Your AI strategy is really just your data strategy. Christian J. Ward, CDO and EVP at Yext, has built his career on that principle, from NLP-driven signal analysis on Wall Street to synchronizing brand data across hundreds of endpoints. Training your own models is not where he would be spending his time.

Christian walks Saket through how opening Yext's data via MCP servers transformed client interactions overnight, why he tracks token usage by department and model, and how correlating weather data with same-store sales identified patterns across 10,000 locations.

Topics discussed:

• Your AI strategy is your data strategy
• Using knowledge graphs to boost AI visibility and accuracy
• Opening enterprise data via MCP servers to clients
• Inverse prompting and the tyranny of the blank box
• Treating tokens as currency through a token exchange model
• Protecting derivative data sets in data partnerships
• Tracking token usage by department as a CDO metric
• Building organizational consensus to drive AI adoption

Terry Miller, Vice President AI & Machine Learning at Omada Health, calls frontier models "very expensive commodities." With 14 years of longitudinal data for over a million members, the real competitive advantage lives in proprietary data and the workflows built on top of it.

Terry walks Saket through how reusable templates and tightly bounded applications make agent deployments succeed, why combining traditional machine learning with LLMs unlocks new capabilities, and what "edge healthcare" means for decentralized care from the home.

Topics discussed:

• Frontier models as commodities, data as the real moat

• Building reusable agent templates for production systems

• Tightly bounded applications as the path to agent success

• Edge healthcare and decentralizing care to the home

• Combining traditional ML with LLMs for new capabilities

• Preventing the next wave of high-cost patients

• Deep personalization as table stakes for enterprise AI

• Policy gaps around autonomous AI in regulated industries

For Anusha Dandapani, Chief Data & AI at UNICC, what keeps her up isn't model accuracy. It's the asymmetry of consequences. A wrong commercial AI call costs money, but in a humanitarian context, it can mean aid never reaches the people who need it.

She tells Saket why she'd rather run a less sophisticated model with rigorous decision architecture than drop in a state-of-the-art system, and why nobody wants to talk about data lineage until it's too late.

Topics discussed:

• Weighing asymmetry of consequences in humanitarian AI decisions
• Avoiding grand unified architecture in favor of minimum interoperability layers
• Cost-sharing a shared AI platform across ten UN organizations
• Baking responsible AI into design instead of retrofitting later
• Treating data lineage as the real foundation for AI trust
• Drawing the line between reversible and irreversible AI decisions
• Prioritizing decision architecture over model sophistication
• Reframing data quality gaps as risk decisions for leadership

Most enterprises know AI has value. Getting it into the financials is a different problem entirely. Asif Mujahid has spent over a decade in healthcare data and AI, and his read is direct: the technology problem is largely solved. What remains is trust, governance, and accountability, and most organizations are still working out how to get there.

Asif breaks down why centralizing data and AI ownership is non-negotiable, how he frames the CDO role across three distinct archetypes, and what it actually takes to move AI from value identification to value realization in a regulated, high-stakes environment.

Topics discussed:

• Why AI is now a risk governance problem, not a technology problem

• The three pillars required to build stakeholder trust in AI models

• Healthcare AI use cases across administrative, supply chain, and care management

• How generative AI expands volume, velocity, and variety of patient data

• The privacy and bias risks that derail AI programs before launch

• Why CDO success depends on centralizing AI ownership end to end

• The three CDO archetypes and how their importance is shifting

• Moving from "actionable insight" to joint ownership of business outcomes

• Bridging academic models and real-world operational constraints

Most data leaders chase distributed infrastructure they don't need, paying what Mahesh Mishra calls a complexity tax. As VP of Artificial Intelligence at Cloudera, he argues most enterprise data isn't big data at all, and platform readiness, more than model quality, is the real barrier to production AI.

Mahesh breaks down knowledge engineering as the missing layer beyond prompt and context engineering, explains why traditional data catalogs are becoming obsolete, and maps out how Iceberg and Lance are reshaping the lake house for agentic workloads.

Topics discussed:

- Distinguishing knowledge engineering from prompt and context engineering

- Building decision flow and decision tracing languages for agents

- Why data catalogs must evolve into semantic knowledge graphs

- Combining Iceberg and Lance for structured and unstructured data

- Three reasons AI projects stall before reaching production

- Treating AI agents as identity-bound team members for governance

- Asking whether your data actually qualifies as big data

- Forecasting agent-driven decision making and falling engineering costs

Justin Heller spent nearly 11 years building an enterprise data management program from the ground up, most recently as former SVP and Chief Data Officer at Synchrony Financial. He joins Saket to unpack the discipline shift that most organizations miss: structured and unstructured data have to be managed differently.

The rise of generative AI has exposed a relevance gap, the piece most data quality programs were never built to measure, tied to the absence of a records management discipline capable of categorizing, labeling, and disposing of files at scale.

