Marriott International is one of the world’s largest hospitality companies, operating a global portfolio of hotel brands and a loyalty ecosystem built around delivering consistent, personalized guest experiences. In that environment, enterprise data readiness and AI-ready data foundations are no longer optional; they determine whether personalization, governance, and real-time decisioning actually scale.

The first part of this four-part interview series discussed how enterprise data leadership has shifted from stewardship to measurable value creation. The conversation explores how modern data leaders translate technical complexity into business outcomes tied directly to loyalty, revenue, and customer experience.

In this second installment, Nitin Kumar, Director of Data Science and GenAI at Marriott International, sits down with Ben Blanquera, VP of AI and Sustainability at Rackspace Technology and a CDO Magazine Editorial Board Member, to unpack the industry shifts shaping enterprise AI: why governance becomes an engineering reality, how democratization works with guardrails, and where agentic solutions are headed.

Three enterprise AI shifts shaping data readiness

Kumar says multiple transformations are unfolding at once, but at uneven speeds. “Three simultaneous shifts are happening at a different pace.” He frames the moment as a convergence of three forces: engineered governance, democratized access with safety controls, and the emergence of agentic solutions.

From model-centric ai to data-centric thinking

Kumar observes a major reset in how organizations think about AI. “People are moving away from a model-centric to a more platform and data-centric thinking,” he says, pointing out that the biggest constraint is rarely the model itself.

He explains why: “The models are rarely the bottleneck.” Instead, he says, the bottleneck is whether “your data is  trustworthy, accessible, governed, and usable in real time.”

GenAI makes the stakes harder to ignore: “Think about any rack solution that you have, and if the underlying data is incorrect, you’re sharing incorrect information with your customers.”

Responsible AI moves from policy to engineering reality

Kumar describes another shift of how responsible AI stops being a policy statement and starts becoming a build requirement. He characterizes it as “responsible AI moving from just a policy statement to an engineering reality,” including “the right guardrails, right monitoring, automated all processes, and human insight or oversight.”

At the same time, he says the “cool demo” era is fading. “The shift is about whether that model is actually able to improve the ROI or not,” he says. “Because if there is no moving of the needles for the ROI, then that demo may not be very useful.”

Why data readiness isn’t a checklist

When asked how Marriott approaches data readiness, Kumar calls it foundational: “It is the most fundamental problem every company has to focus on and be able to solve for.” And he rejects the idea that readiness is a simple scorecard: “Data readiness isn’t about a checklist or a quality score.”

Instead, he frames it as a strategic posture: “It’s a robust, strategic state that puts your company on a growth mindset rather than a segment or a fixed mindset.”

He says Marriott anchors readiness around three principles:

• Architecture
• Thinking of data as a product
• Proactive governance

Kumar explains what “data as a product” solves: teams repeating the same wrangling steps for every new use case. The alternative is shared, reusable products that teams can just plug into.

The customer 360 that combines structured and unstructured

To make the concept concrete, Kumar walks through what a unified customer view could look like at enterprise scale, bringing transactional data and unstructured case history together.

Here’s a minimally edited version with clarity and grammar tightened, while keeping the original voice and structure intact:

He offers an example: “Let’s say if I build a Customer 360, which takes transactional data, as well as your personal data, and then your unstructured case data, and tie it together to provide a unified view.”

With that in place, he says teams can answer both structured and narrative questions quickly: “With a simple query, you can actually get structured information like how many stays you had or which property you stayed at,” while also being able to trace service context: “Like what problems you had in the past, so that we can rectify or double-check when you are checking in the next time.”

The goal is a semantic layer that makes data easier to use and more actionable: “It will give you a unified platform to actually query your data more easily, and you’ll be able to answer or cater to your customers better.”

Democratization with guardrails

Kumar argues the next wave depends on enabling the broader workforce, not bottlenecking every use case. He believes the near future centers on broad enablement: “Empower your workforce to be able to use GenAI or AI to solve their business problems.”

At Marriott, he says that empowerment starts with structured rollout and tiered tooling. Marriott empowers employees to use AI by starting with familiar copilots, then offering a tiered toolkit, from no-code options to fully coded builds, so teams can choose solutions based on problem complexity and user skill level. That approach supports everyday workflows like email summarization and content generation, including no-code marketing content creation through tools like Writer.

Training that scales fluency

Kumar describes enablement as an ongoing cadence, not a one-time workshop, especially because GenAI is new for everyone. He also stresses the risk of getting the education wrong at scale: “Because if we teach you something incorrectly, the impact would be huge because you’re teaching a lot of people on that.”

Further, Kumar says the next wave of enterprise AI will move from large, generic LLMs to smaller, highly fine-tuned models built for precision on a company’s own data. Beyond that, he expects agentic systems, AI that can reason, remember, use tools, and take actions, to absorb repetitive work as “digital twins,” escalating to humans only when help is needed. He also notes that multimodal AI (vision + text + voice) will unlock use cases that weren’t previously solvable.

Marriott invests in practical training and safety behaviors, including:

• Prompt education
• Sensitive data awareness
• A multiplier model
• Filtering use cases

The point is to keep the AI team from becoming the choke point. “Otherwise, in a traditional IT shop, there is no way a handful of people can solve all the business problems without becoming the bottleneck.”

Kumar, however, concludes with the condition that makes democratization safe: “Anyone can do it, with proper guardrails. Don’t forget that.”

CDO Magazine appreciates Nitin Kumar for sharing his insights with our global community.