Can banks improve experiences with GenAI?

With advancing computing power and the availability of vast datasets, Artificial Intelligence (AI) has made remarkable progress over the past decade. Applications based on transformers, a type of neural network architecture optimised for natural language processing, and especially large language models, have captured the public imagination since 2023: This is Generative AI (GenAI).
Today, Generative AI has already been integrated with many aspects of business and is becoming a transformative force reshaping the financial industry.
To expand on these developments, we have collaborated with Citi on the “Delivering Improved Experience Through Generative AI” whitepaper, which explores potential applications of GenAI in global payments, focusing on practical implementations that could improve banking services for clients.
Why are banks looking to GenAI?
The banking sector is a prime candidate for the rapid adoption of GenAI technologies. With vast text-rich data sets alongside a high volume of text-based customer interactions, the banking industry could see the biggest impact as a percentage of their revenues from the technology.
GenAI can draw on banks’ process and product knowledge bases to tailor its communication with clients, providing fast and accurate answers, which might cover anything from payment format setup to payment status information. Further enabled by accelerated document digitisation, banks can offer faster, higher quality and more relevant services, such as onboarding and updating client mandates.
From a regulatory perspective, while GenAI can aid in enhancing various compliance processes in the banking sector, it will be critical for the banks to carefully oversee GenAI capabilities through appropriate safeguards, and factor in the full catalogue of potential risks. Having these processes in order as soon as possible will be critical to the long-term benefits of GenAI and roll out across the sector.
As banks continue to modernise their applications, improvements made possible by GenAI will have a significant impact on speed and consistency. This progress, in turn, will open an opportunity for new product lines with a faster time to market.
The use cases of Generative AI for seamless banking
Use cases include:
• Onboarding – onboarding for larger corporate clients is complicated by a heavy reliance on documents. Once digitised, these documents can be converted into vector databases which allow GenAI models to retrieve only the data specific to a query, ensuring documents remain relevant and up to date.
• Document metadata and context extraction – removing the need for the same documents to be provided multiple times – all relevant documents would be easily searchable because context would be extracted and indexed in the knowledge database.
• Aligning bank offerings with client needs – clients often need guidance to match their needs against a bank’s product offering and find it difficult to ascertain which solutions will offer the greatest value and an improved banking experience. GenAI’s strength is in rapidly analysing vast amounts of non-structured data, finding patterns, and making connections.
• Straight Through Processing (STP) – the elimination of manual touchpoints from banking processes is a key goal to deliver a seamless client experience and achieve near 100% STP. Improvement of such processes involves leveraging advanced analytics and machine learning techniques, including mapping client journeys.
Challenges and considerations for Generative AI implementation
Among the concerns associated with the implementation of GenAI is the matter of bias. If the training data contains bias, then that bias can be replicated and spread at scale. This bias can lead to content being created that reinforces cultural, racist or gender-based stereotypes.
If the model has inherited biases during its training and tuning, its outputs could discriminate against certain groups or lack diversity and inclusion. This presents a risk as an organisation: Using GenAI may generate output at odds with the organisation’s cultural beliefs and policies.
As most suppliers of the Large Language Models that underpin GenAI will not publish details of the data used to train their models, it is important that GenAI applications are thoroughly tested to ensure that outputs are unbiased and accurate.
Other challenges include:
• Data accuracy – if a model is trained on data that contains factual errors or bias, the model may repeat the inaccuracies.
• Data completeness – if the dataset lacks the necessary information to answer a question, the model might generate incorrect responses.
• Input context – if a query is unclear or lacks necessary contextual information, the response is more likely to be unreliable.
To minimise mistakes, domain-specific data used by an organisation to train a model must be accurate, complete, free from bias and not infringe intellectual property rights.
All eyes on the future
Generative AI marks a seismic shift in how banks can engage with corporate clients. By leveraging this technology in feedback mechanisms, financial institutions can move from a reactive to a predictive approach, and from generalisation to customisation. This transformation is key to building enduring client relationships based on co-creation, and delivering exceptional value in the competitive corporate banking landscape.
However, the technology needs to be implemented responsibly. Systems must be thoroughly evaluated to ensure they do not generate inaccurate or unethical responses. The ability for systems to communicate using natural language necessitates more extensive testing than past systems might have required.
To learn more on the potential applications of GenAI in global payments, download the Delivering Improved Experience Through Generative AI whitepaper from Citi and Icon Solutions.