Nan Li, AI leader and former Nationwide VP, AI/ML and Statistical Practice, speaks with Gavroshe Founder Derek Strauss, in a video interview about the role of human-centric AI, the importance of aligning models with business objectives, addressing data quality challenges, promoting AI literacy, making trade-off decisions in model development, and the need for collaboration to drive meaningful solutions.

From being a biologist to getting an MBA, Li started her career in database marketing at a time when analytics was not even a popular word. Eventually, when data analytics had just taken off, she pivoted to marketing analytics at Nationwide.

Then, Li led the solution and analytics team at Cardinal Health, where they pioneered cloud-based big data analytics solutions, uncovering sales opportunities across medical product lines. Post that role, she wore multiple hats at Bread Financial, established new data analytics functions, and implemented enterprise-wide platforms and new practices that brought in million-dollar benefits in digital marketing and operations.

In her next role, Li led the AI/machine learning and statistical practice in the enterprise analytics office at Nationwide, championing the vision, strategy, execution, and governance of advanced analytics. She also implemented scalable AI and machine learning solutions, including generative AI (GenAI), to enhance enterprise operations by automating and optimizing processes throughout the insurance value chain.

Reflecting on her trajectory, Li states, “I feel very fortunate about my journey. But right now I want to spend time taking the learnings from my experience and translating them into clean frameworks and developing my own leadership in the human-centric AI approach.”

With so much buzz around technology and AI, she stresses that it is critical to remember that the idea to work on AI was germinated to improve business, society, and life in general.

Sharing key lessons that shaped her notion around human-centric AI, Li says that there are three key components of a successful solution: Data, Models, and Technology.

She notes that while data is the right ingredient for an intelligent model that requires technology, in essence, all of them depend on decisions made by humans.

Delving deeper, Li says, “Data comes from human behavior and processes designed by humans to collect the data.” She recalls an instance when she was working on a data governance project for retirement plans at Nationwide Insurance and encountered challenges in standardizing and simplifying definitions.

Next, Li shares how one conversation with her middle school daughter led to an epiphany that it is critical to work with the people and process to fix the data quality issue. “Data quality is not a technical problem. It’s not data people’s problem. It represents the business quality,” she adds.

The first lesson, according to Li, is that data is related to people, and the focus must be on people and processes.

Speaking of models, she shares how often leaders and business partners ask her to build models to solve business problems. Breaking the assumption about the simplicity of the task, she explains the intrinsic nature of a model. 

One key aspect is that every AI model requires a clearly defined objective, which must be determined by humans. Additionally, models operate based on assumptions that need human validation, and all these model calculations inherently come with errors.

These errors cannot be completely eliminated, says Li; “…it’s all about making trade-off decisions…” deciding which aspects to minimize and which to optimize. A lack of understanding in this case can create gaps in expectations when collaborating with business partners, she adds.

Recognizing this, Li mentions adjusting the approach to engage with business partners by actively involving them at every stage. Before a project begins, they are educated on AI capabilities, complexities, and limitations, ensuring they are aligned with the process.

Li further adds that the business partners are presented with trade-offs, such as deciding whether to minimize errors at the cost of increased risk or maximize profitability while potentially reducing retention rates.

This approach has fostered strong collaboration, in-depth discussions, and mutual understanding, enabling effective communication in a shared language, she maintains.

This experience also highlights the need for AI literacy, says LI. “It’s not about building the best models. It’s about aligning the model with business objectives to create the values and the collaborative approach that can help you breach the expectation gap and lead to long-term success,” she affirms.

When it comes to technology, Li mentions evaluating several models and selecting the right one. However, in the GenAI space, with new models coming up every other day, it is a different scenario.

Shedding light on the newer models like Mistral, Llama, and DeepSeek, she notes, “You will never be able to finish your project. If you keep chasing the latest.” Li urges data leaders to stop chasing at a certain point, take the model that works best for that use case, and take it to the finish line.

AI models are never truly finished, says Li, as they evolve through continuous iteration. So rather than endlessly refining, she launches MVPs with key stakeholders and improve them over time.

Concluding, she says, “We focus, not on the technology. We focus on people, what they need most, and we solve for them and what our people are capable of doing.” Instead of chasing an ideal, Li confirms making the most of the resources at hand to provide the best possible value.

CDO Magazine appreciates Nan Li for sharing her insights with our global community.