Unleashing the next storm of BFSI transformation with Generative AI

One of the most significant risk factors of GenAI adoption is the risk of biased outcomes. AI models have been found to deny mortgages and issue expensive insurance coverages to people from marginal communities, even in cases where an algorithm was trained to ignore features pertaining to them. However, sophisticated algorithms like GenAI may discover correlations from non-protected features, which can cause bias to creep into the output. This can not only lead to unfair practices but also attract regulatory action.

Proprietary models could be black-box algorithms, which makes bias mitigation even more challenging. Therefore, sophisticated bias mitigation techniques are essential to safely deploy GenAI solutions in the BFSI industry.

As GenAI solutions become readily available, multiple banking organizations may adopt them for use cases like securities trading. This may amplify the market's volatility and cause systemic risks as various organizations execute correlated trading strategies. While regulators are likely to issue statutes to address such risks, organizations must factor such risks into their adoption strategies.

Each financial services organization is different in the customers it caters to, the services it sells, and the operating model it employs. The same set of processes might need additional training datasets and models for different clientele across the same organization to bring about focused personalization. GenAI use cases must be piloted and refined for these deployment scenarios, and models must be fine-tuned to meet the organization's specific needs.

GenAI offers a valuable edge in some use cases – like the ones listed above. However, it is not a quick-fix solution for all problems. For instance, deploying the technology through a chatbot will not magically solve all customer experience issues – especially if the pain points lie elsewhere. It's a long-term strategy with initial effort into training on the exact need and ensuring statutory laws are abided by.

Like all AI algorithms, GenAI is prone to errors, too. While a newly deployed model will be reasonably accurate, its accuracy will climb over time. Nonetheless, it is prudent to employ GenAI and a human-in-the-loop to mitigate the consequences of inaccurate outputs. Some regulations also necessitate such frameworks in the use of AI.

Unlike early hard-coded robotic process automation (RPA) systems, GenAI doesn’t operate like a cold, indifferent algorithm. It may not bring the emotional quotient of a human being, but recent advancements bring emotional intelligence capabilities to GenAI. A leading example is Google’s Gemini 1.0, which can infer the tone, mood, or emotion in voice, text, or video streams.