Topics discussed:

• Building a data management operating model from the ground up
• Applying industry frameworks to establish a common data management language
• Detecting versus preventing bad data through quality controls
• Managing metadata across structured and unstructured repositories
• Redefining ROI as relevance of information in the age of generative AI
• Why records management is having a resurgence because of AI
• Avoiding a police-state approach to data management
• MVP-first approach to scaling AI foundations enterprise-wide

What happens when a company built on 16 acquisitions tries to run on one trusted data model? At Precisely, Dave Shuman has spent six years building the data infrastructure needed to take the business from a $300M mid-market firm to a billion-dollar global software company, and the lessons from that journey cut straight to the heart of what makes or breaks an enterprise AI strategy.

Dave outlines for Saket why most AI initiatives fail before they start: organizations skip the unglamorous work of cataloging, semantic alignment, and quality observability in a rush to get to the model. He breaks down how Precisely is building the semantic layer that lets agents operate autonomously, why data governance needs to feel like a supple leather glove rather than an iron fist, and what the CDO role has to become if it wants to stay relevant in an AI-first organization.

Topics discussed:

• Why most transformative AI starts with a data catalog, not code

• Building a semantic layer so agents can operate autonomously

• Managing AI model overconfidence in production rollouts

• Distinguishing active governance from passive governance in agentic workflows

• Structuring data products across raw, component, and trusted zones

• Integrating unstructured documents into structured data pipelines

• What the CDO role must become in an AI-first organization

• Lessons from scaling through acquisitions and closing on one set of systems

What happens when half the room at a Gartner executive workshop raises their hand to say they've shipped AI to production and tracked ROI? Conor Jensen, Global Field CDO at Dataiku, uses that moment to reframe the entire "AI is failing" narrative and get specific about what separates the companies making it work from the ones still stuck in prototype mode.

Conor walks Saket through the compounding mistakes he sees across enterprise AI programs, from skipping legal and governance early, to misreading which problems data can actually solve, to deploying tools company-wide before proving a single use case. The conversation covers data products, agent management, the limits of code generation tools for data teams, and why the "citizen data scientist" framing has always been slightly wrong.

Topics discussed:

- Why close to half of executives at a recent Gartner workshop reported tracked AI ROI

- Engaging legal and compliance early as a speed accelerator, not a blocker

- Building a use case prioritization process as a core organizational capability

- Treating AI outputs as data products that require ongoing ownership and maintenance

- Semantic and context layers as the foundation for AI-ready data products

- Managing agents deployed out of the box in enterprise platforms versus custom-built agents

- Why code generation tools are less effective for data engineering than software development

- The meteorologist reframe: domain experts gaining new tools, not becoming data scientists

- Where 50 to 60 percent of data team backlogs can be self-served by the business

- Why predictive analytics projects hit dead ends in non-tech companies with limited data volume

Most enterprises are racing to build AI labs while the real problem sits in their data layer. Santiago Guillotti, Chief Data and Analytics Officer at Chubb LATAM, leads a 130-person data organization and his conclusion is direct: no vendor can fix your data problems for you. AI platforms are replaceable. The data you generate is not.

Santiago outlines for Saket how GenBI is creating outsized efficiency gains in mature companies where 35 to 40% of FTEs are still tied to BI-related tasks, and where cloud migration is only the beginning of a much larger transformation.

Topics discussed:

• GenBI as an efficiency lever in mature enterprise organizations

• Cloud migration from on-prem legacy systems to Azure and Databricks

• Building data teams differently in startups vs. large regulated companies

• Hiring for cultural fit across Latin America, Asia, and the US

• Scaling from 10-person to 100-person data organizations

• AI's impact on data engineering roles and intern programs

• Why data, not AI platforms, is the real competitive advantage

• Three pillars of AI transformation in the insurance sector

• Automating end-to-end claims flows with conversational AI

• Getting POCs approved in slow-moving regulated industries

What happens when data breaks in production and no one sees it coming? At Recce, Dori Wilson is building the systems to make data reviews as systematic as software code reviews, and the gap between those two worlds is bigger than most teams realize.

Dori Wilson, Head of Data at Recce, walks Saket through how data validation has historically lagged behind software engineering, why dashboards are becoming obsolete, and what it actually takes to build stakeholder trust when the numbers tell a story no one wants to hear. From her time at Uber, Mux, CircleCI, and a YC-backed startup she co-founded, Dori brings a practitioner's lens to every layer of the data stack.

Topics discussed:

• Comparing data validation gaps to software DevOps and CI/CD maturity

• How Recce's AI data review agent surfaces lineage impact before merging code

• Using human-in-the-loop review alongside automated distribution checks

• Why dashboards are becoming obsolete and what replaces them

• Segmenting feature adoption data to avoid selection bias in retention analysis

• Building stakeholder trust through anticipatory storytelling and appendix slides

• Using skills and contextual memory layers to improve AI agent performance

• Career advice for data professionals: abundance mindset and not self-limiting