The 16th Data Management Summit New York City immersed delegates in a deep dive into the implications of artificial intelligence and associated agents as the technologies dominate discussions among data chiefs of the world’s financial institutions.
In the opening Data Leaders Panel, The Evolving Data And Ai Operating Model – Synergy, Governance And Accountability, the panel took in the changing role of the chief data officer and the emergence of the chief AI officer, discussed challenges of governance and accountability as well as the importance of data literacy and noted that risk management now must encompass data quality assurance.The panel concluded that four pillars of a modern data management needed to align to scale AI safely and deliver business value.
AI was characterised as the engine, data the fuel, governance the steering wheel and the operating model design the road network.
The panel brought together Yinghua Michelle Zhou, Head of Finance Data and AI at State Street; Andrew Foster, Chief Data Officer at M&T Bank; Amy Kabia, Head of Wholesale Client Data Strategy at Wells Fargo; Richard Chudley, Chief Data Officer for SG Americas at Société Générale; and David Zimmerman, Chief Data & Analytics Officer, Americas Division at SMBC Corporation. It was moderated by Julia Bardmesser, Adjunct Professor at NYU Stern School of Business and Founder and Chief Executive at Data4Real.
Governance and Accountability
Some of those challenges were debated in depth during the next panel session, which was entitled Redefining Governance And Accountability For The AI Era.
The transition from critical data elements to AI-ready datasets took up a large proportion of the discussion, with the panel noting that now “everything is critical” and that governance had shifted towards ensuring all datasets were AI-ready. The challenges of managing unstructured data were also brought into the mix, including the necessity of creating a validation layer to verify extracted data.Moderated by Andrew Delaney, President and Chief Content Officer at A-Team Group, the panel comprised: Brian Greenberg, Senior Director – Data Operating Model Lead and Data Management Practice Product Manager at BNY;
Nirav Shah, Enterprise Data and AI Governance Director at MetLife; Manal Alimari, Executive Director, Chief Data and Analytics Officer Business Partnership and Product Management at SMBC Corporation; Michael Curry, President, Data Modernisation at Rocket Software; John Joseph, Director, Sales Engineering at Xceptor; and Gille Halle, Financial Services Data Integration Leader at IBM.
Among the conclusions, the panel advised against the creation of a separate AI governance framework and instead organisations should design an “integrated governance platform” that encompasses both traditional and AI governance, a concept that one panellist described as “governance by design”.
Semantic Layers
The design and construction of semantic layers occupies a lot of data-focused literature as organisations seek to bring trust to AI systems as they are scaled. In the panel chat entitled Building The Intelligence Layer – Semantic Architecture For AI-Ready Data Products, panellists discussed how it was critical that careful thought went into the design of these architectures, noting that they comprised four components:
- A storage foundation where raw data resides.
- A semantic core containing metadata, context and standardisation frameworks to unify data representations.
- A knowledge layer that maps entity relationships alongside natural language definitions and documentation from domain experts.
- A distribution and governance layer comprising APIs, security policies and access guardrails.
Among the recommendations that emerged were that they should be built for humans with AI on top of those systems; that explicit evaluation metrics and feedback loops should be established to continuously measure accuracy against real business outcomes; that starting small with high-value domains was the best way to create immediate value; and that operationalising governance would build institutional trust.
The panel, moderated by Dessa Glasser, Independent Board Member at Oppenheimer and Co, comprised Sri Bhupatiraju, Director, Principal AI Engineer at BlackRock AI Labs; Rohan Kodialam, Co-Founder of Sphinx; Matt Katz, Field Chief Technology Officer at Arcesium; and Michael Hom, Head of Financial Services Solutions at InterSystems.
Data Trust
In the panel Scaling With Trust – Data Quality Engineering As A Driver Of Business Performance, expert speakers cast an eye over the long-standing challenge of ensuring information is fit for purpose in an AI context.
New pressures placed on data managers, including ingestion of real-time data and its deployment in AI systems, have forced a re-evaluation of how information is obtained, cleansed and managed. Post-hoc reconciliations and reporting dashboards are now redundant as the criticality of the outputs of those AI systems has risen, the panel discussed.
The use of near-autonomous agents had also raised the ante but had made building a business case for investment easier because there are more compelling metrics that can be cited in appeals to the C-suite.
The panel concluded that data quality should be treated as an engineering discipline with continuous observability signals; that trust in data requires systematic, automated testing, alerting and monitoring; and that internal communications and steering committee presentations must be aligned to highlight the exact business metrics leadership cares about.
The panel comprised Sumanda Basu, Head of Data Quality, Data Risks, Controls and Data Domain Lead at Société Générale; Michael McCarthy, Head of Asset Management and Product Data at Manulife John Hancock Investments;
Naveen Chavali, Director – Product Solutions, NeoXam Americas at NeoXam;
Mick Hittesdorf, Cloud Product Architect at KX; Lydia McCurdy, Informatica Account Solution Engineer at Salesforce; and moderator Brian Buzzelli, Director, Head of Data and Digital Transformation at Meradia.
Verifying Outputs
Accountability has become a hot topic as AI systems become more autonomous, placing new imperatives on compliance teams. In the panel Proving Your AI — Building A Defensible Audit Trail For Agentic Workflows In A Regulated Environment, experts discussed accountability concerns as multi-agent interaction proliferated without human interaction and the mismatch of systems with regulatory models. The importance of lineage was also highlighted in a discussion on the difficulties of BCBS 239 compliance.
Speakers said organisations should define the specific operational failures they want to safeguard against, then build audit controls backwards from those goals. Human-AI collaborative workflows should be designed proactively upfront rather than retrofitting auditability onto shiny pilot projects. And AI governance should be treated as an enabler and continuous discipline that combines technical audit trails, model risk management, and workforce literacy.
Moderated by Pete Harris, Editor, A-Team Group, the panel comprised Vanessa Jones-Nyoni, Chief Data Officer North America at Commerzbank; Balasubramaniam Iyer, Head of AI Governance – International at Nomura; Alex Golbin, Technology and Data Executive, and author of “Governing AI Risk — The RIVER Charter: An Enterprise Resilience Standard”; Samuel Anani, Head of Strategic Account Management – North America at Behavox; and Craig Tutterow, Director of Data at Edge and Node.
Agentic Pathways
The final panel session of the day, entitled Decisions As A Product – From Static Assets To Intelligent Agents, tackled themes such as the real-world uses of agents, the deployment and verifiability of the structures, vendor licensing arrangements and the cultural shift required to implement them.
Governance was evolving, the panel discussed, with humans now providing parallel oversight and the panel warned against validating an AI agent’s output solely with another AI agent.
The speakers concluded by predicting that institutions may soon purchase entire operational units – such as accounting or compliance teams) – as turnkey AI entities and that sales operations will shift away from static slide decks toward live prototypes built on the fly or deposited into client systems prior to meetings. Human judgment will remain essential for complex, high-stakes decisions like corporate restructurings and debt reorganisations, they said.
The panel was moderated by Brian Greenberg, Senior Director – Data Operating Model Lead & Data Management Practice Product Manager at BNY and featured Peggy Tsai, AI and Data Management Leader; Tyler Frieling, Director of Applied Technology at Blackrock; Janelle Veasey, Chief Executive of 3d innovations; Yatharth Sejpal, Chief Executive and Co-Founder of KNOWIDEA Technologies; and Tom Bradley, Chief Product Officer at Fitch Solutions.
Subscribe to our newsletter
