Generative AI is entering its second phase, and parts of the financial services industry are already being reshaped. But the ultimate benefits of generative AI won’t be distributed evenly.
Phase one was dominated by large language models, which offer assistance based on prompts by users. Phase two is the agentic AI revolution, with new systems capable of planning, deciding and executing multistep workflows autonomously on behalf of users.
Already, more than half of financial services executives report using AI agents, according to a recent Google study based on a survey of more than 3,000 leaders of global enterprises.
Yet in my work advising senior leaders of European financial services firms, I am seeing wide variation in agentic AI readiness and implementation. Despite the flurry of activity, many obstacles are impeding long-term transformation, from infrastructure and operational limitations to trust gaps. Addressing these barriers is vital to accelerating agentic AI’s integration. The firms that get it right — not necessarily those that do it first — will reap the biggest rewards.
Sizing Up the Opportunity
Agentic AI promises to eliminate manual tasks and accelerate decision-making, and firms are eager to get started, with more than 80% of industry respondents expressing willingness to work with AI agents to increase efficiency.
Agentic AI calls upon tools, uses persistent data and takes actions. Besides cost saving, it promises to unlock new revenue opportunities by powering services that can access underserved markets or fill advice gaps.
It can also improve customer satisfaction via autonomous services that meet customers’ changing preferences instantly, raising their comfort level for future agentic interactions.
Three Scenarios in Financial Services
The adoption of agentic AI will give rise to three economic models, each offering banks and fintechs novel opportunities to deliver more relevant, personalised and effective services.
One is what might be termed the Assistance Economy, in which AI agents deliver entire customer experiences and manage multistep workflows to assist in users’ daily activities. For example, agents could look for opportunities to refinance a mortgage and invest the proceeds based on the consumers goals and risk tolerance, handling much of it autonomously.
The second model is Adaptive Customer Experience, where customer journeys are shaped by real-time agentic AI insights, drawn on contextual data such as the specifics of a product search or user behaviour. For example, an adaptive banking app could prioritise showing the quick-pay option on a customer’s banking app right after large purchases.
The third is the Agentic Twins model, featuring digital counterparts of customers that manage personal data and interact with providers on their behalf. The model will empower customers to make faster and more efficient decisions, such as moving savings to the best providers or reallocating portfolio funds, ensuring efficiency, control, value and trust for users. And agents will adapt over time, protecting customers against fraud and new risks.
What’s Getting in the Way of Progress
Infrastructure is maturing and becoming widely available, but there are challenges in getting high-quality data, managing security, getting access to the right capabilities, and hiring the right talent.
Incumbent banks lag behind fintechs in these areas. Their slower pace stems from their legacy architectures, which favour incremental tool layering over dramatic overhauls. And despite their prioritisation of AI in customer facing services, only 32% say they have realised tangible returns from these efforts, according to Oliver Wyman research.
Confidence and trust are also hurdles. According to Oliver Wyman research, 44% of consumers cite privacy concerns and 21% worry about inaccurate data, leading to discomfort with autonomous task delegation. A lack of transparent, ethical and responsible AI guidance has heightened these concerns.
How To Clear the Hurdles
Progressing from experimentation to broad adoption requires robust data pipelines and an effective analytics layer that empowers AI agents to serve customer needs.
Financial institutions also must identify high-impact use cases and then continuously refine systems to optimise value. For example, using AI to resolve customer pain points or to fix dysfunctional internal processes like inefficient fraud detection can lead to improved efficiencies and cost-savings if done thoughtfully.
To promote trust, institutions need to integrate traditional regulated models, open banking and digital ID for secure, transparent data sharing and authentication. Banks in particular have natural advantages in trust when compared with fintechs and others, and they should continue to foster customer confidence through education and upskilling. Industry-wide collaboration is essential.
Firms that surmount these obstacles will reap the greatest rewards during phase two of the generative AI era. Those that don’t might have trouble making it to phase three.