Chun Schiros PhD, SVP, Head of Enterprise Data Science Group at Regions Bank, shares how a focus on trustworthiness is helping her team drive value with AI at the banking giant
Banks have invested heavily in projects to integrate AI with their core business processes in recent years. But in light of scrutiny from consumers and regulators alike, financial services sector executives are increasingly focused on ensuring AI systems are fair, trustworthy and explainable.
As SVP, Head of Enterprise Data Science Group at Alabama-based banking giant Regions Bank, Chun Schiros PhD is a key player in this movement. She joined the bank as a Model Validation Analyst in 2014 and is now responsible for its AI strategy roadmap.
In this week’s Business of Data podcast episode, Dr Schiros outlines how she’s putting ‘trustworthy AI’ front and center as she deploys the technology to provide superior customer experiences and drive business value across the bank.
Optimizing Commercial Banking Processes with AI
Under Dr Schiros’ leadership, Regions Bank is implementing an AI-driven analytics platform for its commercial banking arm. She says the platform will provide capabilities including early warning indicators for customer attrition, liquidity alerts and credit risk alerts.
“This tool really gives them a leg up,” she says. “We have a much more streamlined process for identity to find new opportunities and close them faster. Now, we are on version four, where we are adding a new pricing analytics model.”
This platform also gives the bank a more holistic view of its customers. In turn, this is enabling Regions Bank to apply predictive models to common customer pain points, allowing it to provide better customer experiences and more personalized services.
Dr Schiros explains: “Not every client likes going to the branch, for example. Young people are used to digital banking. So, how can we optimize their channel preference to provide the right services, through the right channel, at the right time?
“We provide real-time predictive analytics and can use this for problem resolutions, so that we understand where customer pain points are before the customer even contacts us.”
Prioritizing Trustworthy AI at Regions Bank
AI systems must be accurate and operate in ways that are easy to understand if executives expect customers and company employees to trust and use them. For these reasons, Dr Schiros believes in an open, transparent methodology and an end-to-end business value approach when delivering projects.
“We started right from the beginning, taking the user along with us when designing the product,” she explains. “We start with an MVP [minimum viable product] with minimal features, so that we can iterate the product throughout its cycle and use the feedback we get to improve the product. This way, we can really understand the users’ needs and create ownership for the user.”
“When identifying business problems, we start by understanding what needs to be solved and how we can solve it through data science,” she says. “We also consider the type of data we’ll need to best solve the problem. We manage the quality of the solutions through the entire machine learning lifecycle from target definition to specific use cases, data prep, model evaluation to post-production model monitoring.”
“We measure the quality of the solutions through data quality, conceptual soundness, model stability and model accuracy,” she adds. “There’s also post-production model performance, such as accuracy drift.”
Managing the end-to-end lifecycle of AI models in this way requires enterprises to develop mature functions for AI governance, model development and model operationalization.
It’s a huge undertaking, and one that many enterprises have yet to complete. But as Dr Schiros and her colleagues say, it’s also essential for building trust in the business decisions AI models recommend.
- Trustworthy AI is key. Driving value with AI depends on promoting the adoption of AI systems, which in turn depends on ensuring stakeholders trust them
- Incorporate user feedback into AI development processes. An effective business transformation relies on collaboration and encouraging a sense of data ownership within business units
- Adopt an agile approach. Start with an MVP and iterate AI systems based on user feedback to deliver value incrementally over time