Bank of America and S&P Global on why AI success starts with governance and data

Good morning. In highly regulated industries, the focus is not on whether companies can deploy AI technology, but whether they can trust it to make decisions that impact customers, markets, and the economy at large. At Fortune's inaugural AIQ Summit, leaders from Bank of America and S&P Global emphasized that AI adoption should be based on a strong foundation of governance, data, and accuracy, rather than simply attaching the technology to every process.

Sally Moore, Chief Client Officer and Co-Head of Market Intelligence at S&P Global, highlighted that data is the bedrock of their approach. By collaborating with a major bank, S&P Global demonstrated how combining proprietary intelligence, subject-matter expertise, and implementation support can significantly enhance operational efficiency. The partnership led to a remarkable reduction in the bank's time to market and improved accuracy in work involving multiple data sources, from 60% to 98%.

However, Moore stressed that accuracy alone is insufficient. It is equally crucial to ensure that the underlying intellectual property informing AI decisions can be traced back, especially in high-stakes applications. Hari Gopalkrishnan, Bank of America's Chief Technology and Information Officer, explained that the company evaluates AI implementations across 16 risk dimensions, such as privacy, bias, workforce implications, and intellectual property.

The first step is not the model itself but understanding the client's need and determining if AI can genuinely address the problem. Gopalkrishnan cautioned against rushing to adopt AI when simpler deterministic models could achieve the same results.

He emphasized that the most costly mistake may not be moving too slowly but rather investing in complex AI solutions where simpler technologies would provide a more reliable and lower-risk outcome. Once AI is deemed appropriate, continuous governance is essential. This includes setting guardrails, assessing outcomes, and monitoring systems for unintended consequences.

For instance, a fraud model that disproportionately affects certain demographic groups or a chatbot that fails to serve customers with specific accents could lead to customer, compliance, and reputational risks—not just technical issues.

In summary, successful AI implementation hinges on a disciplined approach grounded in governance, high-quality data, and unwavering accuracy.