Cindy (Xin) Tu, Director of IT and Data Audit at Discover Financial Services, speaks with Mudit Gupta, Partner, FS Consulting AI Practice Leader at EY, in a video interview about balancing innovation and risk management in financial services, the importance of stakeholder engagement, building a culture of digital fluency, and enterprise-wide data and AI training.

GenAI in financial services — Balancing innovation and risk

Tu begins by sharing her insights on how financial organizations can successfully navigate both priorities of innovation and risk management, especially as they incorporate generative AI (GenAI) technologies.

Speaking of balancing robust risk management with the need for technological innovation, she says, “Without innovation, we could become obsolete pretty soon.”

One of her key recommendations is fostering strong trust and collaboration with stakeholders across the organization. Tu emphasizes that second-line and third-line partners, including legal, compliance, and IT, should not be afterthoughts but integral parts of the innovation journey.

She advocates for involving stakeholders early and consistently throughout the process, particularly when moving from proof-of-concept (POC) stages to production deployment.

“All of these stakeholders have to be involved in the process to assess the overall risk.”

To build confidence in GenAI models, particularly large language models (LLMs), Tu suggests involving stakeholders in monitoring and validating the model outcome. In addition, she mentions demonstrating the impact of GenAI models in increasing productivity, reducing costs, and accelerating innovation.

This way, stakeholders will have transparency on the overall process, have a say, and witness what a model can do from a productivity and outcome perspective. “They can have more trust, and maybe they can be a champion then in the future GenAI use case,” she adds.

Ultimately, Tu envisions pushing these efforts through a governance framework wherein the company balances data security, privacy, and overall risk mitigation while still championing innovation and proposing new GenAI business use cases.

Building a culture of digital fluency: Why enterprise-wide data and AI training is essential

“I am a strong believer in digital fluency in terms of not just AI, but also data. It has to be a training program for everyone in the company,” says Tu.

She emphasizes that data and AI literacy must not be limited to specialized roles like data stewards or data owners. The entire workforce, regardless of job function, needs to have at least a baseline understanding of the data they interact with.

Tu also points out that even roles like auditors, who may not be building data products, handle sensitive and confidential information regularly and should therefore understand data risks just as well. “You can’t just say, ‘We are not the ones producing any data product; therefore, we don’t need data training.’ That’s not the case.”

Core elements of an enterprise training program

Tu believes that a successful data and AI training initiative must be structured and scalable across all levels of the organization. She outlines several key components:

• Awareness of data ownership and classification: Understanding what data is being handled, its classification, and how it should be protected.
• Risk awareness: Training should help individuals grasp the specific data risks relevant to the financial services sector.
• Alignment with strategy: Employees must understand how the data and AI strategy connects to the broader business goals.
• Defined roles and responsibilities: Everyone in the company should be clear on their part in helping the organization succeed in its digital initiatives.

Highlighting that different roles require different levels of training, Tu recommends persona-based learning paths. For instance, she says data producers need detailed training on data quality, governance, and product development. However, auditors or non-technical staff require a foundational understanding of data risk and compliance practices.

Despite these differences, Tu insists the training should be enterprise-wide and ideally integrated into existing compliance platforms to ensure adoption and consistency.

Wrapping up, Tu states that without a well-trained workforce, even the best data or AI strategies are likely to fall short. “Without the right talent in the right place, any of the data programs or AI programs are not going to be successful. We need people to make it happen.”

CDO Magazine appreciates Cindy (Xin) Tu for sharing her insights with our global community.