other dimensions.

The second hurdle: regulators. The potential perils of AI span the enterprise, including operations, technology, legal, compliance, process, data, technology and reputational risks. Banks need an enterprise-wide framework to holistically manage these risks. Government bodies have offered guidelines on best practices in the US, the EU, the UK and Hong Kong. When these guidelines turn into hard rules, we estimate fines for lack of governance could approach 6% of the industry’s global revenue.

Another hurdle is the technology itself. Today’s models lack desires and self-directed learning. They have extensive knowledge of the world, but don’t “know what they know,” and lack any sense of truth. As for reasoning, models’ abilities remain brittle and likely to fail unexpectedly, especially when asked to apply logic and knowledge in new contexts. And they don’t yet offer predictability, with unwanted outputs creeping into models frequently.

As a result, significant productivity improvements from generative AI will take time. Learning curves are steep, there is still insufficient scale of adoption, and model tweaks and redesigns have been slow. Firms must navigate a collision course: In one direction, ongoing advancements are likely to drive more widespread use as tools become integrated into our daily lives, much like the iPhone. In the other direction, regulatory bans could lead to unsupervised and unsafe AI tool usage, and the potential for employee misuse needs to be managed. Past technologies have overcome such obstacles. It took e-commerce 20 years to reach 10% of retail sales. The personal computer took 10 years to get to 42% usage across US households. Electrification took 40 years to deliver measurable productivity gains across the UK. There is good reason to believe generative AI one day will be ubiquitous in the financial
services sector and throughout society.

Pathway to generative AI in financial services 

In the short term, firms should encourage safe engagement through training and limiting access to safe use cases. Over the longer term, companies will be able to target AI capabilities
toward key business pain points, invest in technical training and tailored upskilling to improve usage. Eventually generative AI will drive large-scale organizational transformation, helping
companies understand inter-department workflows to integrate technologies across the enterprise and adjust people and processes as needed.

The ultimate possibilities lie in AI’s ability to help institutions reorganize entire business units based on the wants and needs of their customers. There is little doubt that customerfirst platforms, powered by AI and offering rich, flexible user experiences, are the industry’s future. The race is on to get there first.

This contribution has been co-written by Sian Townson, David Waller, and John Lester.

Find the original piece, here.