It would not be an understatement to say the banking industry is in the middle of a perfect storm.  On the one hand, digital disruption is forcing banks to reimagine newer business models – whether it is related to cashless payments, greater access to global markets, elevated service expectations from the customer, financial inclusion goals, lowering the cost of products and services or even mitigating risk. 

On the other hand, there is still the bugbear of antiquated systems, technical debt accumulated over years, changing regulatory requirements related to capital adequacy norms, anti-money-laundering, data privacy. 

Amidst this disruption and dynamic situation, banks are also co-opting fintech partners and external ecosystems rather than depending on internal platforms alone.

In this backdrop, the Generative AI wave, with the spree of announcements from big tech (Google, Microsoft and AWS), is beginning to intrigue the banking industry enough for them to have diverted their IT spend towards a range of exploratory projects and have even taken some of these into production.  Gen AI is bringing in the promise of a much-enhanced customer experience, introducing greater agility and lowering the cost of internal operations and revenue maximization through cross-sell, up-sell opportunities.

Some of the top use cases that are coming to the fore are around ‘virtual assistant’ for the frontline staff and personalized marketing content. Gen AI based tools can quickly retrieve information on relevant products or services or insights on the customer’s financial health, transaction patterns and such. LLMs and other foundation models can improve customer click-through rates with generated, personalized content that is human-like, closer to the customer’s real-life context, adjusting for tone, empathy and supporting multiple languages.  AI-based financial advisories are another use case where multimodal Gen AI models can be applied to engage with customers through images, videos, text, recommend the next best action, provide personalized guidance to achieve financial goals. However, these solutions would need to mature through active learning, human verification and training to ensure greater accuracy of the output.

Synthetic data creation through GANs (Generative Adversarial Networks) is a use case that has been in vogue for some time now specially to enhance accuracy of credit scoring models or support credit decisioning for customer segments where data sets are sparse or even in cases where data privacy or sovereignty has to be preserved for meeting regulatory requirements.  

India today accounts for nearly 46 per cent of the world's digital transactions (as per 2022 data). The digital payments market of India is expected to grow at a CAGR of 50% and exceed 400 billion transactions in FY2026–27.  Payment exception processing becomes a very relevant use case for application of both traditional AI and Gen AI in this context.  As payment transactions get declined, observability around what part of the workflow or what part of the digital stack caused the transaction to fail will become critically important.  Banks know that they cannot afford slip-ups - failed or slow payments, inability to handle sudden spikes or instances of fraud can have a hugely damaging impact on customer experience, compliance, brand reputation and revenue.

Most organizations are at a juncture where the value and impact of Gen AI is clearly understood.  There are some concerns that are clearly emerging though as organizations look to adopt Gen AI at scale within the enterprise. Some obviously are around data leakage or inadvertently exposing enterprise private data to the LLMs.  Other concerns are around risks and costs incurred of running Gen AI applications being commensurate with the outcomes. Uncontrolled use of Gen AI applications and shadow IT becomes an immediate challenge. There are concerns around whether the Gen AI applications are kosher to use from a perspective of meeting any audit requirements such as traceability, accuracy of the outputs generated, accountability in terms of human involvement or feedback to the process.  In many organizations, lack of adequate talent to drive such disruptive innovation sustainably is a major hindrance.

It's also important to understand that not all use cases have to be looked at from a Gen AI lens.  Many use case families – prediction, anomaly detection, planning and optimization, decision intelligence, autonomous systems – can be much better addressed by traditional AI techniques. The hype surrounding Gen AI can lead IT leaders to apply it where it is not a good fit, increasing the risk of higher complexity, diminishing business value and failure in their AI projects.

Given all of the above, a comprehensive LLMOps framework is essential to overcome these governance challenges, drive adoption at scale and drive to a better time-to-market and time-to-value with these Gen AI infused solutions.  Besides, of course, the human capital that is needed to lead wisely in such a transformation. As always, a significant part of driving any innovation agenda is having the right level of sponsorship, thought leadership and the ability to take on big bets and build breakthrough solutions, having the right mix of tooling and talent to back it all up.  As the technology and ground rules change rapidly, it is wise to take a co-creation approach both with key stakeholders within the enterprise and co-opting the ecosystem of service providers and technology providers that can reliably deliver outcomes.   As a famous proverb goes, “If you want to go quickly, go alone. If you want to go far, go together”.